|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 ANALYSIS OF THE SPATIAL DYNAMICS AND DRIVERS OF FOREST COVER CHANGE IN THE LEMPA RIVER BASIN OF EL SALVADOR By HECTOR CASTANEDA A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2009
2 2009 Hector Castaneda
3 To my wife Evelyn
4 ACKNOWLEDGMENTS I thank the School of Natural Resources & Envi ronment of the Univer sity of Florida and the Compton Foundation for funding support for th is study. I would also thank the Land Use Land Cover Change Institute of the Univers ity of Florida for providing software and infrastructure needed for my analyses. I woul d also like to mention the Sistema Nacional de Estudios Territoriales (SNET) of the Ministry of Environment of El Salvador and the University of Florida Digital Map Library for helping in the acquisition of key sa tellite images and GIS layers used in this work. Finally I would like to thank the members of my doctoral committee for their support and guidance. Finally I would like to thank my advisory committee Hugh Popenoe, Peter Waylen, Jane Southworth, Richard Stepp, and Stephen Perz for all their advice and time invested into this dissertation.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ............................................................................................................... 4LIST OF TABLES ...........................................................................................................................8LIST OF FIGURES .........................................................................................................................9ABSTRACT ...................................................................................................................... .............11 CHAP TER 1 THE FALL AND RISE OF FO RES TS: RECONSTRUCTING THE HISTORICAL INTERACTIONS BETWEEN HUMANS AND FOREST COVER IN EL SALVADOR ... 13Introduction .................................................................................................................. ...........13Geographical Location ............................................................................................................16Prehistorical and PreColonial Period .................................................................................... 16Colonial Period. ......................................................................................................................23Post-Colonial Period. ..............................................................................................................27Twentieth Century ..................................................................................................................30Conclusion. ................................................................................................................... ..........342 PATTERNS OF FORESTATION AND DE FOR ESTATION IN EL SALVADOR: ADDING QUALITY TO QUANTITIES IN FOREST TRANSITION STUDIES. .............. 44Introduction .................................................................................................................. ...........44Forest Transition Theory and its Implications for El Salvador. ...................................... 45Materials and Methods ...........................................................................................................49Land-Cover Classification ............................................................................................... 49Land Cover Change Analysis ..........................................................................................52Determining Autocorrelation and Hotspots for Reforestation and Deforestation ...........52Forest Type and Age ....................................................................................................... 53Results and Discussion ........................................................................................................ ...54Classifications and Accuracy Assessments ..................................................................... 54Forest Cover Distribution and Dynamics ........................................................................54Types of Forest and Forest Age. ......................................................................................55Overview and Future Trends ..................................................................................................583 FOREST REGENERATION IN EL SALV ADORS LEM PA RIVER BASIN: A QUANITATIVE APPROACH TO THE SOCIOECONOMIC DRIVERS OF FOREST TRANSITIONS ......................................................................................................................71Introduction .................................................................................................................. ...........71Forest Transition Theory .................................................................................................71SocioEconomic Factors and Fore st Transition in El Salvador. ..................................... 72
6 Methods ..................................................................................................................................75Land Cover Classification ...............................................................................................75Multiple Logistic Regressions ......................................................................................... 76Dependent Variables ....................................................................................................... 77Independent Variables. ....................................................................................................77Results .....................................................................................................................................80Discussion .................................................................................................................... ...........82Conclusions .............................................................................................................................874 APPLICATIONS OF GIS-BASED LOGI STIC M ODELS TO EXAMINING THE PHYSICAL DRIVERS OF FOREST COVER CHANGE AND SCENARIO BUILDING IN THE LEMPA BA SIN OF EL SALVADOR. ................................................ 95Introduction .................................................................................................................. ...........95Forest Cover Change and Forest Transitions in El Salvador .......................................... 96Logistic Regressions in La nd Cover Change Studies ......................................................97Spatial Scale of the Study ................................................................................................98Research Questions. ........................................................................................................98Methodology ................................................................................................................... ........98General methods .............................................................................................................. 98Dependent Variables ....................................................................................................... 99Independent Variables ..................................................................................................... 99Construction of Probability Maps. ................................................................................101Scenario Construction. ..................................................................................................101Logistic Models for Forest Cover Change ............................................................................ 102Deforestation ................................................................................................................. 103Reforestation ................................................................................................................. .106Stable Forest Cover: Permanent Presence or Absence of Forest Cover Throughout the Study Period. ........................................................................................................108Spatial Distribution of Forest Cover Change Probabilities ..................................................110Applications of Spatial Logistic Models to Scenario Building ............................................111Conclusions ...........................................................................................................................1135 INFLUENCE OF PRECIPITATION VARIATION ON THE SPATIAL DISTRIBUTION OF F OREST COVER RE GENERATION IN THE LEMPA RIVER BASIN OF EL SALVADOR. ............................................................................................... 123Introduction .................................................................................................................. .........123Mapping Forest Cover Change in the Lempa River Basin ................................................... 125Reconstructing Monthly Precipitation Distribution Thr ough Linear Regression. ................ 125General Precipitation Regime in El Salvador ....................................................................... 125General Precipitation Regi me of El Salvador ............................................................... 125Linear regressions as a tool for reconstructing historical precipitation distributions .... 128Geographic variables ..................................................................................................... 129Temporal variables. ....................................................................................................... 129Combination variables ...................................................................................................130Predicting Land Cover Change Ba sed on Rainfall Distribution ........................................... 134
7 Logistic Regressions for Testing the Proba bility of Reforestation during the Study Period. ....................................................................................................................... .134Results for Logistic Models of Fore st cover Change Based on Climate ....................... 135Discussion .................................................................................................................... .........139Conclusion .................................................................................................................... ........1426 GLOBALIZATION AND FOREST TRANSITIONS IN SMALL DEVELOPING COUNTRIE S: LESSONS FROM THE CASE OF EL SALVADORS LEMPA RIVER BASIN ..................................................................................................................................155Introduction .................................................................................................................. .........155Caracteristics of Forest Transiti ons Resulting from Globalization ...................................... 156Forest transitions in Small Tropical Countries ..................................................................... 158War and Other Triggering Factors of Forest Transitions ..................................................... 160Importance of Employment Opportunities in the Forest Transition Process ....................... 161Climate Change and Forest Transitions ................................................................................ 163Impact of Spatial Factors on the Dist ribution of Forest Regeneration .................................164Implications of Forest Transitions Resu lting from Globalization in Small Developing Countries ..................................................................................................................... ......165LIST OF REFERENCES .............................................................................................................167BIOGRAPHICAL SKETCH .......................................................................................................179
8 LIST OF TABLES Table page 2-1 Error matrix for the land cover classification of forest cover in the Lem pa River Basin of El Salvador. .........................................................................................................70 2-2 Accuracy assessment of the land cover cl assification of forest cover in the Lem pa River Basin of El Salvador. ............................................................................................... 70 3-1 Market accessibility index values. ..................................................................................... 93 3-2 Best logistic regression model summary fo r the probability of net forest gain in the Lempa River Basin for the period of 1979 1990. ........................................................... 93 3-3 Logistic model summary for the probability of net forest gain based on social data for 148 m unicipalities in th e Lempa River Basin for the period of 1990 to 2003. ............ 93 3-4 Second Logistic model summary for the proba bility of net forest gain based on social data for 148 municipalities in the Lempa River Basin for the period of 1990 to 2003. .... 94 4-1 Results for forest change models ..................................................................................... 121 4-2 values for the variables in the deforestation m odels ..................................................... 121 4-3 values for the variables in the reforestation m odels...................................................... 122 4-4 values for the variables in the logistic m odels regarding stable forest cover. .............. 122 5-1 Linear models for mean monthly precip itation for the Lem pa River basin for the period 1971-1999. ............................................................................................................152 5-2 Logisitc models relating the probability of a 23% increase or greater of forest cover happening in the Lem pa River basin base d on precipitation, slope and land use capability data. .............................................................................................................. ...154
9 LIST OF FIGURES Figure page 1-1 General map of El Salvador. Source : SNET geographic Database, Ministry of Environm ent of El Salvador............................................................................................... 38 1-2 Population distribution at th e year 1550 in El Salvador .................................................... 39 1-3 Population distribution at th e year 1770 in El Salvador .................................................... 40 1-4 Change in population densities El Salvador from 1550 to 1770. Light. ............................41 1-5 Estimated historical population growth for El Salvador .................................................... 42 1-6 Estimated historical forest cover and ma in events affecting its distribution for El Salvador. ............................................................................................................................43 2-1 Geographical location of the Study region. ....................................................................... 61 2-2 Kuznet curve. Behaviour of fore st cover in tim e according to FTT. ................................. 61 2-3 Total forest cover registered for the study area during the period 1979-2003. .................. 62 2-5 Changes in percent forest cover over the period 1979-2003 in the Lem pa River basin. ... 64 2-6 Distribution of Spatial Au tocorr elation (Local Morans I) A) spatial autocorrelation for deforestation B) spatial auto corrlation for reforestation ............................................. 65 2-7 Change in area of forest cover accordi ng to forest type (Life zone) in the Lem pa River basin for the dates 1979, 1990-91, and 2003. .......................................................... 66 2-8 Forest cover in the Lempa River Basin for the year 2003 according to forest type (life zone). ..........................................................................................................................67 2-9 Total area of forest cover in the Lempa River basin for 2003 according to forest ag e. .... 68 2-10 Distribution of forest cove r in the Lempa River Basin acco rding to f orest age in the year 2003. ...........................................................................................................................69 3-1 Study area: Departments, municipalities and extension of the Lem pa River Basin in El Salvador. ........................................................................................................................89 3-2 Changes in net forest cover by munici pality for the period 1979 to 1990 in the Lem pa River basin.. ........................................................................................................... 91 3-3 Changes in net forest cover by munici pality for the period 1990 to 2003 in the Lem pa River basin.. ........................................................................................................... 92
10 4-1 Probability map for reforestation accordi ng to the logistic m odel for the period 19792003 in the Lempa River Basin of El Salvador. .............................................................. 115 4-2 Probability map for deforestation according to the logistic m odel for the period 1979-2003 in the Lempa River Basin of El Salvador. ..................................................... 116 4-3 Changes in the probability of forest re generation accord ing to two possible scenarios for the period 2003-2027. ................................................................................................117 4-4 Changes in the probability of deforesta tion according to two p ossible scenarios for the period 2003-2027. ....................................................................................................118 4-5 Estimated changes in forest cove r for the period 2003-2027 according to two different scenarios. ...........................................................................................................119 4-6 Predicted trend of forest cover in the Lem pa river basin according to the proposed scenarios. .................................................................................................................... ......120 5-1 Location of study area: Lemp a River basin, E l Salvador. ...............................................143 5-2 Average monthly precipitation in El Salv ador based on rainf all data from 32 rain gauges for the years 1979 2003. ...................................................................................144 5-3 Differences in rainfall distribution for El Nio, L a Nia and normal years in the Lempa River basin over the period 1971-2005. ............................................................... 145 5-4 Location of rain gauges and geographi cally dif ferentiated rainfall areas in El Salvador. ..........................................................................................................................146 5-5 Distribution of mean tota l prec ipitation in the Lemp a basin during the dry season (December April) for the years 19711999 according to precipitation models. ........... 147 5-6 Distribution of mean tota l prec ipitation in the Lempa basin during the beginning of the rainy season (May June) for th e years 1971-1999 according to precipitation models. ....................................................................................................................... ......148 5-7 Distribution of mean tota l precipitation in the Lem pa basin during the mid-summer drought (July) for the years 1971-1999 according to precipitation models. .................... 149 5-8 Distribution of mean total prec ipitation in the Lempa ba sin during the second part of the rainy season (August-November) for the years 1971-1999 according to precipitation models. ........................................................................................................ 150 5-9 Actual Vs. Predicted increments in forest cov er (>23%) in the Lempa basin according to the climate based fo rest cover change model. ............................................. 151
11 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ANALYSIS OF THE SPATIAL DYNAMICS AND DRIVERS OF FOREST COVER CHANGE IN THE LEMPA RIVER BASIN OF EL SALVADOR By Hector Manuel Castaneda Langlois August 2009 Chair: Hugh Popenoe Major: Interdisciplinary Ecology This work studies the changes of forest c over that have happened in the Lempa River Basin of El Salvador during the period 19792003. Although historically the trend has been towards the loss of forest cover since colonial ti mes, over the period of st udy a large increase in forest cover was detected. The main tool of ev aluation was the analysis of LANDSAT satellite imagery. Images for the dates 1979, 1990-91, and 2003 were classified into forest and noonforest land covers. Then the changes in land cover were analyzed to determine what were the social, geophysical and climatic drivers determ ining why and where these new forest appeared. The results indicate that there has been an overa ll increase in forest cover from 20% in 1979 to 43% in 2003. Although there has been extensive deforestation, this has happened mostly around the main urban centers within the basin. In the more rural and remote areas, the tendency has been towards a resurgence in forest cover. The increase in forest was found to be significantly related to remittances, inaccessibi lity to roads and markets, density of urban populations, poverty and the civil war of the 1980s. Among the geospatia l factors that determined where deforestation and reforestation happened were distance to roads and urban centers, slope, elevation, land use capability, and irrigation potential. The results in dicate that the tendency in the future will be towards further reforestation but at a slower rate. Although reforestat ion and deforestation
12 happened simultaneously, there are clear differences in the spatial patterns that each of these phenomena follow. In terms of climate, it wa s found areas subjected to inter-annual rainfall extremes due to El Nio Southern Oscillation, particularly areas with lo w agricultural potential, were more likely to be abandoned and left to revert to forest than those with more stable rainfall. The results of this study s upport the hypothesis that El Salv ador is undergoing a Forest Transition process, that is a recuperation of forest cover due to urbanization, migration and economic growth.
13 CHAPTER 1 THE FALL AND RISE OF FO RES TS: RECONSTRUCTING THE HISTORICAL INTERACTIONS BETWEEN HUMANS AND FOREST COVER IN EL SALVADOR Introduction El Salvador has often been depi cted as a country with virtua lly no natural forest cover. Being the sm allest and most densely populated country of the mainland Americas, many authors have pointed it out as a Malthusia n disaster regarding natural re source management, particularly forests (FAO 2001; Hampshire 1989; Dull 2007). Recent studies indicate however, that, while there is deforestation go ing on near urban areas, new secondary forest growth is appearing in former agricultural areas (Hecht et al. 2005). This is apparently due to abandonment of fields triggered by a diverse assortment of factors (Hech t et al. 2005). The truth is that forest cover changes are not a simple one-way trajectory. When we observe long term historical trends in forest cover, often cycles of de forestation and reforestation are the rule instead of any linear tendencies pointing towards inevitabl e deforestation. El Salvador pres ents an intere sting case of interaction between a changing hum an society and an adaptable, highly anthropogenic vegetation cover. This study seeks to reconstruct the countr ys forest cover history under the scope of a changing society. Early forest ecology was centered around the ideal of primary or virgin forests, referring to forests in a primeval state, undisturbed by the hand of man This concept, however is slowly disappearing as archaeological rese arch has begun to discover that even the vegetation of the densest, most remote of forests of regions su ch as the Amazon basin, l ong considered the last stronghold of pristine tropical forests, have been shaped and molded by thousands of years of human occupation and management (Heckenberger et al. 2003). The occide ntal concept of the duality of man and nature is slow ly evolving into the re alization that our speci es is, and has been since its origins, a part of the plan etary ecosystem. Even since prehistorical times, there are few,
14 if any, places on earth that can be said to be fr ee of human influence. The impact of our species on vegetation in the past has larg ely defined the composition and di stribution of forest cover in the present. Similarly, the concept of a climax vegetation, an ecological state of balance to which forests should tend to revert to if left undisturbed, has been found to be an illusion. It is our perspective, based on our short life spans, which fail to capture the state of constant change in which forest structure constantly varies at a br oader time scale than we can normally perceive. While climax states may appear to be stable for relatively short-term periods (geologically speaking) of hundreds or thousands of years, in the long term, ecosystem s tend to be dynamic, very often evolving into completely different syst ems and never returning to a static point in the past. For example, the Mayan forest, one of the largest rainforests left in Mesoamerica, once thought to be an epitome of climax vegetation, has also been proven to be the result of the Mayan civilizations enormous impact on the local ecosystems (Abrams and Rue 1988). Highly developed agriculture and use of native trees by the ancient Mayans prom oted the survival of certain useful or culturally important specie s over others during a period when this region consisted mostly of urban and agricultural lands capes. When the Mayan civilization collapsed these trees became predominant in the forest that reclaimed the abandoned fields (Curtis et al 2004). It is highly unlikely that th e structure and composition of th is ecosystem will ever revert to the same state that existed before the time of the Mayas. Another misconception about human interactions with forest cover is the concept of unidirectional reduction of fore st cover. Often the interacti ons of humans with forests, particularly in developing count ries, are perceived to consist mainly of forest removal and agricultural expansion. While this is definitely true, th e interactions between human societies and
15 forest cover are far more complex. Climatic, economic, ecological and social conditions, among others, also lead to the abandonmen t of fields which often tend to revert back into forests. This means that both deforestation and reforestation (whether anthropogenic or otherwise in nature) are always taking place simultaneously. The magnit ude and rate at which events happen varies according to the global, regional and local factor s that influence human decisions regarding land use. Although timber production is not a big industry in El Salvador, tree cover is an essential if often unacknowledged part of the countrys functioning. In this semi -industrialized nation, forests provide its poorest inhabitants with a source of fuel wood, construction materials, and non-timber forest products. Besides this direct usage of forest products, tree cover creates the ecological framework for all of Salvadorian soci ety to function. By increasing water infiltration during the five-month rainy season, forests give the country a stea dy supply of drinking water, fuel hydroelectric plants and reduce sedimentation due to erosion in dams which account for 33% of the countrys power and 66% of the countr ys drinking water. Tree cover is also essential for the production of shade coffee which is one of the most important crops in the Salvadorian economy. In order to understand the stat e of contemporary Salvadorian forest cover it is important to understand the changing interactions human societies have had up until the present day. Radical changes in both the culture and economy of Salvadorian society have reflected on its relationship with the environment. From the preh istoric period of early human occupation to the industrialization period star ted during the twentieth century the pr essures that have shaped forest vegetation in the country have been under consta nt change, exerting evolutionary pressures on plant species and ecosystems in general and shaping them into their current state and
16 distribution. This work explores historic, archeological and pa leonthological evidence in an effort to understand the composition, distribution and trends that have shaped forest vegetation in El Salvador thr oughout its history. Geographical Location El Salvador is loc ated in the Pacific waters hed of Central America, bordering to the West with Guatemala and to the North and East with Honduras (Map 1). Aver age annual precipitation ranges from 1141 to 2886 mm per year. Elevations range from 0 to 2678 meters above sea level. The variation in topography a nd rainfall regimes makes it an ecologically diverse region. According to Holdridges life zones it contains eight distinct ecological zones with several distinct ecological associations due to variation in soils. Major land uses on the present vary widely including industrialized agriculture, tradit ional subsistence agriculture, cattle raising, urban and sub-urban areas, natural forest prot ected areas, highland and lowland forests, and extensive shade coffee plantations (SNET 2005). Prehistorical and Pre-Colonial Period It is often a first assumption when thinking of Mesoamerica that lush tropical forests are and were the default vegetation before humans had an impact on it. This however is not supported by palynological evidence. Pollen records indicate a diverse, discontinuous land cover existed during the Holocene indicating the presence of not only forests, but also of vast non-forest formations such as wetlands and savannahs. The first set of factors affecti ng El Salvadors land cover has been natural events such as earthquakes, volcanic eruption, and, to a lesse r degree, hurricanes a nd tropical storms. Earthquakes, mostly caused by the ef fects of the San Andres fault, started being recorded by the Spanish since 1524, particularly in the central hi ghlands and coasts. The effects of major quakes are widespread landslides in steep slopes and occasional destruction of human settlements. A
17 recent example of deforestation due to earthquakes is the two 2001 seismic events which in total caused 576 major landslides, particularly in th e shade coffee growing regions of the central highlands. (Dull 2007). Volcanic eruptions also play a relevant role in land cover change. The first role of volcanoes is in the direct elimin ation of vegetation and, in some cases human populations. Evidence of this can be observed in the numerous lava flows foun d throughout the basin or in archeological sites buried under volcanic ashes. The second, less obvious role volcan ic activity plays in land cover change is the deposition of fertile minerals and ashes, dictating where the most fertile soils are and thus the areas where human agricultural soci eties settled in the pa st. (Sheets and Woodward 2003). Although not much is known of land cover in prehistoric El Salvador, paleozoological studies suggest that during the Pl eistocene period a large part of the territory supported large numbers of savannah fauna including ground sl oths, mastodons and bison. From the The presence of these animals one may infer the exis tence of savannah vegeta tion with an extensive distribution. (Janzen and Martin 1982; St ratton and Gaesley 1949). Until the late Holocene period, it is believe d humans had little impact over vegetation, existing in populations with lo w densities and surviving through hunting and gathering. It was not until about 2000 BC that sedentary agriculture spread through the region increasing the influence of humans on vegetation. (Blake et al. 1995; Love 1999; Neff et al. 2006). Palynological studies show eviden ce that the main vegetation fo rmations at this time were anthropogenic savannahs and thorny scrub formati ons (likely derived fr om agricultural and natural fires), tropical forests in the coastal lo wlands and slopes, and pine-oak forest in the northern highlands. Dull points out that the central valley and highlands in particular show strong
18 evidence of anthropogenic vegetation (maize, usef ul tree orchards, and savannah vegetation such as Cresentia alata, Curatella am ericana, Byrsonyma crasifolia ). This pattern of land use is likely to have continued until the arrival of the Spanish with varying degrees of intensity. Peleobotanical studies show ev idence of the formation and disappearance of anthropogenic savannahs implying periods of defo restation and reforestation as human settlements rose and fell (Dull 2007). It is hard to determine which lands within the Lempa basin, if any, where ever free of human occupation in the past 2000 years. This can be se en in the many archeological sites found within protected areas once thought to be the last remnants of undistur bed forests in the country. The region now encompassed by El Salvadors bord ers was a major center of population during the pre-colonial period. Within its borders lie more than 671 registered archeo logical sites. (Amaroli 1996). The country was occupied by Olmec, Lenca, Pocomam and Nahua (Pipil) (Fowler 1988; Pohl et al. 2002) peoples at different periods prior to the Span ish invasion, the Pipil being the dominant culture from 800 AD to the 1500s AC. While the northern pine forests with lower soil fertility have few archeological sites, the Zapotitan valley, and the highlands surrounding present day San Salvador are surrounded by numerous archeo logical sites such as San Andres, Joya de Cern, Cihuatan, and Guija which te stify to highly developed agricu ltural societies. By the time of the Spanish arrival, population in the study re gion, according to early records of the area, was then concentrated mostly along th e central slopes and fertile volcan ic valleys of the country, and away from the central Lempa basin and the northern slopes. This distribution of human settlements might in part be due to the edaphic conditions in the region. Th e volcanic soils in the central mountains have a much higher potential fo r agricultural production than the older, poorer soils found in the central and lower Lempa valley or those of the northern highlands. They also
19 retain humidity better during th e dry season making settlement conditions more favorable (Browning 1971; Sheets and Woodward 2003). The first Spanish records from the area describe indigenous communities organizing themselves in dense population centers in the foothills of the central valleys with numerous but sparsely distributed farming households in the h illy surrounding areas. The c limate in this area is also milder in the sense that it does not have a pronounced drought as do the lower valleys. (Browning 1971). Milpa cultivation, a system s till practiced today in some areas, was the main method of food production throughout this period. This system was pr acticed in all kinds of terrains, from the steep mountains to the lower valleys. Indigenous soil management techniques still in existence today allowed pre-colonial farmers to sustain the permanent or semi-permanent agriculture in the steep slopes the Spaniards first described. Soil mana gement activities such as irrigated tablon terraces, camellones or ridged fields, rock-walled terraces, rais ed beds, organic amendments, and agro-forestry among other tec hniques used in this period lim ited the loss of soils in broken topography and allowed for the cultivation of steep topographies. (R ojas-Rabiella 1985; Whitmore and Turner 1992; Wilken 1987). Evidence from archeological sites such as Joya de Cern suggest that the milpa practiced during that period was not an extensive shifting cultivation practice but rather a more permanent, intensive and highly diverse agro-ecosystem. This thesis is supported by the fact that the native peoples of El Salvador did not possess metal t ools which would make s hort cycle slash and burn agriculture as practiced today ex tremely difficult. It is likely that metal tools transformed the intensive milpa system into the simpler, sort-fallow extensive corn production system ( ) that is
20 more commonly practiced in present day El Sa lvador (Mathieu and Mayer 1997; Wilken 1987; Sheets and Woodward 2003). Archeological and ethnographical data also indicate the importance of orchards and managed forest successions for indigenous comm unities. Managed fallows were an important source of secondary products such as medicine, wild or semi-domesticated food plants, building materials and fauna among others. Evidence of such systems can be found stil l in the practices of other nahua-speaking groups such as the Huastecs in Mexico (culturally rela ted to the Pipil that dominated El Salvador), or in the balsam cultiv ation systems in western El Salvador. Diverse, stratified canopy home gardens st ill found today throughout the Amer icas also seem to have played an important role in food pr oduction and land use. (Alcorn 1981). Crops grown in pre-colonial El Salvador included maize beans, squash, cotton, cacao, indigo manioc, chili peppers, ujuste ( Brossimum alicastrum), and a high diversity of other roots, fibers, fruits, gourds, and tree crops. Evidence at the Joya de Cern archeological site also indicates that complex agricultu ral practices such as ridged fi eld cultivation, stratified home gardens, and irrigation were present at this tim e. (Lentz et al. 1996; Lentz and Ramirez-Sosa 2003). Cacao was an important crop during the precolonial period. There is evidence that the region had extensive cacao planta tions which were exported as a commodity to the Mayan citystates and later to the Aztec em pire. The cultivation of cacao, as described by the Spanish in Guatemala, was an agroforestry system that em ployed nitrogen fixing shade trees such as madre cacao (Gliricidia sepium ) and Pepeto (Inga spp) as shade for the crop. These species are still used in the same manner today for lowland and midland coffee production. However cacao
21 requires gentle slopes with deep fertile soils and high rainfall which limited its cultivation to the western regions and coasta l areas (Bergman 1969). In general, agricultural practices in pre-col onial El Salvador were characterized by their intensiveness. Intensiv e agriculture focuses on using high labor inputs to maximize production per unit area; this is particularly useful if ther e are no metal tools or large domesticated animals to reduce the costs of land pr eparation. Intensifi cation also involves complex intercropping practices, soil conservati on structures, long managed fallows, agroforestry systems and cultural pest controls. Modern agriculture on the other hand, is extensive, that is, it seeks to minimize labor inputs by using large s cale monocropping, chemical fertilizers, pesticides and mechanization of large tracts of land. Thus we can conclude from these facts that agriculture as practiced prior to colonization demanded less land and therefore exerted less pressure on forest ecosystems than agriculture as practiced in El Salvador throughout colonial period up until the present. Although few studies have been done directly on the aspect of climate changes in El Salvadors pre-colonial past, studies from Guatemala and Belize suggest that climatic changes were among the major forces that might have tri ggered changes in regional forest cover. During the IX century AD, major changes in climate are believed to be re sponsible for the collapse of the Maya civilization and the massi ve regeneration of forests in the Peten region (Curtis et al. 1998; Hodell et al. 2001). The way these changes affected El Salvador are not yet known, however likely changes may have be en the collapse of larger settlements such as Tazumal in the west of the country as well as a possible influx of migration from the Mayan region as people migrated to escape harsh climatic and socio-politi cal conditions.
22 The information from archeological sites and th e descriptions and record s of the first Spanish explorers suggest that the central and north ern Lempa basin should have been covered by relatively continuous deciduous tropical forests, savannahs or fire resi stant low thorny scrub (which is currently the vegetation that tends to re generate in these areas when left fallow), dotted with sparse agricultural mosaics of milpa based communities around year round water sources. This pattern is found in other parts of Mesoam erica with pronounced dr y seasons (Bullard 1960). An exception would have been th e areas surrounding major population centers in water-rich soils where irrigation and flat terrain allowed more extensive grain cultivati on. Even in the denser populated highlands, the use of orchards, strati fied home gardens, and managed forests and fallows would have meant the landscape would have been composed of a mosaic of forested and cleared land rather than mostly open agricultural sy stems. The coastal lowlands were fairly well populated indicating forests were cleared to give way to agriculture. The coastal slopes in the west, on the other hand, are likely to have been mostly covere d by semi-deciduous and evergreen forests and sparse settlements. Finally, the northe rn highlands are likely to have been dominated by less disturbed pine-oak woodlands and forests with few settlements. Although forest cover likely decreased with the advent of agriculture, the concentration of population densities in the central highlands and the use of intensive agriculture points to a large portion of El Salvador possessing a signifi cant percentage of forest cover up until colonial times. This includes a large portion of highly anth ropogenic forests (cacao agro-forest, managed fallows, stratified home gardens). The expansion and collapse of settlements due to social or climatic changes also likely trigge red periods of deforestation and reforestation in certain regions which maintained a balance between constant fore station and deforestation patterns. Evidence of
23 this are the large city-states such as Tazuma l, San Andrs and Cihua tan, which were already abandoned and overgrown by the time the Spanish arrived. Colonial Period. By the time of the Spanish arrival to El Sa lvador, the regions population was estimated at 750,000 (Fowler 1988). Judging by the di stribution of archeo logical sites and Spanish records, most of this population was concen trated in the central highlands and fertile valleys with milder climate and favorable rainfall regimes as well as th e coastal lowlands in th e east (Figure 1-2). The Spanish invasion brought along drastic ch anges in land use for El Salvador. The introduction of disease, new technologies, ne w economic and land ownership practices, and exotic species played a relevant role in the tr ansformation of Mesoamerican landscapes to what they are today (Whitmore and Turner 1992). The large scale production of indigo, for instance, is one of the activities that affected Salvadorian landscape. The cultiv ation of this plant, previously a minor indigenous product, became widespread to satisfy European demand for the dye (Escobar 1965). Indigo cultivation required exte nsive clearing of forest, bur ning of the undergrowth and subsequent abandonment after three years of cultivation. The official method for indigo cultivation, dictated by the Spanish crown, consis ted of a mixed system of cattle-indigo-maize which required a constant rotati on and burning of active and fallow fields and took advantage of the fact cattle would not eat indigo plants. El Salvador became the second largest colonial producer of indigo during this pe riod after the French colony of Saint Domingue (present day Haiti), and most of the west-centr al Lempa river valley, an area wi th little signs of pre-hispanic habitation, was converted into Indi go-cattle haciendas. Historical records describe the landscape in the northern San Salvador and La Libertad departments as de nuded of all trees except for the
24 riverbeds during this period. This extensive and intensive cultiva tion system greatly defined the landscape for this area up until present day (Smith 1959; Alden 1965; Browning 1971). The advent of indigo production also changed th e settlement patterns pr esent at the time of the Spaniards arrival. Population shifted from the indigenous highland settlement pattern to the colonial hacienda pattern which expanded into the dryer regions of the Lempa valley. Still the dryer, less fertile areas around the centr al basin remained fairly uninhabited. An important difference in habitation patterns between colonial and pr e-colonial periods was the density of human settlements. Whereas indige nous populations tended to be widespread into thinly populated centers, colonial habitation pa tterns concentrated population in fewer but densely populated urban centers (Figure 1-3). The reason for this was the hacienda system which worked in a semi-feudal structure and re quired a large centralized labor force. In regions not suited for i ndigo, the more humid lowlands al ong the Cordillera del Balsamo remained dedicated to the production of cacao, which had become another profitable colonial commodity. The importance of this product was such that many of the cacao cultivation areas such as Sonsonate and Izalco were allowed to retain their indigenous autonomy in order to maintain the production of cacao stable. The slop es of this humid areas were used for the production of balsam ( Myroxylon balsamum ), a multi-use aromatic resin of which El Salvador has been the main producer even until present day (Can ales 1985; Castaneda 1999). 24Both balsam and cacao production were responsible for the retention of a managed tree cover in the southern slopes of the Lempa basin and the western coastal slopes. Later on, these systems would give way to the current shade coffee plan tations present today in that same region. Sugar cane was another relevant crop introdu ced into the Lempa valle y. The irrigated lands of the Zapotitan valley and part s of the central Lempa valley were dedicated to this crop.
25 Although of lesser importance than indigo, sugar cane generated a lot of deforestation. This forest removal came not only through direct clea ring but by increasing the risks of forest fires during the burning season, a phenomenon still common today This area was particularly known for its wetland swampy forests, an extinct ecosyst em today due to the channeling of waters for irrigation systems. Of the exotic biota introduced by the Spanish, cattle had the greatest impact on the colonial transformation of El Salvador. Cattle ranching became a predominant land use previously unknown to native cultures, particularly in the no n-arable lands throughout the basin as was the case in other parts of Mesoamerica. This new land-use contributed to the clearing of previously depopulated areas. Burning of fields for openi ng pastures, damage to indigenous farming systems and infrastructure by free roaming herds, and the suppression of forest regeneration by trampling are just some of the impacts cattle had on the native landscape (Brand 1961; Chevelier 1963; Cook and Borah 1979; Morisey 1951; Super 1988). Another important change in land use and set tlement patterns brought by the Spanish to the New World was the spread of disease. For exam ple, the introduction of water-related disease such as malaria and yellow fever, caused the depopulation of the coastal lowlands in the east leaving them virtually deserted during colonization. Other diseas es such as measles and small pox are believed to have contributed to the ma ssive depopulation of the eastern valleys of the country leaving these areas to re generate their forests while de velopment concentrated along the western highlands. The same phenomenon occurred throughout the Americas as the European colonizers advanced. The intensive indigenous farming systems practiced throughout the region likely collapsed due to the lack of labor (Lovell 1992; Cook and Borah 1979).
26 Technologies such as metal tools are another factor that significan tly contributed to the transformation of land in Mesoamerica in general. Metal tools allowed fo r easier clear ing of land and increased the efficiency of agricultural labor s in general. Their intr oduction allowed a more dynamic shifting cultivation system and lowered th e need for intensive lo ng-term agriculture or management of secondary vegetation in fallows for secondary forest products. Together with cattle, metal tools also allowed fo r new practices such as tilling. T illing favored the settlement of low slope areas with deep fertile soils in large farm plots. This contra sted with the indigenous system which was more adapted to shallow, steep soils and small intensive plots. Metal technologies also changed the size of farm plots a nd the types of soils that could be used. It also favored the proliferation of shifting cultivation in previously unexploited lands (Hurtado and Hill 1989; Mathieu and Mayer 1997). The landscape in El Salvador in the late 1700s would have varied with regards to its initial stage in the 1550s (Figure1-4). Although in general populations te nded to decline and concentrate in the Central Highlands, extensive agriculture involving the massive use of fire, cattle, and metal tools replaced the intensive milpa system This gave to the mass removal of forest cover in the central and western valleys The land management practices of the Spanish and their export driven colonial economy re quired more land than the intensive system of the native people. Cattle and indigo played a key ro le in shaping the denuded landscape of the Lempa valley that exists in many areas up to th e present day. Even traditional milpa growers, now with access to metal tools, abandoned their la bor intensive systems in favor of the short rotation shifting cultivation system still practiced in the present. The eastern Lempa valley, on the other hand experienced a loss of population due to disease and other factors, and probably experienced regene ration of forest cover with regards to its pre-
27 colonial state. Similarly, the eastern coastal plains, previously well inhabited are likely to have experienced an increment in forest cover due to the loss of population br ought about by disease and conflict with the Spanish. The secondary fo rest that resulted from this abandonment remained, until the twentieth century, among the last strongholds of dense tropical forests in the country. Animal species such as tapirs, monkeys, scarlet macaws, and jaguars were still abundant in the coastal forest of Usulutn and San Miguel, in the eastern part of the country up until the early twentieth century. Eventuall y, these eastern forests were clea red to give way to extensive cotton plantations that characterized the area during the post-colonial period until the late twentieth century. Post-Colonial Period. By tim e El Salvador became independent from Spain (1821), agricu lture was intensified particularly in the flatter areas of the Lempa valley. Indigo still remained the main cash crop of the area with a stable demand in foreign markets. Extensive corn and bean production, with the added influence of metal tools, was still an important part of the countrys main grain production. In the central mountain range, tree cover remained in place due to cacao agroforestry and balsam extraction and also because of the lack of highland export crops. Due to a relatively low population density and abundance of arable land, most of the Lempa basins higher volcanic slopes were still undisturbed. It was not until the 1820s, when coffee was introduced to the country that agriculture starte d actively spread ing into the volcanic highlands above 1000 meters from sea level. The cultivatio n of coffee from that period onto present times, was mostly done under shade to protect the crop from the heat of the dry season. In El Salvador the usual practice was to clear th e understory of natural forests and replace it with coffee. This resulted in an agroforestry system composed of a diverse overstory of trees and a coffee
28 understory. Many of the cacao lands, suffering fr om disease and low prices, were easily converted into lowland coffee by eliminating cacao and replacing it with coffee plants. In the case of balsam stands, the transi tion was easier, the loosely manage d stands of balsam and other timber species were left in place and the understo ry replaced with the cash crop. This system of cultivation is still in use today in some regions. In some cases, however, coffee was responsible fo r the destruction of many of the forests that remained in the country. Even if coffee required shade, larg e amounts of native forest were burned to the ground and later repl aced by more uniform inga trees. The reason for this was the governmental policy of granting land ownership to those who cleared the land. Even if it was not necessary for coffee plantations to remove the en tire canopy, ownership was granted only to those who fell the forest completely before planting. (Browning 1971). Coffee plantations in the country in the pres ent vary from traditional diverse canopy systems similar in composition to native forests to less co mplex systems with a lo wer diversity of trees. Although coffee plantations will neve r have the diversity of natu ral forests, the tree cover provided by them in El Salvador has served as a substitute for many forest ecosystem functions such as erosion and sedimentation control, wate r infiltration, biological co rridors, and refuge for wildlife. Unlike more sophisticated shade systems such as those used in countries like Costa Rica and Colombia, where a very well pruned, single speci es with low density tree cover is used; most of El Salvadors shade coffee systems are relativ ely diverse. Biodiversity studies in three coffee farms in El Salvador found a total of one hundred sixty-nine species of tr ees belonging to forty six different botanical families (Komar 2003; Cuellar and Rosa 1999; Mendez and Bacon 2005). Coffee production slowly started to gain ground while indigo began to loose its demand with the advent of artificial dyes. Gradually, cultivatio n of the latter slowly disappeared and the open
29 lands in the indigo-producing re gions which continued to be us ed for cattle raising and sugar cane plantations. Many of the former indigo farmers and workers spread into the northern highlands of Chalatenango. Due to the poor fertility of its soils, this region remained sparsely settled and dotted with shifting cultivation and low productivity cattle. Both soil conditions and high wind intensity made the area unsuitable for coffee plantations like those in the central mountains, thus pine-oak woodlands remained relatively untouched. Meanwhile, a new industry was rising in the co astal lowlands of the country. The hot humid plains here were found to be suitable for co tton production. This crop was responsible for the clearance of most of the lowland forests that unt il this point remained in that region, and would remain a major land use until its collapse in the 1970s. Besides extensive deforestation cotton brought with it other environmental hazards such as the overuse of pesticides. This abuse resulted in massive intoxications of farm worker s, outbreaks of malaria du e to the proliferation of DDTresistant mosquitoes and the degradation of ma ngrove systems and fisheries downstream. The general trend in land use in the Lempa basin for the nineteenth century until the midtwentieth century was of aggressi ve agricultural expansion. It is during this period, an extension of the same trends from the colonial period, that great losses of forest cover happen with few if any cases of forest regeneration. Among the major activities responsible for the demise of forests in the country are the cultivation of sugarcane in the more fertile flat lands, rain-fed subsistence agriculture and cattle in the foothills and northern mountains, and coffee in the higher slopes of the southern mountains and volcanoes. The coastal areas at this time were wide ly used for the cultivation of cotton, grains and cattle, and most of the coastal forests were cleare d for these uses. Natural forest cover
30 became relegated to the northern pine-oak woodl ands and the more inaccessible parts of the central mountain range. Even if coffee was responsible for the elimina tion of a large part of the native forests cover, it is important to emphasize its relevance as a la nd use since it maintained 7.9% of the country under tree cover and has been serv ing as an ecological substitute for natural forests (PROCAFE 2004). Although population in the basin had always c oncentrated in the ce ntral highlands, during this period it slowly spread in to the western and central Lemp a valley, the Lempa floodplain and even to the northern highlands. Twentieth Century Although population in the basin had always concentrated in the central highlands, during this period it slowly spread in to the western and central Lemp a valley, the Lempa floodplain and even to the northern highlands. El Salvadors population grew rapidly during the latter part of the twentieth century (Figure 1-5). It is very likely that this growth is due in part to th e increased production of food brought about by the green revolution in the late 1960s throughout the developi ng world. However this rapid transfer of technology also brought about problems such as agricultural frontier expansion to non-suitable lands, erosion due to over cultivation, population growth and subsequent land scarcity. El Salvadors population grew rapidly during th e latter part of the tw entieth century (Table 1). It is very likely that this growth was due to the increased production of food br ought about by the green revolution in the late 1960s throughout the developing world. However this rapid transfer of technology also brought about problems such as agricu ltural frontier expansion to
31 non-suitable lands. The result was widespread er osion due to overcultivation, population growth and subsequent land scarcity. The conditions of shortage of arable lands unequal distribution of wealth, and declining grain yields present during the 1970s contributed to the brewi ng of the twelve year armed conflict that was to engulf th e country from 1980-1992. The civil war period was extremely influential on the patterns of land use and settle ment and can be considered a turning point in land cover trends for the country. One major ch ange it brought about was the beginning of massive migrations from the rural areas into the cities, and towards foreign countries (mainly the United States) which have continued even afte r the conflict en ded (Kandel 2002). Over twelve years people abandoned agri cultural lands to escape the violen ce and the four main urban centers of the country sprawled with the influx of refugees. The effects can be seen in the changes in the urban/rural population ratio from 1930 to 1992 according to the data from the national population census (Table 1-1). Urban population has increased at a higher rate than rural population. The census data also shows an increa se in population density concentrating mostly along central highlands, the wester n Lempa basin, the western coast line, and the eastern coastal plains. This massive city sprawl caused by the ur ban population explosi on, and the preference for settling in the central mount ain range has caused the conversion of coffee agroforests into urban areas, thus becoming the main cause of forest cover loss in the c ountry during the 1980s and 1990s. On the other hand, this reduction in rural population diminished the demand for land in rural areas leaving thousands of hectares to regenerate into secondary forests due to abandonment. The construction of several hydroelectric dams also contributed to the displacement of agriculture and loss of arable lands (Blackman et al. 2006).
32 Foreign migration also had its effects. By 2000, the countrys main sour ce of foreign income had ceased to be agricultural exports and had sw itched to remittances from the United States. This input is believed to have caused rural households receiving it to lower the intensity of farm work and invest the money in commerce or real-estate. Studies have found that families receiving remittances are more likely to invest in e ducation of their young people; consequently, young people with more education tend to move to cities in search of higher income jobs as opposed to staying on the family lands and working in agriculture (PRISMA 2002; Andrade-Eekhoff and Kandel 1998; Zilberg and Lungo 1999). One indirect effect of external migration is the change in the gender composition of the population. In 1971 the majority of the country was dominated by male population. By the time the war was over in 1992 the majority of the countrys population was female due to the fact that most external emigrants are male. This change in population composition has ma ny implications. For one part, traditional gender roles of women in rural El Salvador ar e different than those of men. While many times women will work the fields, more often they wi ll focus on small livestock, home gardens and commerce activities. In general women tend to assume the role as head of households when their partners migrate, yet retain their traditional gender-related occupations. The money from the remittances is mostly used to invest in propert y, education, and small businesses. In this manner households where the male members who have em igrated and send remittances are less likely to work their agricultural fields. Blackman et al. s (2006) research suppor ts this hypothesis by finding that there was a significant negative correla tion between the percentage of women in the population per canton (smallest political unit) and the amount of deforestation during the 1990s in coffee regions (Mahler 2000; Lungo and Andrade-Eekhoff 1997).
33 Another important factor in land use patte rns was the land reform of 1980. In hopes of diminishing support for the guerrilla movement, the government decided to give land to poorer rural families. The process of land reform had its setbacks. Large land owners were forced to sell part of their property to the government to la ter be divided among poor families in the region. The result of this was that the lands given to th e agrarian reform tended to be the marginal, less productive ones. In the end the av erage property size of the re-distr ibuted land ranged from 1.7 to 5.8 hectares. In general this measure was not enough to alleviate poverty (Diskin 1996). Marginal lands in the country mainly are used for cattle production, however such small plots did not allow enough cattle to become an alternative to the beneficiaries of land reform. This lack of planning led to the use of poorly managed, intens ive swidden agriculture to be carried out in unsuitable lands often leading to erosion and declining yields. Land scarcity, declining rural wages and low so il fertility further co ntributed to migration, causing the abandonment of rural areas. In othe r cases land reform led to the formation of cooperatives. In many cases cooperatives actually promoted the conservation of natural resources within communal lands. Another significant effect of landreform was that it discouraged investment in agriculture from the wealthy upper class. Since th e government set a cap on private farm size (245 ha), private large scale agriculture d eclined. Instead investments in industry increased, creating more jobs in the urban and suburban areas (Gomez et al. 2005; Hecht et al. 2005). Policy changes during the 1990s also had thei r effect on land use. The period from 1960 to 1980 was characterized by strong subs idies to extensive and industria l agriculture such as sugar cane, cotton and cattle-raising. These policies promoted the massi ve removal of forest cover
34 where ever land was apt for such activities. Th e 1992 peace accords brought about changes in these sectors. Subsidies on cattle and grain pr oduction, the two main land uses of non-arable land, were removed, allowing for cheaper imports from abroad to compete with local producers and adding to the diminishing profitability of agricultural work (Faber 1993; Paige 1999; Utting 1994). One last factor that contributed to this situa tion was the instability in grain and coffee prices due to fluctuations of the global economy, as well as climatic anomalies such as El Nio phenomena and hurricanes such as Mitch and Fi fi. Major earthquakes su ch as those in 1970, 1986 and 2001 also played a role in encouraging the abandonment of agri cultural production in favor of migration. The result of these complex changes triggere d in great part by the civil conflict created a pivot point in the nature of fore st cover change. The general trend of deforestation in favor of agricultural expansion that can be tracked all the way to the colonial period, drastically changed. The main cause for deforestation became urban e xpansion, mostly responsible for the removal of lowland and midland coffee agro-forests. On th e other hand, a significa nt trend in forest regeneration due to field ab andonment appeared; a trend that had not happened since the centralization and decline of indigenous populations during the early colonial period. Recent studies indicate that during th e period of 1979 2003, there was an increment of 23% in the forest cover of the Lempa basin. The distributi on of these new forests coincided with social, climatic, and geophysical variables that reduce the demand for marginal agricultural soils. Conclusion. Forest cover dynam ics in El Salvador have been anything but static or lin ear in nature. Since precolonial times both natural and anthropogenic factors have shaped the landscape. Periods of
35 forestation and deforestation have occurred in diff erent locations of the country according to the geophysical, social, and economic conditions present at the time (Figure 1-6). The anthropogenic factors that have affect ed the distribution of forests have changed throughout history. Soil fertility and climate influenced the distribution of indigenous populations and land use patterns. Large agricultural states deve loped in these areas, their expansions created periods of hi gh deforestation rates, while their collapse allowed forests to reclaim these lands. The coming of the Spanish culture also brought about drastic changes for the environment. While demand for products such as indigo and the introduction of cattle contributed to deforestation, disease and cult ivation of cacao, balsam and, la ter on coffee, contributed to conservation of tree cover in certain regions dur ing the colonial and post-colonial periods. During recent years, war, migration, economic po licies, and fluctuations in local and global markets are again driving change s in forest cover in El Salvador. Economic growth in the industrial and construction sector s, and migration due to war seem to have promoted the abandonment of fields and subsequent regeneratio n of forests triggering a process of forest resurgence. The same phenomena, however, are al so responsible for the re moval of tree cover in areas adjacent to cities as urban population grows. Even though high population dens ities are often related to deforestation in developing countries, there have been documented cases where industrial development, education, and migration have triggered forest increase in populated countries. Such are the cases of mid twentieth century Puerto Rico, ni neteenth century Europe, and Ea stern United States, and more recently in countries such as Mexico. This proce ss has been coined fores t transition, it implies a recovery of forest cover triggered by econom ic growth and industrialization. El Salvador
36 presents many similar conditions to Puerto Rico during its industrializa tion phase (rural-urban migration, external migration, remittances, in creasing levels of education, and economic integration to global markets). In an increasi ngly globalized world, it is possible, and often necessary, for land-scarce countries such as El Salv ador to rely on agricult ural imports and invest in labor intensive industries th ereby diminishing the demand for agricultural land. (Angelsen and Kaimowitz 1999; Grau et al. 2003; Ma ther et al. 1999; Klooster 2003). El Salvadors new forest mass is composed of multiple fragments of secondary vegetation, coffee agro forests, and to a lesser degree pine woodl ands in the north and mangroves in the coast. It is very likely that many of the same fire and drought resistant tree species that once populated the prehistoric savannahs are now re-colonizing those lands. The new rise of forests in the country bri ngs with it many questions. What is the species composition of these forests? What is the extent of these forests? How will low-priced grain from the United States affect land abandonment under the new free trade agreements? Is this a temporary phenomenon or is it a long term trend towards a forest transition? What exactly are the socioeconomic, climatic and geophysical fact ors driving this proces s? What is the impact these new forests are having on Salv adorian ecology and society? For conservationists in the re gion, a look at forest cover hist ory also brings up the question ( of ) which forest is the right fo rest to preserve or restore? Contemporar y conservation schemes in El Salvador are still based on the ideal of a static state of climax vegetation towards which forests tend to revert. But given that at any given point in hist ory vegetation has always been different, is there a partic ular state to which new forest should be brought up to or is this just a construct of our own perception of what they should be? Should these new forests resemble those present during the Holocen e period, the anthropogenic precolonial forests, or the
37 secondary forests resulting from the fall of na tive civilizations, or something different all together? Regardless of the answers to these questions, the study of forest cover history gives us a hopeful perspective that forests ar e resilient ecosystems, able to adapt and recover if given the opportunity. By understanding the conditions that prom ote the recovery of forests, we can better manage our resources and the impacts social change has on our environment. Countries undergoing forest transitions like El Salvador should focus not only on preserving what forests they have left but also in ma naging the new generation of forest s that are beginning to emerge.
38 Figure 1-1. General map of El Salvador. [S ource: SNET geographic Database, Ministry of Environment of El Salvador.]
39 Figure 1-2. Population distribution at the year 1550 in El Salvador [Found in: Colonial Archives, cited by Browning 1975.]
40 Figure 1-3. Population distributi on at the year 1770 in El Salvador. [Found in: Colonial Archives, cited by Browning 1975.]
41 Figure 1-4. Change in population densities El Salvador from 1550 to 1770. [Created by Author based on data from colonial Archives, cited by Browning 1975.]
42 0 1 2 3 4 5 6 175017751800182518501875190019251950197520002025 YearEstimated population (millions of inhabitants) Figure 1-5. Estimated historical population growth for El Salvador. [Found in Colonial Archives, National Population Census.]
43 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 10,000 bc2,000 bc800 ac1550 ac1770 ac1900 ac1970 ac2003 acYearPercentage forest cover Figure 1-6. Estimated historical fo rest cover and main events a ffecting its distribution for El Salvador. A) Pre-agricultural period, stable savannah-for est cover. B) Advent of Agriculture, period of defore station around fertile areas of the territory C) Fall of large city-states (San Andres, Tazumal, Cihuatan, among others) marked by decline in agriculture and forest regeneration D) Arrival of the Spanish, a period of deforestation due to new agricultural pr actices accompanied by reforestation due to relocation and depopulation due to disease and conflict E) Co lonial and Post Colonial Trend of agricultural expansion into margin al areas F) Civil war, marks a period of urbanization and migration, starts a process of forest transition with reforestation in marginal areas.[ Source: a Created by author.] A B C D E F
44 CHAPTER 2 PATTERNS OF FORESTATION AND DEFO R ESTATION IN EL SALVADOR: ADDING QUALITY TO QUANTITIES IN FOREST TRANSITION STUDIES. Introduction In the pas t decade, many studies have focused on forest transitions, a process which explains widespread reforest ation as a product of economic development and other socialeconomic conditions (Perz, 2007; Rudel et al ., 2005; Mather, 1992, 1999; Aide and Grau, 2004). According to forest transition th eory (FTT), one path towards reforestation is when development shifts human activities from agri culture to industry thus allowi ng fields to revert back to forestland. Both the United States and Europe e xperienced them during the industrial revolution (Mather et al. 1999); later on countries such as Puerto Rico during the 1940s (Grau et al, 2003), and parts of Mexico (Klooster, 2003), India (Foster and Rosenzweig, 2003) and South America (Rudel et al ., 2002; Batista and Rudel, 2006) have been found to be undergoing a similar processes in the last decades. New aspects of forest transitions are constantly being explored to increase our understanding of the process. For example, wh ile land-cover change may show in crements in forested areas, deforestation is often going on simultaneously (Sloan, 2008). Examining the simultaneity and spatial interaction of deforestati on and reforestation within forest transitions can help us better understand the process as a whole. Another aspect that can be added to enhance forest transition studies is forest composition. It is easy to oversimplify the term forest by ju st describing its quantit y, and assuming a uniform mass of vegetation in terms of composition and ecosystems. Secondary forest vegetation that colonizes abandoned lands is very different fr om older successional ve getation (Finegan, 1996; Grau et al., 1997; Guariguata and Ostertag, 2001;Lucas et al ., 2002; Chazdon, 2003) even if the regeneration is occurring in the same eco-zone as the one deforesta tion is happening in. The
45 difference in floristic composition is even more relevant if the regeneration process is going on in different ecosystems than the deforestation pro cess. In this case, the new forests will not be functional substitutes for the ones that are being lo st. This is particularly relevant in regions where a high diversity of ecosystems is concentrated in a very small area such as the case of El Salvador (Holdridge, 1975). In this study, El Salvador, specifically the Lempa River Basin (Fig. 1), was used as an example to illustrate the use of qualitative and sp atial characterization of forests to enhance the study of forest transitions. The Lempa basin provid es El Salvador with 33% of its electrical power (Lozano and Cottin, 2002) and it supplies about 66% of the countrys population with water. Average annual precipitation ranges from 1141 to 2886 mm (MAG, 1998; SNET, 2006). Elevations within the basin range from 0 to 2678 meters abov e sea level (NASA, 2000). The variation in topography and rainfall regimes make it an ecologically dive rse region. According to Holdridges life zones it contains 8 distinct ecological zones with several distinct ecological associations due to variation in soils. The objectives of this study are to answer the following questions: Is there a forest transition taking place in the Lemp a River Basin? If so, what is the magnitude and distribution of forestation and deforestation within the basin? And, what are the ecological and age characteristics of the Lempa basins forests? Forest Transition Theory and its Implication s for El Salvador. Forest transition is a term that was introduced in the early 1990s to describe the process by which a regions forest cover tends to increase as a result of socioeconomic conditions that affect land use. According to FTT, there are several path s that can lead a country to a forest transition. The first one is called the economic development path. The theory is that industrialization and urbanization take place as a result economic de velopment (Mather et al. 1999, Rudel et al.
46 2005). The second path for forest transition has been called the forest scarcity path. In this scenario, the high demand and low supply for fo rest products triggers an initiative by the government to establish large scale plantations to meet this demand (Mather, 1992; Drake, 1993; Grainger 1995). Finally, Rudel (2002) also menti ons a third path to fo rest transitions through globalization. The changes in glob al demands and supplies for products as well as migration and remittances are thought to induce changes in agri cultural production which may lead to increases in forest cover. According to FTT, forest cover tends to follow an inverted U shaped curve in time, also known as a Kuznets curve (Rudel et al., 2005) (Figure 2-2). Forest cover begins to decrease as agricultural frontiers begin to e xpand into new areas as populati on and economic growth begin in a country. FTT also states that the rate of deforestation eventually slows down and is followed by a gradual increase in forest c over as industrial activ ities such as construction, services and manufacturing replace agriculture as a means of livelihood. Although this is the ge neral principle of FTT, there are a diversity of other factors that may lead to alternate pathways of forest cover increase. Initial works in forest transitions focused on th e factors that lead to forest transitions in now developed countries such as the United St ates (Foster, 1992) and France (Andre, 1998; Mather et al ., 1999). More recent works on forest tran sition have focused on cases in the developing world where forest tran sitions are believed to be goi ng on. Unlike case studies in the past, developing world forest transitions seem to have a greater diversity of underlying causes which vary from case to case. War, free trade ag reements, international migration, environmental policy and diminishing soil fertility are only a fe w examples of these (Southworth and Tucker, 2001; Klooster, 2003; Grau et al. 2003; Hecht, 2005).
47 Buttrick (1971) cites three main reasons fo r agricultural land reverting back to forest land: 1) loss of fertility 2) changes in economic conditions rendering cultivation unprofitable and 3) discovery that certain lands are not suited to agricu lture. In El Salvador all these conditions are present and a fourth possible reason may be added to this list: armed conflict which triggered the abandonment of ag ricultural lands (Hecht et al, 2005; Hecht et al, 2007; Kandel, 2002). Another alternate mechanism of forest transitions is the effects of globalization (Rudel, 2002; Klooster 2003). In his paper, Rudel describe s globalization as destroying first nature and creating second nature. By this he means that in many countries affected by globalization older forest growth tends to be depleted while ma ny agricultural areas te nd to be abandoned giving way to secondary vegetation growth. This pathway is particularly relevant to El Salvador in the view of its open trade policies and its dependency on foreign income from remittances. Loss of soil fertility has been a recurrent issue in El Salvador The intensive cultivation of marginal lands with steep slopes, short fallows, a nd little or no use of eros ion control have led to large rates of soil loss in the past decade s (MAG-DGRNR-PAES-IICA, 1997). Yields of maize and beans have been reported to be declini ng in many rural areas of the country (Hecht, et al. 2007). At the same time, economic conditions favoring the traditional agriculture of the country have been on the decline. Maize and bean prices have had a steady downfa ll during the past three decades (PRISMA, 2002). At the same time, the real minimum salary for agricultural labor tended to decrease during the 1984-1994 period (F USADES, 1998). Coffee, an important source of temporary work for small farmers and landless laborers, also had a sharp decline in prices during the 1990 (PRISMA, 2002). These conditions ha ve made agricultural work in post-war El Salvador less attractive than it was in the past.
48 While the agricultural sector has declined, th e industrial and cons truction sectors have grown significantly in past years (FUSADES, 19 98) creating new jobs in urban and suburban settings. This instability of agricultural activities and openi ng of more attractive income opportunities has resulted in increas ed migration to urban centers and towards the US (Kandel, 2002). Migration is directly responsible for the decrease in popul ation in the countryside (Hecht et al ., 2005). Another effect of migration is the appearance of a new sour ce of income for rural families: remittances (Lungo et al., 1997). Over 21% of families in the countryside were reported to receive remittances from relatives in urban centers or the US (Briones et al ., 2005; Acevedo et al ., 2005). These remittances often allow households that receive them to live without the need to clear forests for cultivation and to invest the money in more profitable ventures such as small scale retail and services (Ka ndel, 2002; Andrade-Eekoff and Kandel, 1998; Rodriguez, 1999; Zilberg and Lungo, 1999; Allen et al. 2006). The effects of the 1980s civil war also had a strong influence on land use in El Salvador. Violence in certain areas caused mass populatio n displacements and increased out-migration (CNR, 1992; Kandel, 2002; CCP 2007). One common path for tropical forest transiti ons is called the hollow frontier effect (Rudel et al ., 2002; Sloan, 2008). This path to forest transition occurs when pioneers clear forestlands and later abandon it wh en fertility declines. After this they move forward abandoning the land and forest regeneration begins. This pa ttern, common in land ri ch countries of South America, does not seem to be the case for El Salvador. With a populat ion density of about 371 inhabitants/km2 and its shortage of agri cultural lands, the agricultural frontier probably stopped expanding some time in the mid-twentieth century (Castaneda, in review). However, it cannot be
49 said that the case is comparable to those of France or the US where a solid improvement in employment and income from industry lowered the demand for agricultural land. Instead, El Salvador seems to follow a pattern more similar that of Puerto Rico (Grau et al ., 2003) or Mexico (Klooster, 2003) where out-migration combin es with some industrial growth to lower the demand for land. Other studies (Hecht et al., 2005, Hecht et al. 2007) strongly support the hypothesis that a forest transition is occurring in El Salvador. Materials and Methods The spatial s cale of the analysis is at the wa tershed level, more specifically the Lempa River Basin which covers 9,868.45 km2 (approximately 47.6% of the country s total area). Areas of 57 x 57 m in areas were used as the main unit for the land cover analysis. The temporal scale covers the period from 1979-2003, divided into two sub periods: 1979-1990/1 and 1990/1-2003. Land-Cover Classification In order to answer the research questions, I ba sed the study on analysis of LANDSAT satellite im agery for the years 1979, 1990-91, and 2003. Two imag es per date were obtained to cover the majority of the study region which was spatially defined as the portion of the Lempa river basin within the political boundaries of El Salvador (areas not within the two images or with cloud cover were excluded from the study). In total th e study area encompasses 91.6% of the total area of the basin. These images were radiometrically calibrated, geo-referenced to the official digital road map of El Salvador (SNET, 2006) and cl assified into the follo wing land cover classes: forest, non-forest, and water. The process of classification was done combining a rule-based classification (Daniels, 2006; Lucas et al .; 2007; Sader et al ., 1995) with supervised classification (Jensen, 2005). The following defi nitions were used for taking the training samples:
50 Forest cover was defined as a surface cove red by a trees with at least 25% canopy cover and a height of 5 meters or more. This may incl ude man-made forests such as timber plantations, as well as densely planted agroforestry systems such as home gardens shade coffee plantations. Non-forest land cover was defined as all land covers not included in the previous category since it is not serve the purpose of the st udy to differentiate between them. Built areas, grasslands, pastures, crops, irri gated crops, bare fields, dry lava beds, and water will all be included as such. While these classes were not se parated for the analysis, the field data phase recorded them for possible future uses. For the classification process, 320 training sa mples were taken in the field using a GPS unit and a combination of randomized and oppor tunistic sampling. Th ese training samples incorporated all of the major la nd covers found in the basin. The training samples were randomly divided in half. The first half of the samples we re used for the classification process while the second half were used for an accuracy a ssessment of the land cover classification. The first step in classification was done through a rule based method based on the Normalized Differential Vegetation Index ( NDVI) (Jensen 2005). Examination of the NDVI histograms revealed that the image was composed of a bimodal distribu tion with a large amount of low NDVI surfaces and a smaller but we ll defined amount of high NDVI surfaces. Through field observation, it was determined that the high NDVI surfaces corresponded to evergreen vegetation and irrigated fields, while the lo w NDVI surfaces corresponded to seasonally dry vegetation and non-forest land covers. This patter ns corresponds to the dry season phenology of the local vegetation, where in many areas forests lose their leaves, and fields are left uncultivated until the rains arrive.
51 In order to separate these two types of surf aces, the image was divided into surfaces with high photosynthetic activity and surfaces with low photosynthetic activity. From a random sample of evergreen forest areas, it was determin ed that most of these forests had an NDVI of greater than 0.35. Hence this was used as a cut poin t to classify surfaces into high or low NDVI. The resulting classifications were used to mask the satellite image in order to isolate the two types of surfaces. The two resulting subset s were then classified separately using a supervised classification maximum likelihood algorithm. The high NDVI surfaces were classified into evergreen forests and irrigated fields (the two major land covers that composed this category). The low NDVI surfaces were cl assified into deciduous forests and bare/built surfaces (since the images used were taken dur ing the dry season, non-irrigated agriculture and pastures are included in the categ ory). Water was incorporated into the classification afterwards from a supervised classification of th e original image into water/non water. Further rule-based classification was used to refine the irrigation and evergreen forest classes by incorporating slope a nd irrigation layers as logical classifiers (no irrigation was allowed to exist above a slope of 5% a nd all high photosynthetic surfaces within known irrigation districts was re -classified as irrigated fields (non-forest). A final category of forest was created by addi ng the evergreen forests and the seasonally dry forest land cover classes. The rest we re classified as nonforest land cover. The classification method was conducted and te sted on the most recent image (2003). Once the accuracy of this map was tested, the same pr ocess was applied to the other images. In order to confirm the NDVI values were applicable, a sample of pixels from 1979, 1990 and 2003 taken from a protected area that remained forested throughout the stury period was compared using a comparison of means. Once it was confirmed th ere was no significant difference between the
52 NDVI values in this area, the same thresholds of NDVI were used to classify the other images. For the classification of these images, only sa mples that were land use was confirmed through owner information, local interviews or observation of vegetation age (tree ag e) to have remained unchanged throughout the study period were used for the classification. The period of study was divided into tw o sub-periods one from 1979 to 1990-91, and another from 1990-91 to 2003. The area analyzed wa s dictated by the available imagery for the given period. Only areas with images available fo r all dates within the sub-periods were used. As a result, the portion of the watershed analyzed for the first sub peri od is slightly smaller than the one used for the second sub-period. Areas that pres ented cloud cover in any of the images were also excluded from the analysis. Land Cover Change Analysis A change trajecto ry of the three classifications was created to detect changes in forest. cover for both periods. The result was a change trajectory map of areas that had undergone forestation, deforestation or no change for the period. These maps, in combination with the vegetation type classifications were used to estimate the age of forests in the country. Due to small pixel size (57 x 57 m) it is hard to visualize the spatial pa tterns of forest cover change in the region. In order to aid in analysis and visualizati on a forest cover density change map was created using ArcMAP software (E SRI, 2006). A circular moving window of 1784.150 m radius (10 km2) was used to estimate number of pixels per km2 under forest cover for each date. The results were multiplied by a factor of (57.125 2/100000) to convert pixels into km2 to obtain a percentage of forest cover for the region. Determining Autocorrelation and Hotspots fo r Reforestation and Deforestation In order to determine if there was any spatia l autocorrelation (Rogerson, 2001) for the patterns of forest cover change the land cover maps were further analyzed for spatial autocorrelation.
53 First a separate sub-set layer for deforestation and reforestation was creat ed for each sub-period in order to isolate each type of change. Then a local Morans I using FRAGSTATS software (McGrail et al ., 1995) was applied to each sub-set in orde r to determine whether the distribution of deforestation and reforestation was random or followed a spatial pattern. A queens case 20 pixel (1020m) moving window was used for this purpose. Forest Type and Age Once the to tal forest cover for the three dates was calculated, fo rested areas were classified according to Holdridges life zone classifi cation (Holdridge, 1978) using existing maps (Holdridge, 1975). Holdridge clas sifies life zones according to biotemperature, precipitation and potential evapotranspiration, he does not refer to actual vegetati on cover but to the type of vegetation that would exist in an area if left undisturbed. In order to covert Holdridges Life Zone map into actual vegetation covers, the la nd cover classification was combined with the existing eco-climatic maps by Holdridge in order to identify forest covers as specific types of ecosystems. In addition to Holdridges life zones, shade co ffee areas were classified as highland (above 1200 m above sea level), midland (between 800-1200m) and lowland (below 800). The distinction between shade coffee and forests was done using a decisi on based classification combining ancillary data from the official ag ricultural map for El Salvador (CIATPNUMA 1998), expert knowledge of the area, and spectral data from the sa tellite images. This was done for all three dates. The main assumption in determining forest ag e is that forest vegetation (with a canopy greater than 5 meters and a cover greater than 25%) has to be at least 5 years of age to be detected as such. Based on this assumption, vegeta tion present since 1979 was classified as being 29 years or older. Forests not present in 1979 but present throughout the 1990-2003 sub period
54 were classified as being between 18-24 years of age. Forests detected only until 2003 were assumed to be less 18-5 years old. Based on these categories, the 2003 forest cover was classified according to age. Results and Discussion Classifications and Accuracy Assessments The overall accuracy of the land cov er classification in 2003 was 92.85%. The error matrix and accuracy assessment are presented in Tables 2-1, 2-2. Forest Cover Distribution and Dynamics Over all, the total forested area in the Le m pa River basin increased during the study period (Figure 2-3) from 20% in 1979, to 31% in 1990-91 to 43% in 2003. The average annual rate of reforestation went from 1.53% during the period of 1979-1990, to 1.74% during 1990-2003. At the same time, the average annua l rate of deforestation also increased from 0.58% during 19791990, to 0.75% during 1990-2003. In 1979 most of the forest cover was concentr ated in the coffee growing regions in the central mountain range as well as pine-oak fo rests of the northern mountains. This pattern follows the trend that was started during the colo nial and post-colonial pe riods when the central and western valleys of the Lempa basin were cl eared for indigo, milpa (beans and maize) and cattle production (Browning, 1971). However, dur ing the period of 1990-2003 forest distribution changed, expanding into some areas previously dedi cated to agriculture and retreating from areas previously forested. Deforestation appears to be concentrated mostly along the densely populated central highlands and valleys (Figures 2-4 and 25). Even forest cover removal and regeneration are happening at the same time, the spatial distri bution and pattern of both forest cover loss and regeneration are very distinct.
55 Deforestation presents a high spatial autoco rrelation (Local Morans I = 0 0.87) and concentrates in intense hotspots around the main urban centers (Figure 2-6). There were also some deforestation foci in certain regions in the northern highlands. Reforestation on the other hand tended to happen with less spatial autocorr elation (Local Morans I = 0 0.26) spread throughout the eastern parts of the basin. The spatial distribution of Mo rans I indicates that defore station was a more focalized phenomenon than reforestation. This difference in clustering can be due to the nature of the factors underlying each trend. Deforestation is a direct result of human decisions, for example, the decision to systematically remove trees and urbanize or star t a new farm plot happens in very well defined spots and results from a direct and planned deci sion. The concentrated hotspots of deforestation around urban areas and norther n pine forests coincide with the th esis of urban growth. The direct expansion of cities and demand for constructio n materials is probably responsible for the removal of forest cover in those particular areas. The scattered distribution of reforestation, with significant but low values of spatial autocorrelation, points to a less defined, perhaps more complex set of drivers of change. Climatic variability, intensity of the armed conflict, lo w agricultural productivity, and migration are all existing conditions in these parts of the country which may contribute to field abandonment. Further studies are needed to determine which fa ctors have a statistical significance in driving forest cover changes in the region. Types of Forest and Forest Age. Overall dry, hum id subtropical and humid tropical forests te nded to increase (Figure 2-7). Other natural forest types such as lower montane subtropical forests (mostly composed by pineoak associations) and very humid montane subtr opical forests (Oak dominated highland forests)
56 tended to decrease in area. Coffee agroforests, whic h traditionally have accounted for most of the forest cover in the central highlands tended to decrease. Lowland coffee suffered the greater loss of area followed by midland coffee while highland coffee remained mo re stable. This stability of highland coffee may be due in part to its higher ma rket price, but also to the steep elevation and remote areas where it is grown. By examining the types of forest that have tended to increase we notice that dry forests have a particularly high rate of increase. These areas are perhap s more likely to be abandoned for their lower productivity, propens ity to droughts during El Nio years, and lower agricultural potential. On other hand, highland natural forests present the highest rates of deforestation. This means that although net forest cover is increa sing, endangered ecosystems such as pine-oak associations and cloud forests ar e still in danger of disappearing. Lowland and midland coffee are also losing area, mostly due to conversion to urban areas (Figure 28). Although this is not good for the local ecology, in term s of functionality, humid tropi cal and sub-tropical forests regenerating at high rates coul d eventually become ecological substitutes of the type of vegetation found in these agroforestry systems. The age composition in of the forests in the Lempa region (Figures 2-9 and 2-10) gives an idea of the successional stage most of these fo rests are found in. Of th e forest existing in 2003, 50.4% of the were estimated to be less than 18 years old, indicating that the period from about 1985 to 1997 was when most of the forest regeneration happened in the basin. It is important to mention that many of these younger forest successions are likely to be part of shifting cultivation systems under temporary fallow periods and not necessarily permanen t reforestation.
57 A total of 23.2% of the forests detected in 2003 have an estimated age of 18 to 29 years. The greater part of these forests are located in the hilly lowlands surround ing the inner valleys of the basin. Only about 24.4% of the forests present in 2003 were forests with an estimated age of more than 29 years, representing in great part the coffee agroforests which have been stable throughout the period. The rest of the area is mostly composed of highland forests (pine and oak) in the northern region. Only a small portion of natural forests in this age category were found in the lowlands. The age of forests also helps us understand th e role these new forest are playing in the regional ecology. The fact that most of the forest s detected are under 18 years of age means that the species of plants and animals they can house are limited to those adapted to early forest successions. Also, many of the Lempa basins ne w forests have regenerated from isolated fragments of forest which as Spellerberg (1996) found, may contribut e little to increasing biodiversity. An exception to this can be forest in proximity to old growth forest remnants (Zimmerman et al 2000; Holl 1999; Wijdeven and Kuzee 2000) where there is enough biodiversity of plants and animal s to re-colonize these new fore sts. In either case, forest transitions can help reduce fragmentation and prev ent from further loss of biodiversity (Schelhas and Greenberg 1996; Ferraz et al. 2003) Typically these forests have a high dens ity of shrub and vine biomass and a canopy between 5-9 m of height. Although in terms of infiltration and erosion control these forest provide important ecosystems se rvices, it may be some time before these forests can be considered as suitable habitats for conservation of biodiversity or source s of tropical hardwoods characteristic of more mature forests. Hi gh biomass and a pronounced dry season also make
58 these forests very vulnerable to fires which ma y also affect the flor istic composition found in them. The distribution of young forests, particularly dry tropical forest and humid sub tropical forests, might also affect their diversity. Since forest regenerati on is happening in places where there was little, if any forest in 1979, it is possible these new forests composed mainly of wind dispersed pioneer species. Future studies should be conducted regarding the relation of forest location with regards to older forest fragments and their floristic diversity. Overview and Future Trends The data in this study strongly supports th e hypothesis that a f orest transition is happening in the Lempa basin. This transition in particular is ve ry dynamic, with high rates of forestation and deforestation happeni ng at once. The pattern appearing in El Salvador fits in with Rudels second path to forest tr ansitions (2002) where the effects of globalization seem to be the main cause. In this case, massive deforestation is occurring with high spatial autocorrelation around the cities, while simultaneous secondary growth is appearing in a mo re dispersed, yet also spatially autocorelated pattern in the countrysi de. Although there has been some industrialization in the country, it seems that migration, remittances and global grain and coffee prices may have a higher influence on forest cover changes and population pattern s in this case. Rudels hypothesis is that gl obalization destroys older forests and promotes secondary forest regeneration. In the case of El Salvador, there were few old grow th forest left at the start of the study period. However the statement holds true in the sense that lowland coffee agroforests and highland pine forests were the oldest, more stable covers up at the beginning of the study period and are the ones that were mostly depleted. Secondary growth, on the other hand, occurred in previously deforested ar eas, often away from population centers.
59 Even though by 2003 the Lempa basin had more than double the forested area it had in 1979, highland forests which hold a large part of th e countrys biodiversity are still in danger of disappearing. Midland and lowland coffee agro forests in the central mountains which are essential for providing drinking water to the grow ing urban population and habitat for local fauna (Gallina et al., 1996; Cuella r, 1999; Fimanow, 2 001; Komar, 2003; Mendez and Bacon, 2005) are also in danger of disappearing due to urban sprawl. On the other hand, dry and semi-humid lowland forests in the eastern parts of the basin have a clear tendency to increase. Although these new forests are mostly composed of secondary vege tation less than 18 years old, they have a great poten tial for providing ecosystem se rvices to the country. For one part they help control erosion and sedimentation that goes into the Lempas hydroelectric dams. They can also have a good potential as wildlife ha bitat in the future or as part of regional initiatives such as the Mesoamerican Biologica l Corridor. As secondary forests with high biomass increase rates, they also have a good potential for the growing carbon offset market. If El Salvador follows the forest transition pa ttern of similar countries such as parts of Mexico and Puerto Rico, these dynamic changes are likely to stabilize an d both deforestation and reforestation rates should slow down. Future fore sts will likely be located in the hilly, dryer regions of the basin while the forests of the cen tral mountains will probably tend to yield to urban growth. Good agricultural land, with fertile soils, gentle slopes, a nd access to irrigation is likely to remain under such use or, in some cases transform into urban areas. Further research needs to be undertaken to fu rther understand the fore st transition process that is going on. What social, economic, geophysical and climatic factors are driving the forest transition process? What is the floristic compos ition of these forests? What is the statistical relation between forest cover changes and so cial, climatic and geophysical factors? How
60 permanent is this forest transition? What are the values these new fo rests contribute to the countrys economy as sources of ecosystem servi ces? What are the factors driving El Salvadors forest transition at commun ity or household levels?
61 Figure 2-1. Geographical lo cation of the Study region. Figure 2-2. Kuznet curve. Behaviour of forest cover in time according to FTT. Forest Cover Time
62 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 197919902003 YearPercentage of area under forest cover Figure 2-3. Total forest cove r registered for the study area during the period 1979-2003.
63 Figure 2-4. Forest cover change trajectory ove r the period 1979-2003 in the Lempa River basin. The three letters in the traj ectory stand each of type of forest cover present years under study (1979, 1990, and 2003 respectively). F = forest cover, N = No forest cover.
64 Figure 2-5. Changes in percent fo rest cover over the period 1979-2003 in the Lempa River basin.
65 Figure 2-6 Distribution of Spatia l Autocorrelation (Local Morans I) A) spatial autocorrelation for deforestation B) spatial au tocorrlation for reforestation. B A
66 1 10 100 1000 10000N o f o r e s t D r y F o r e s t L o w e r M o n t a n e S u b T r o p i c a l F o r e s t H u m i d S u b t r o p i c a l F o r e s t H u m i d T r o p i c a l F o r e s t V e r y H u m i d L o w e r M o n t a n e S u b T r o p i c V e r y h u m i d S u b T r o p i c a l F o r e s t V e r y H u m i d M o n t a n e S u b T r o p i c a l F o r e s t L o w l a n d C o f f e e M i d l a n d C o f f e e H i g h l a n d C o f f e eType of ForestArea (km2) 1979 1990 2003 Figure 2-7. Change in area of fore st cover according to forest type (Life zone) in the Lempa River basin for the dates 1979, 1990-91, and 2003.
67 Figure 2-8. Forest cover in the Lempa River Basin for the year 2003 according to forest type (life zone).
68 0.0% 10.0% 20.0% 30.0% 40.0% 50.0% 60.0%29 + 18 to 29 5 to 18 Forest agepercent of total forest detected in 2003 Figure 2-9. Total area of forest cover in the Le mpa River basin for 2003 according to forest age.
69 Figure 2-10. Distribution of forest cover in the Lempa River Basin according to forest age in the year 2003.
70 Table 2-1. Error matrix for the land cover classifi cation of forest cover in the Lempa River Basin of El Salvador. Field data Classified data Forest No-Forest Grand Total Forest 57 462 Non-Forest 7 86 92 Grand Total 64 90 154 Table 2-2. Accuracy assessment of the land cover classification of forest cover in the Lempa River Basin of El Salvador. users accuracy Forest 0.90 Non-forest 0.95 Producers accuracy Forest 0.93 Non-forest 0.93 overall accuracy 0.93 kappa 0.85
71 CHAPTER 3 FOREST REGENERATION IN EL SALV ADORS LEM PA RIVER BASIN: A QUANITATIVE APPROACH TO THE SOCIOECONOMIC DRIVERS OF FOREST TRANSITIONS Introduction Loss of forest cover has been a widesp read phenomenon in the developing world during past decades (Ehrhardt-Martinez 1998; Angelsen and Kaimowitz 1999; Cropper et al. 1999). However, just as human activities cause deforestation, sometimes social, economic or environmental factors propitiate the abandonment of land and allow forests to regenerate. The resurgence of forest cover at a large scale is ofte n linked with economic development through a process called forest transition (Mather 1992; Mather et al. 1999; Grainger 1995). Although the statistical relationship between socioeconomic factors and deforestation has been well studied (Skole & Tucker 1993; Mather et al. 1998; Marten and Lambin, 2000; Angelsen and Kaimowitz 1999), less is known about how socioeconomic aspect s affect the probabilities of reforestation. The objective of this study to a pproach the socioeconomic drivers of forest transitions from a quantitative point of view. It builds on past research in forest transitions and uses logistic regressions as a tool for test ing hypotheses regarding the soci oeconomic drivers of forest regeneration. For this purpose we focused on the ongoing forest transition that has been happening in the Lempa river basin of El Sa lvador (Figure 2-1) during the period 1979-2003 (Chapter 1). Forest Transition Theory Forest transition theory began with the study of reforestation in now developed countries such as the United States (Foste r 1992), France (A ndre 1998; Mather et al 1999) and Norway (Staaland et al 1998). These works describe forest tr ansition as a process by which economic development triggers an increase in forest cove r. According to general forest transition theory,
72 when countries or regions begin development, there is usually an expans ion of the agricultural frontier that results in a large-s cale removal of forest cover. However, as industry develops and the population becomes more educated, marginal ag ricultural activit ies tend to be abandoned in favor of better paying activitie s in service and industry. This causes a transition period when deforestation slows down and gradually begins to expand again as population abandons the countryside and begins to c oncentrate around urban centers (Foster & Rosenzweig 2003). Recently, forest transition studies have fo cused on transitions happening in developing countries (Rudel et al. 2002; Pe rz & Skole 2003; Sloan 2008). Most authors agree that forest transitions in these countries do not always follow the same pa ths that developed countries followed in the past. Different conditions such as migration, war, free trade agreements, tourism, and environmental policies have been found to trigger forest regene rations in developing countries (Southworth & Tucker 2001; Grau 20 03; Klooster 2003; Hecht et al. 2005; Babtitsta & Rudel 2006; Kull et al. 2007). The factors that stimulate increments in forest cover seem to be case specific (Aide & Grau 2004; Kull et al. 2007 ; Perz 2007) and it is hard to generalize a recipe for forest transitions. In many cases, de veloping countries may present partial forest transitions, where there may be increments in fore st cover, but not necessarily an improvement in socioeconomic conditions that can ensure thes e changes to be permanent (Rudel et al. 2002; Klooster 2003). SocioEconomic Factors and Fores t Transition in El Salvador. El Salvador is the sm allest and most densel y populated country in the mainland Americas. The country has long had the reputation of bei ng almost denuded of natural forests cover (FAO 2001, Terborgh 1999). This is partially true in th e sense that there are very few old-growth forests left. However the countrys shade coffee i ndustry has allowed for extensive tree cover to
73 remain standing, and in the past decades it has experienced a significant increase in secondary forest cover (Hecht, 2007; Chapter 2). The Lempa basin is of great importance to El Salvador. Covering nearly half of the countrys land area, it provides about 33% of its electrical power (Lozano and Cottin, 2002) and supplies nearly 66% of its drinking water. Av erage annual precipitation ranges from 1141 to 2886 mm (SNET, 2006). Elevations within the basin range from 0 to 2678 meters above sea level (NASA, 2000). Eight very di stinct life zones or ecosystems can be found within the basin (Holdridge, 1975), making it a very di verse region in spite of its sma ll size. Also, a large part of the countrys urban and industrial areas are found with in the basin. Just like many other forest transitions going on in the developing world, the factors underlying its increase in forest cover vary from the essential forest transition theory. Although there has been economic growth in the country, this alone may not account for the whole process. Buttrick (1971) mentions three reasons for ag ricultural land reverting back to forest land: 1) loss of fertility 2) changes in economic c onditions rendering cultivation unprofitable and 3) discovery that certain lands ar e not suited to agriculture. Current forest transition theory mentions three important paths to reforestati on. The first one is the Economic development path, which states that as a country develops low paying farm jobs ar e abandoned in favor of higher paying jobs in industry, leaving fields to regenerate in the count ryside. The second path has been labled the Forest scarcity model. Th is model explains foret cover recuperation by the establishment of forest plantations at large scale as a response to low availability of forest products. The third path mentioned by Rudell (20 02) involves the effect s of globalization and changing markets in the global economy that affect the way people manage the land.
74 With the exception of the second path, all these conditions are present in El Salvador to a certain degree, but one central event that s eems to have played a key role in changing Salvadorian society is the civi l war of the 1980s (Hecht 2005). This conflict began brewing in the 1970s and extended up until 1992. It transforme d El Salvador from a mainly agricultural society to one based on industry, services and remittances. During the war period, migration increased from the rural areas towa rds the cities as well as towards foreign countries (mainly the United St ates). The urbanization of Salvadorian society is believed to have resulted in the abandonment of large tracts of land (Hecht et al., 2005). By 1992 nearly 20% of El Salvadors population wa s living abroad (CNR 1992), the remittances these immigrants sent home soon became the la rgest source of foreign income for the country (Lungo et al. 1997). According to some st udies (Briones et al., 2005; Acevedo et al., 2005) Over 21% of families in the countryside receive remittances from relati ves in the US or urban centers. This steady flow of income has reduced the demand on land for subsistence agriculture, and allowed families to invest in more profita ble ventures such as commerce (Kandel, 2002; Andrade-Eekoff and Kandel, 1998; Rodriguez, 1999; Zilberg and Lungo, 1999; Allen et al. 2006). In addition to the armed conflict, other condi tions are indicative of a reduction of the demand for agricultural lands. The land distribution patterns that characterized the XX century forced many small producers to conduct subsis tence agriculture on marginal lands (Seligson 1995). The combination of short fallow periods, steep slopes, and little or no use of erosion control led to elevated rates of soil loss (M AG-DGRNR-PAES-IICA, 1997). As a result, yields of maize and beans, declined dramatically dur ing the war and post-war period (Hecht, et al. 2007). Basic grain and coffee pri ces, have also fallen during past years (PRISMA, 2002). In
75 terms of off farm labor, real minimum salary for agricultural labor decreased during the 19841994 period (FUSADES, 1998). Coffee, an important source of temporary work for small farmers and landless laborers, also had a shar p decline in prices during the 1990 (PRISMA, 2002). These conditions are likely to have disfavored farming, particularly in the more marginal lands. The trends of land use in El Salvador follow very similar patterns to those of Puerto Rico (Grau 2003) and parts of Mexico (Klooster 2003) wh ich have experienced forest transitions as results of migration, industriali zation and income from remittances. Past studies have quantified the increase in forest cover in the Lempa Ba sin from 20% to 43% during the period of 19792003 (Castaneda in press). Although many of the socioeconomic conditions in the country seem likely to be favoring a forest tran sition, no studies have yet confirme d if they have a statistically significant relation with the increase in forest cover. This study concentrates on the socioeconomic drivers of forest increments within El Salvadors Lempa river basin. This basin covers approximately 50% of the countrys territory and provides a significant part of its drinking water, arab le lands and electricity. Methods Land Cover Classification In order to obtain the data for this anal ysis, LANDSAT satellite im agery from the years 1979,1990 and 2003 were used to conduct a land-cover cl assification (Chapter 2). Only the parts of the basin located with in the boundaries of El Salvador were used. About 9% of the total area was excluded due to lack of satellite imagery or cloud cover. The temporal scope of the study was divided into two sub periods: 1979-1990 and 1990-2003. The images were classified into two main land covers Forest and NonForest. Forest being defined as an area of at least 57 x 57 m with a canopy cover of more than 25% and a canopy
76 height greater than 5 m. Besides natural forests, this classification also in cludes artificial forests such as timber plantations, and as densely pl anted agroforestry systems such as home gardens and shade coffee plantations. The overall accuracy of the classification used as input was 0.92 with a kappa statistic of 0.85. Since the socioeconomic data for the country was available only at the level of municipalities (smallest political unit), the data from the land c over classifications was aggregated to match this scale. The percentage of forest gain and lost for each municipality within the basin was calculated for each of the sub-periods. In tota l 148 municipalities were used in the study. Multiple Logistic Regressions The m ain statistical tool for analyzing the da ta was logistic regre ssion which have been used extensively in land-use change modeli ng (Vledkamp & Fresco 1996; Geoghegan et al. 2001; Agarwal et al. 2002). Through this method the probability of a binary dependent variable happening was related to the socioeconomic vari ables mentioned above. The logistic models were used to predict unusually large increases in forest cover (greater than 10%). This threshold was chosen based observation of the bimodal dist ribution of the data for the two study periods (Figure 3-2). The frequency distribution shows there is a group of munici palities in the study had mild or negative increments in forest cover and an other, very distinct one with increases above 515%. The cut point changes from 5% in 1979-1990 to 15% in 1990-2003. For analysis and comparison purposes a midpoint threshold of 10% wa s chosen to define the dependent variable. The basic logistic equation is the following: P = e (A1X1 + A2X2 + ...+ C) 1+ e (A1X1 + A2X2 + ...+ C) Where: P = Probability of event happening
77 e = exponential constant An = regression coefficients Xn = Independent variables C = regression constant Logistic regressions have the advantage of easily incorporation many types of data. There are no multicolinearty or normality assumptions for th e independent variables so it is not hard to combine dummy variables, continuous and discre te data, percent and index data (Menard 2002). One limitation to this methodology is that logi stic regressions cannot predict continuous variables, this limits the study of gradual changes over time. Dependent Variables A binary variable was created to rep resent forest cover changes in each municipality during each of the two sub-periods based on the land data. The values of these two variables were defined as follows: 1: if net increase in forest cover is > 0.10 0: if net increase in forest cover is <= 0.10 Where : Reforestation = percentage of pi xels detected as changing from non-forest to forest new forest cover for each municipality for the sub-period under study. Deforestation = percentage of pixels detected as changing from forest to non-forest for each municipality for the sub-period under study. Independent Variables. Existing data from various sources were co llected to create a geographic information system of variables representing the main drivers of forest cover dynamics in the country. The variables included in the study were selected from a series of factor s that previous works cite as potential catalysts of forest transitions in developing countries in the region (Aide & Grau 2004; Hecht 2005; Perz 2007). The selection of each variable was based on a specific hypothesis regarding its relation to forest cover regeneration. The variables evaluated were the following:
78 Intensity of Armed Conflict: It is hard to estimate the intensity of the civil war in a spatial sense since most of the military records of that period are not available to the public. As a proxy for this variable I used the United Nations official data for human rights violation reports (U.N. 1993) during the war at a departmental level (largest political division). Although this data has very coarse spatial resoluti on, and it is subject to errors su ch as unreported cases or overreported areas, it was considered th e best country-level data availa ble. It was expected that the municipalities located within departments of intense armed conflict would have a higher probability of forest regeneration due to the abandonment of land. Poverty : The data from the Poverty Map of El Sa lvador (Briones et al. 2005) was used to represent this variable. The poverty index in this map classifies each municipality as follows: 1 = low levels of extreme poverty 2 = moderate levels of extreme poverty 3 = high levels of extreme poverty 4 = severe levels of extreme poverty Forest transition theory relates forest rege neration economic growth, there fore it was hypothesized that there would be an inversel y proportional relation between the poverty index and the probability of forest cover increments. Changes in Population Density: The underlying hypothesis is that municipalities that presented higher population densities in general would be less likely to present increases in forest cover. For this total change in population dens ities (inhabitants per squared kilometer) were calculated for each municipality. Also rural and urban population densities were tested for any significant influence on th e dependent variables. Changes in Urban/ Rural Population Ratios: The hypothesis about this variable is that places with higher rural populations ar e less likely to have net gains in reforestation rates. This is because rural populations in El Salvador tend to depend on agriculture which places a higher
79 demand on land, thus lowering the probability of land being abandoned and left to regenerate forest cover. Income from Remittances per Household: It was hypothesized that municipalities with higher proportions of households receiving rem ittances are more likely to stop working their fields and hence allowing for ne t increases in forest cover. Changes in Male/Female Ratios : According to the literature gender labor divisions in the country indicate was expected that municipa lities with lower male/female (likely due to migration to the US) ratios would be more likely to present net gains in reforestation rates. Access to Markets: This variable was measured usi ng an accessibility index. This was calculated using the Arc GIS (ESRI 2006) function fo r cost-weighed distance. I used the distance from each point in the study region to the neares t urban center (market) weighed by the average slope the terrain presents along that rout. The in dex was fitted to a scale of 1 to 10 Table 3-1. After this, the average market accessibili ty index (MAI) was calculated for each municipality. The reason for using the MAI instead of the actual slope weighed distance was to facilitate the interpretation of the results sin ce the actual value of th e numbers was extremely high and lacked units. My hypothesis was that municipali ties with higher MAIs (that is lower access to markets) would be more likely to present net gains in forest cover than those with lower MAIs. This would be because the cost of producing agricultu ral and transporting prod ucts in inaccessible regions would be higher than those in more accessible regions, hence people would be more likely to migrate from these regions leaving their fields to regenerate into forests. Access to Main Road Network: As with markets, it was pr edicted that places with less access to main roads are more likely to be abandone d in favor of migrating to better lands, urban
80 areas or outside the country. As a measure of acce ssibility to roads the av erage value km of main roads per km2 were calculated for each munici pality using ArcMAP (ESRI 2006). Fuel Wood Consumption : It is often been assumed that fuel wood consumption is one of the reasons for deforestation (Cline-Cole et al. 1990). The census data shows that approximately 40% of Salvadorian homes within the basin rely on fuel wood for cooking (CNR 1992). My hypothesis in this case is that municipalities with higher fuel wood consumption will have a lower probability of having a net gain in fo rest cover than those with lower fuel wood consumption. These data were obtained from the populat ion censuses of 1972 and 1992 (CNR 1972; CNR 1992) as well as from official road a nd city maps (SNET 2005), and NASAs SRTM elevation data (2000). Results The spatial location of municipalities with ne t increases and d ecreases showed definite aggregation patterns for the period of 1979-1990, th e majority of municipalities presented increments greater than 10% in forest cover (Figure 3-3). Only a few of them, mostly concentrated in the far north and the southwestern regions pr esented net decreases. For the period of 1990-2003 reforestation seems to have more aggregated patterns (Figure 3-4). Deforestation concentrates mostly in the southw estern region of the basi n where there are more urban areas, and the new industr ial areas and good farmlands in the Zapotitn valley. The municipalities where negative or little increas e in forest cover was detected are located along the central mountain range. This is the area that historically has always tended to have the largest populations, the densest concentration of communities and, in the past century, has largely been dedicated to the production of lowland and midland coffee and irrigated agriculture (Browning, 1975). This same area has experienced a high population growth rate resulting in the
81 expansion of urban areas into lowland and midl and coffee as property values increase around the main population centers. It was observed that ne t forest increase was more common during the first sub period (1979-1990) th an the second (1990-2003). Another factor that coincides with this patt ern is the development of free trade industrial parks in the southwestern areas. These parks physically changed land use from coffee agroforests to industrial areas as well as attracting rural populations in search of employ ment to those regions. The result was the sprawl of formerly small rural communities into large urban centers. Another phenomena happening during this period was the spread of irrigation agriculture in the Zapotitn valley. This practice had been hampered during the war period due to violence and the risks that investments in infr astructure and machinery had du ring the war period. Once the war ended many irrigation districts we re re-established and new ones bui lt in areas that could support such farming practices.The result s of the logistic regression mode ls are presented in Tables 3-2, 3-3 and 3-4. The independent variables in this model (Table 3-2) were rejected if they had a P value greater than 0.05, however, since the civil war was such an influent ial factor during this period it was decided to include the war intensity variable for this analysis. Its P value for this variable was 0.09, perhaps because of its coarser spatial resolution, but it does give insight into the reasons for forest regeneration during this period For this model, the overall -2 log likelihood was 65.71, the Cox & Snell R squared was 0.51 and the Nagelkerke R squared value was 0.74. For the sub-period 1990-2003 two models were chosen out of 13. The reason for this is that they both provide useful in formation about different variable s that have significant effects on forest regeneration. The first model is the simplest one (Table 3-3).
82 The -2 log likelihood ratio for this mode l was 71.13, the Cox & Snell R squared was 0.50, the Nagelkerke R squared value was 0.72 a nd an overall prediction accuracy of 90.5%. This model indicates that the places with high po verty indices, high income from remittances and low road densities were the ones most likely to present fo rest regeneration. The alternate model (Table 3-4) is less simp le than the model on Table 3-3 but includes different variables that help explain the process of forest regeneration. The -2 log likelihood ratio for this mode l was 65.12, the Cox & Snell R squared was 0.52, the Nagelkerke R squared value was 0.75 a nd an overall prediction accuracy of 89.8%. This model indicates that the places with high consumption of fuel wood, high income from remittances and high decreases in rural population densities were more likely to present forest regeneration. On the othe r hand municipalities with high population densities, low road densities and bad access to markets had less probability of poverty indices, high income from remittances and low road densities were less lik ely to present increments in forest cover. The difference in the models may be due to the fact that certain variables may present high levels of multicolinearity and thus be proxies for the same driving factor. For example, rural poverty is very closely linked to fire wood consumption and market access. In this way, place with high indices of rural poverty are probably also the same ones with high percentage of fuel wood consumption and bad accessibili ty, resulting in pa rallel models with similar explanatory power. Discussion The hypothesis that the armed conflict that aff ected El Salvador dur ing the 1980s promoted the massive abandonment and subsequent forest regeneration in regions of high violence is partially supported (P = 0.10) for the sub-period of 1979 to 1990. The municipalities belonging to the departments with higher number of huma n rights violations reported due to war had a
83 higher probability of having an increase in net fo rest cover than those in less violent areas. I believe the low significance of this variable is mostly due to the poor sp atial resolution of the data used. In stead of having municipal data, depa rtmental data was used. If this data could be refined it is very likely that both the magnitude and significance of this factor would increase and better explain forest regeneration in the count ry. War had no significant effect for the period 1990-2003. These results are completely opposite to t hose presented by FAO (2000), where war is listed as one of the main causes of deforestation in the country. Th is study suggests that war left previously protected areas open to invasions and forest cover rem oval. Data indicates that there was very little forest lost in most protected ar eas during this period. The da ta tends to agree more with Kaimowitz and Faune (2003) in that wa r triggered a massive depopulation in the countryside triggeri ng forest regrowth The behavior of poverty as a predictor variable helps us unders tand one important aspect of the dynamics of the forest transition process. One of the basic principles of forest transition theory is that economic development leads to increm ents in forest cover. This can be said to be true at a country level, where the GNP per capit a has increased with dir ect relation to forest cover. However, if we look at the local distributio n of forest regeneration, we find that there are significantly higher probab ilities of reforestation in the ar eas with higher poverty indices. This means that the process of forest transition in the Lempa basin is a result of the concentration of wealth and population around urban centers while poorer regions tend to revert back to forests. This change in the spatial distri bution of wealth can be said to be the driving mechanism that reduces the demand for marginal ag ricultural lands and allows fore st regeneration. Therefore, at a municipal scale, the hypothesis that the leve l of poverty is inversel y proportional to the
84 probability of forest cover increments is rejected. Increases in income in urban areas are what is causing migration and land abandonment, but the mani festation of forest regeneration happens in the poorer municipalities which are being abandoned. Although we can say this is partially similar to R udells first path to forest transitions, it is not valid to say El Salvadors forests are increasi ng because of general increments in wealth. It is more likely these results are more related to the effects of globaliza tion, and hence forest regeneration is not neces sarily related to increases in wealth, but in changes in land use due to market changes and migration. The third hypothesis stated th at changes in population densi ties within municipalities for the period 1972-92 would have a negative effect on the probability of forest regenerating. The regression analyses showed that total rural popula tion had a negative effect on the probability of forest regeneration happening. Municipalities where rural pop ulation decreased from 1972-1992 had a significantly higher probabili ty of gaining forest cover. Neither total population nor urban population for 1972 nor 1992 had a significant influence on the dependent variable. Since most of El Salvadors rural populati on obtains its income from agricu lture, this evid ence supports the hypothesis that one of the main reasons behind fo rest regeneration is the lowering of the demand for agricultural lands. In places where land is fe rtile enough to support large rural populations, it is unlikely that land will be left fallow to regenerate forests. The hypothesis that municipalities with highe r incomes from remittances would be more likely to show net increments in forest cover is strongly supported by the data for both subperiods. The income from remittances may be alle viating the need for cul tivation, especially in less productive areas. The data in this study supports findings which suggest the money from remittances goes to education, small businesses or consumer goods rather than into agriculture.
85 Another possible explanation is that the peopl e sending in the remittances are mostly the economically active portion of the population (m ales between 18 and 40). These people would normally be the ones working the fields if they had not migrated. Instea d the people left behind are usually economically inactive po pulation (the children and the elde rly) and to a lesser degree economically active females who traditionally do not dedicate themselves to working agricultural fiel ds (Allen 2006). Allen et al. (2006) found in their work that female-lead households were less likely to clear forests than male-lead households, how ever the data do no support this hypothesis. Male/female ratios did not prove to be significant in determini ng the probability of net forest gains in this study at the munici pal level It is likely that at a household or community level, gender composition could have a more sign ificant effect over land use decisions. Accessibility is definitely an important fact or in forest cover re generation during both subperiods. The MAI and the density of main roads had a negative effect on the probability of net forest increase. This means that the municipalities with more exte nsive road networks and easier access to markets (due to flat topography and dist ance) were less likely to present net gains in forest cover. In regi ons with good access transp ortation costs are lowe r making agriculture a more attractive option. In municipa lities harsher terrain and harder access to markets, migration and remittances might be a more attractive option than low profit or subsistence crops. In this case the hypotheses that road density and market accessibility have a significant influence on the probabilities of forest cover regene ration is also st rongly supported. Finally, the hypothesis that fuel wood consumption lowers the probability of forest cover regeneration was rejected. Fuel wood consumption turned out to have a significant positive relationship with the probability of forest cove r increase. Places where people consumed more
86 fuel wood were the ones where increases in fo rest cover were more likely to happen. This indicates that the use of fuel w ood is not directly responsible fo r significant removal of forest cover. It could be that fuel wood is being used in these places due to its abundance as a result of secondary forest regeneration, in which case forest regeneration w ould be a driver of fuel wood consumption and not vice versa. The effects of fuel wood collection are difficult to measure at the scale of this study and a discrete land-cover classes. Selective harv esting of good fuel woods may result in a diminishing diversity of species and the loss of tree dens ity in forests which do not necessarily lead to the removal of the forest (Klooster 2003). In tropical forests not all tree species are good for fuel wood, so extractive practices tend to be more selective, leaving non useful species standing. Also, the cyclical pruning of shade coffee in many areas is a source of abundant fuel wood. This practice is done to favor coffee during th e rainy season and results in a steady, renewable source of fuel wood. This practice may also explain why forest cover remains even when fuel wood consumption is high. Perhaps the use of continuous measures of forest cover or fuzzy logic land cover classifications may reveal the im pacts this factor has on forest composition. Other non-spatial socio economic fa ctors are also likely to have an effect on forest cover gains or losses. Grain prices for example, are ge neralized for all the country, and they depend on more global phenomena that show no variation at the spatial scale that this study encompasses. During the study period there were several downfa lls in both of these pr oducts prices (Hecht 2005. Decreases in grain prices, as econometric theory would indicate (Hubaceck & Vazquez 2002), are likely to indicate the lo ss of profitability of agricultural lands, particularly those where access increases the cost of producti on hence making them more likely to be left fallow as people invest their time in other inco me opportunities (Butrick, 1971).
87 It is important to note that Rudels second path to forest transitions is discarded. Information obtained from the Ministry of Agricultu re (2005) indicates that there are a little over 2000 hectares of forest plantations in the country. Hence the increases in forest cover detected are not due to this factor. A possi ble explanation for this is the small size of the country allows for low priced timber to come in from Guatem ala and Honduras without having to rely on local plantations. Conclusions The results of this study indicate that the influence socioe conomic factors have on forest transitions can be accurately quantified through statistical analyses. Even if the socioeconomic factors that drive forest cover regeneration are different for count ry, logistic regressions offer a way of quantifying the significance of factors as well as testing hypotheses regarding the drivers of forest transitions. The fact that remittances, migration and war have played such a significant role in the forest transition process puts into question the perpetuity of this surge in forest cover. An armed conflict such as the one in the 1980s (hopefully ) is not likely to happen again. If global economics or US migration politic s reduce the current rates of mi gration and the income from remittances, what will happen to the recently regained forest cover? Will there be a second agricultural expansion into these depleted soils? Unless the El Salvadors government creates the necessary conditions to employ its large and gr owing population, these forests could become an ephemeral phenomenon and disappear once again. Another factor that may reduce forest rege neration is the improvement in accessibility. A large road project funded by the US government is already underway in the northern highlands as part of a development fund.
88 On the other hand, one factor that may increas e the probability of reforestation is the influence forest cover in coming years is the Central American Free Trade Agreement (CAFTA) with the US. Given the similarities in rural livelihoods between El Salvador and Mexico it is possible that the influx of cheap maize may dimini sh local production as it did in that country (Klooster 2003) and lower the demand for farmland and continue the trend in gr ain prices so far. The results of this study indica te that, while a general increase in wealth triggers forest transitions at a country level, the places where fo rest tends to regenerate are the poorest ones. It is important that sustainable uses for this ne w forests such as timber and nontimber forest production, tourism or payment for ecosystem servi ces be developed to allow the people in the newly forested areas to benefit from them. Future quantitative studies at a community of household level could help further understand the influence of socioeconomic factors on forest transitions. Studi es at this scale may also help determine the direct effects of world markets, free trade agreements, and grain prices have on rural livelihoods and how they affect land cover decisions. Also future studies should include separate models of the dr ivers of forestation and deforest ation to add more detail to the causes of forest cover change.
89 Figure 3-1. Study area: Departments, municipalities and extension of the Lempa River Basin in El Salvador.
90 Figure 3-2. Frequency distribution of municipalitie s according to increases in forest cover. A) 1979 to 1990. B) 1990 to 2003. Percentage of increase in forest cover Percent increase in forest cover Frequency B A Percent of increase in forest cover Frequency
91 Figure 3-3. Changes in net forest cover by municipality for the period 1979 to 1990 in the Lempa River basin.
92 Figure 3-4. Changes in net forest cover by municipality for the period 1990 to 2003 in the Lempa River basin.
93 Table 3-1. Market accessibility Index values. Slope weighed distance to markets Market Accessibility Index (MAI) 0 1,316,757 1 1,316,757 2,366,514 2 2,633,514 3,950,2713 3,950,271 5,267,0284 5,267,028 6,583,7855 6,583,785 7,900,5426 7,900,542 9,217,2997 9,217,299 10,534,0568 10,534,056 11,850,8139 11,850,813 13,167,57110 Table 3-2. Best logistic regression model summary for the probability of net forest gain in the Lempa River Basin for the period of 1979 1990. Variable B S.E. Wald Sig. Exp(B) Road Density (km/km2) -1.278 .354 13.053 .000 .279 Market Accessibility Index (MAI) -4.087 1.684 5.886 .015 .017 Rural population density in 1972 (inhabitants/ km2) -.015 .006 5.433 .020 .985 War intensity .001 .001 2.848 .091 1.001 Decrease in rural population density (1972-1992) (inhabitants/ km2) 2.888 1.135 6.474 .011 17.954 Percentage of households receiving remittances 8.483 4.167 4.144 .042 4831.312 Constant 8.597 2.953 8.477 .004 5417.272 3-3. Logistic model summary for the probability of net forest gain based on socioeconomic data for 148 municipalities in th e Lempa River Basin for the period of 1990 to 2003. Variable B S.E. Wald Sig. Exp(B) Poverty Index 1.238 .395 9.810 .002 3.450 Percent of households receiving remittances 9.908 3.990 6.168 .013 20093.744 Road Density (km/km2) -1.074 .292 13.493 .000 .342 Constant -.262 1.192 .048 .826 .770
94 Table 3-4. Second Logistic model summary for the probability of net forest gain based on socioeconomic data for 148 municipalities in the Lempa River Basin for the period of 1990 to 2003. Variable B S.E. Wald Sig. Exp(B) Percent of households using fuelwood 4.178 2.081 4.030 .045 65.244 Road Density (km/km2) -.994 .328 9.165 .002 .370 Market Accessibility Index (MAI) -3.182 1.506 4.468 .035 .041 Rural population density (inhabitants/ km2) -.017 .006 6.945 .008 .983 Percent of households receiving remittances 8.322 4.151 4.019 .045 4113.478 Decreases in rural population density (inhabitants/ km2) 2.571 1.052 5.973 .015 13.081 Constant 4.226 3.065 1.901 .168 68.471
95 CHAPTER 4 APPLICATIONS OF GIS-BASED LOGISTIC M ODELS TO EXAMINING THE PHYSICAL DRIVERS OF FOREST COVER CHANGE AND SCENARIO BUILDING IN THE LEMPA BASIN OF EL SALVADOR. Introduction El Salvador is the sm allest, most densely populated country of the mainland Americas. In the past, the countrys demographi c conditions have put it s natural resources, particularly forest cover under great stress (FAO 2001). The general trend up until th e 1970s was an expansion of the agrarian frontier for the cu ltivation of subsistence and cash crops (sugar, indigo, cotton, maize, sorghum, beans) resulting in corresponding deforestation. There are two exceptions to this trend, one is the cultivation of shade coffee, mostly in the central volcanic highlands. This crop has allowe d a relatively diverse canopy to remain in this area (PRISMA 2002, Mendez and Bacon 2005). The other exception is the remnant forest patches of the northern highlands. These lands, characterized by the presence of pine-oak forests, have very poor soils (MARN 1998) and have ge nerally remained sparsely populated and with more forest cover than the rest of the country (Browning 1975). Forest cover is essential for the countrys ecol ogy in several ways. El Salvadors climate is characterized by a well marked dry season and a copious rainy season. Forest cover increases infiltration thus regulating the flow of rivers avoiding excessive floodi ng during the rainy season and drought during the dry season (Cuellar 1999). Similarly forest cover diminishes erosion, a problem that has lead to soil degradation in many parts of the country as well as sedimentation of major hydroelectric projects. Forests also provid e fuel wood and non-timber forest products that help support the livelihoods of subsistence farmer s. Forests and coffee agroforests also play an important part in providing habita t to local and migratory wildlife that exists in the country.
96 Forest Cover Change and Forest Transitions in El Salvador In the past d ecades, a large amount of secondary forest regeneration has been reported in the country (Hecht 2007, Chapter 2). This represents a reversal in the trend of forest cover dynamics that has existed in the country for the past two hundred years (Chapter 1). The trend towards forest recuperation may be happening due to a forest transition process. The term forest transition was originally in troduced in the 1990s to describe the process by which a regions forest cover tends to increase as economic development leads to industrialization and urbanization (Mather 1992; Drake 1993; Garinger 1995). According this theory the rate of deforestation in developing ec onomies tends to slow down as quality of life improves, and is then followed by a gradual increas e in forest cover as the economy shifts from agriculture to industry. This process is calle d the economic development path to forest transitions (Rudel 2005). The second established pa th is the forest scar city path (Foster and Rosenzweig 2003). In this case forest cover rege neration is triggered by a high demand and low supply for forest products. The increments in forest cover originate from the establishment of forest plantations to meet this demand is the main driver of forest cover increase. In this case the government of the country generally plays a central role in inve stigating plantations. According to Rudel (2002) a third path that may lad to forest transitions is the influence of economic globalization. Rudel explains that globa lization changes the distribution of productive activities in developing countries leading to removal of older forests and the regeneration of secondary forests after due to land abandonment. Unlike the first path, this path does not necessarily involve an improvement of living cond itions. Both the first and third paths seem to partially apply to El Salvador. As with most land use/cover ch anges, the reasons for these processes are often a complex combination of social, economic, climatic and geos patial factors that inte ract to influence human
97 decisions on land use at differe nt spatial and temporal scales (Veldkamp and Lambin 2001). In terms of forest cover, these effects result in a combination of forestation and deforestation processes which happen in different magnitudes and simultaneously, yet with different spatial distributions (Sloan 2008, Chapter 2). Past research (Hecht 2005, H echt 2007, Chapter 3) has identif ied several socioeconomic factors specific to El Salvador as having an influence on the forest transition process. Among them are the 1980s armed conflict, the urbaniza tion of population due to industrialization, the income from remittances from outside the c ountry, the reduction of rural population due to migration are among the factors th at have contributed to this forest transition (Hecht 2005; Castaneda in press). Other economic factors that have had an effect in the region (Mexico) are the fall of grain and beef prices due to regi onal trade treaties or lo wer production costs in neighboring countries, affect the land use deci sions farmers make (Geoghegan et al 2001, Klooster 2003). Since livelihood and socioeconomic conditions in El Salvador are similar to those in these areas of Mexico, it is likely they will have the sa me effects on its forest cover. Logistic Regressions in Land Cover Change Studies One m ethod that has proved to be useful for explaining forest cover change, particularly when studying a single land cover type is spatially explicit logistic re gression (Geogheghan et al 2001, Serneels and Lambin 2001; Veldkamp and Fresco 2001;Verburg et al 2002 ). Through the use of this method it is possible to determine what is the magnitude and significance of the influence of external variables on land cover changes. The use of GIS technologies allows for implementation of these statistical models into sp atially explicit data in the form of probability maps. These maps have been used for studi es such as monitoring wildlife populations (Maldenoff et al 1999), risks of landslides (Ayalew and Yamagishi 2003) and probabilities of karstification (Lamelas et al 2007) among others. The visualizat ion of the pattern of land use
98 change probabilities allows for a better represen tation of how land cover pr ocesses work. In this paper I have incorporated the re sults of logistic re gression into mapping the probabilities of reforestation and deforestation happening at a regional level under different possible future scenarios. Spatial Scale of the Study The area of study for this project is the Lem pa River basin, which covers roughly half of El Salvador (9868 km2) of which only 9041 km2 were used due to lack of satellite imagery for some areas). The temporal scale of the study is from 1979-2003, and the spatial scale is based on cells of 3314.3 m2 (57.57 x 57.57 m) (Figure 4-1). Research Questions. The two m ain research question for this st udy were: what are the effects of geophysical factors on the spatial distribution of forestation and deforestation in the Lempa Basin? And what are the strengths and weaknesses of applying logistic models as a tool for future scenario construction? Methodology General methods A land cover classification for the years 1979, 1990-91 and 2003 was used as the base for this research was the use of land cover maps for the years for the Lempa river basin (Chapter 2). The original data came from Landsat images for the study dates. These images were classified using a combination of supervised and expert based classification (J ensen 2005, Daniels, 2006; Lucas et al .; 2007; Sader et al ., 1995). The land cover of the Lemp a landscape in the images was classified as forest and non-forest pixels. Forest cover was defi ned as a surface covered by trees with at least 25% canopy cover a nd a height of 5 or more meters. This may include man-made forests such as timber plantations, as well as de nsely planted agroforestry systems such as home
99 gardens shade coffee plantations. Each pixel in the map was 57.57 x 57.57 m, the original images for 190, 1991 and 2003 were resampled to match the resolution of the 1979 image. The Overall accuracy of the map was tested and found to be 92.85% with a Kappa statistic of 0.85. A land cover trajectory was then created using the classification for the three dates (Jensen 2005). This trajectory identified where land cove increased, decr eased or remained unchanged within the study area. This information was then used to create the dependent variables used for the logistic regression. Dependent Variables Eight dependent variables were analyzed based on the land cover trajectory. The first three variables were the proba bility of unforested land becoming forest for the periods 19791990, 1990-2003 and 1979-2003. The next three variables related to deforestation, that is the probability of forested land loosing its fo rest cover for the periods 1979-1990, 1990-2003 and 1979-2003. The last two variables we re the probability of land remaining under forest cover throughout the entire period (1979-2003) and th e probability of land remaining unforested throughout the entire study period. Independent Variables In order to explain the dynam i cs of the dependent variables, key geo-spatial factors were used as predictor variables. These were: 1. Topography (slope, elevation) data were obtained from the Nasa Shuttle topography mission (NASA 2000). Topography plays an importa nt role in land use decision, usually higher lands are less accessible to farmers a nd irrigation, or present different climatic conditions that change peoples agricultural production strategies (Ziervogel 2006). Similarly, areas with high slope s are harder to manage and te nd to have a reduced soil fertility due to erosion. 2. Mean annual precipitation. These data were obtained from the SNET (Sistema Nacional de Estudios territoriales) in El Sa lvadors ministry of Environment. Rainfall affects the kinds of crops and livestock that an area can support. In seasonally dry regi ons like El Salvador,
100 it also defines the length of th e production period for rain-fed agriculture. Chap ter 5 of this study will go more in dept h on this important aspect 3. Soil quality and farming potential (Land Use Capability index, irrigation potential, coffee growing potential) Data on Land Use Capabili ty (Klingebiel and Montgomery 1961) was obtained from the Ministry of Agriculture of El Salvador. This i ndex classifies lands according to their agricultural potential wher e class I indcates the highest potential and class VIII the lowest potential. The Irrigation potential was determined by estimated a slope-weighed distance to major water bodies, this is assuming th at irrigation is easier for lands with access to water in flatter topography. Areas with gentle slopes and a higher potential for irrigation can be more productive fo r agriculture, hence it is a factor that will likely affect land use decisions. 4. Accessibility and urbanization (d ensity of roads, density of main roads, distance to main roads, distance to urban areas). Data for thes e variables was derived from official road maps for El Salvador (SNET 2005) and NAS A topography data (2000), using Arc MAP (Esri 2006) software to create distance layers. Urban expansion is one of the major trends in land use change in El Salvadors central valley. The regions where the main cities are expanding are located in traditional lowland, and midland coffee-growing regions, hence a change from forest cover to built surfaces is likely to happen. Roads are very important to land cover changes since accessibility is often an important factor in land cover changes (Thailand, camerron, Indonesia). 5. Regional ecology (Occurrence of fires and distance to existing forests) Fire data was obtained from AVHRR (USGS 2005) data for the years 1998-2000 and MODIS (NASA 2005) rapid fire data for the years 2000 to 2003 and processed using Arc MAP density tool (Esri 2006). Forest fires are a regular occurrence within the study area. Being a widespread phenomenon that involves the dire ct burning of forest biomass, it is likely that they are affecting forest cover change The distance to existing fore sts was calculated based on the classifications executed for the year 1979 and 1990 using the Arc MAP distance tool. Forest regeneration literature emphasizes the importance of remnant fo rests as sources of germplasm for forest regeneration (Holl et al .; Aide et al 1996), hence this variable was included to determine whether this was an important factor in the study region. 6. Protected area system. The data was obtained fr om SNET at the Ministry of Environment for El Salvador. A dummy vari able was used to determine if management by the National System of Protected Areas had a significant effect on forest cover changes throughout the study period A logistic regression was carried out for each of the dependent variables. A random sample of 29,466 pixels (about 1.2%) was used to construct the models. The predictor variables were chosen according to their theoretical importance as well as to their significance in predicting the dependent variable. Regre ssions were run using SPSS.
101 Nagelkerkes pseudo r2 value (reference) was used to estimate the degree to which the variation in probabilities was pred icted by the model. It is importa nt to note this value is not directly comparable to the r2 values in linear regression. The Chi squared value was used to evaluate the overall significan ce of the models. The overall pr ediction accuracy was used to compare models amongst each other in order to c hoose the best ones. This value estimates how effectively the model predicted the independent variable using a value of 0.50 probability as a threshold for determining whether the change in question happened or not Construction of Probability Maps. Once significant m odels were constructed, thes e were converted into spatial probability maps by applying the model to continuous raster data representing each of the dependent variables in the logistic equa tion (Equation 4-1). This is done by creating a raster data set representing each of the independent variables in the model. Using these layers as an input, the model is applied to every cell within the st udy area. The result is a spatially explicit representation of the probabilities of change for each cell in the map. The probability maps for deforestation and reforestation were calculated in this way. The pur pose of these maps is to allow a spatial visualization of how each phenomenon happened during the study period within the study region. Equation 4-1. Logistic equation. P = e(ax+bx+cx++C) 1 + e(ax+bx+cx++C) Scenario Construction. In order to illustrate the a pplica tions for the future, tw o possible scenarios (Peterson et al 2003) were constructed for the period 2003-2020. Scenario 1 is the default scenario. It assumes the same trend of the past decade continues with no significant change in the driver variables. It
102 applies the same models for forestation a nd deforestation for the period 1979-2003 in a projection for the future. The Scenario 2 assumes all conditions remain the same and applies the deforestation and reforestation models taking in to account the construction of the new northern transversal road that is planned to be built in the coming years (2009-2010). The resulting scenarios were compared as to the potential area of forest regeneration and deforestation that would happen. A threshold of 0.50 was used. Any cell with a probability above this value was considered as an indicator that a change would happen in the land cover. The estimated areas of forest lost and ga ined for each scenario were compared. Logistic Models for Forest Cover Change The results for the regression m odels all turned out to be significant (Table 4-1), and they indicate that probability of fo rest cover change can accurately be modeled through the use of geospatial variables in the study region. The logistic models indicate th ere are highly significant rela tions between the probability of forest cover being lost and geospatial f actors for the different study periods. The pseudo r2 values (0-170.21) indicate that there is a gr eat deal of variation in the probabilities of deforestation happening that is not explained by the model. However the fraction of the variation explained is highly significant as chi squared va lues indicate. Similarly the prediction accuracy of the models is also high. The three models of reforestation also preset a similar pattern with pseudo r2 values between 0.09 and 0.23 (the lowest appearing for the period of 1979-2009). The low pseudo r2 value for the first sub-period may be due to the fact that most of the reforestation was registered until the second period. Unlike de forestation which was f ound with higher spatial autocorrelation and happens in a shorter period of time, forest cover regeneration happens gradually and in a less auto-c orrelated pattern, which may re quire longer study periods to accurately model.
103 This combination of statistics indicates that although the models ar e highly significant, there are likely other important factors determining the proba bility of deforestation and reforestation that were not taken into account in these models. However the factors that were included definitely have a significant effect on it. The models regarding the probabili ty of no change in forest cove r tended to have a higher pseudo r2 values, 0.38 for permanent forest cover and 0. 40 for permanent non-forest cover. The models also presented high prediction accuracy, 92.2% for permanent forest cover and 75.1% for permanent non-forest cover, as well as highly si gnificant chi squared valu es. The combination of statistics indicate that the inde pendent variables used in the m odels are accurately capturing a large part of the variation in the data and signi ficantly predicting where forest cover conditions have remained stable. Deforestation The probability of deforestation increased mostly in the vicinity of major urban centers (Table 4-2). During the study period a large increase in urban popul ation (CNR 1992) was recorded by the 2 national census, and, accordin g to the ministry of economy, the tendency continues into the present. This relation follows a logarithmic tr end, meaning that the effects of urban proximity tend to diminish logarithmically as distance increas es. The ongoing transition from a mainly agrarian economy to an industrial one has been mostly responsible for the massive migration to urban centers, and hence the removal of forests (mostly coffee agroforests) that surround the main cities in the study area. This trend was observed throughout the study period. As was expected, the distance to main road s was a factor that promoted deforestation. Proximity to main roads caused a logarithmi c increase on the probability of deforestation throughout the study period.
104 In the agricultural areas, only th e most profitable lands, that is those that have better access to irrigation from large wa ter bodies and can be kept permanently under production, seem to have had a higher probability of deforestati on. Areas depending on rain -fed agriculture were less likely to be cleared during the study period th an those with more access to water sources. In these regions, the main crop grown is usually suga r, which is highly profita ble and requires flat lands with abundant and reliable humidity. Although topographical proximity to water sources was a major cause of deforestation for the whole period (1979-2003) of study, its effect was mostly de tected during the second subperiod (1990-2003). This may be due to the stagna tion of agriculture during the war period. After the conflict ended in 1992 a revi talization of irrigation distri cts may have accounted for the removal of forests in under-exploited lands resulting from insecurity. It is interesting to note that the Land Use Capability index was not significant in predicting deforestation. This would indicate that traditional agricultural expansion (rain-fed pulse crops) is not the main reason for deforest ation. As will be seen later though, lack of agricultural potential is a majo r cause for abandonment. These re sults support the thesis that urban expansion is probably the driving force of deforesta tion in the region. Both irrigation and urbanization have limits with regards to topography. This is why even in areas with higher rates of deforestation, places with steeper slopes tended to be less likely to loose their forest cover. This can clearly be seen in the vicinity of San Salvador, where deforestation on the slope of the San Salvador volcano has been limited only by the steep slopes. An exception to this is the relation betw een slope and landslides due to excessive precipitation such as those due to Hurricane Mitch in 1998 (Crone et al 2001) or seismic activity such as those due to earthquakes in 2001 (Jibson et al. 2004). Landslides that occur in forests are
105 often the cause of ephemeral deforestation, that is they tend to be reverted back to forests or coffee agroforests few years after they happen. Coffee agroforests generally had lower probabi lities of deforestat ion. The general trend was an inverse relation between deforestation and elevation within coffee forests. The higher the elevation, the higher the price of the coffee produced, hence lowland coffee farms are more likely to be affected by lower market prices maki ng them more vulnerable to change in favor of other land uses. However other factors such as proximity to cities actually can cancel out this trend and increase the probabil ity of deforestation, especially in lowland and midland coffee plantations. Fire frequency had no signifi cant relationship to deforesta tion. The data does not support the hypothesis that forest fires are cause of defo restation by transforming forests into savannahs that have existed since prehisto ric times in the area (Dull 2007). More likely, fires are affecting forests in terms of composition of species and tree density as is the case of fire adapted systems such as chaparral forests. These forests are f ound in many fire prone areas and are dominated by the species Curatella americana. Another case may be that fire s are affecting forests may be similar to those described by Cochrane and Shulze (1999) in the Amazon basin. This work indicates that fires are usually not responsible for destroying nativ e forests in a direct manner, but rather by inhibiting tree rege neration, desiccating soils, and making adult trees more prone to disease and pests. The result is a slow change of vegetation composition into fire resistant woodlands and savannas. Further studies should concentrate on the vegetation and faunal composition of these forests and the ecological role they play in the Lempa river basin ecosystem (infiltration, carbon fixation, wildlife conservation, ect.). It is al so possible that the data for fires
106 covered a very short period compared to th e overall duration of th e study period, thus not reflecting many of the changes they may have caused. Reforestation The results for the regression m odels regard ing reforestation all turned out to be significant (Table 4-3). While it wo uld be easy to assume that the opposite conditions that affect deforestation are the ones that promote reforestati on, this is not necessarily true. For one part, the agricultural potential of the la nd was one of the main factors responsible for forest cover regeneration. The lands with poorer soils were more likely to be abandoned than the more fertile ones. So, while rich soils are not a cause of deforestation, poor soil s certainly seem to be a cause for land abandonment and subsequent forest cover regeneration. This is particularly true in the 1990-2003 period where each level of land use cap ability (where 1 signifies the highest agricultural potential and 8 the lowest), increased the probabili ty of reforestation by 0.14 times. Two underlying factors may be behind this trend. First, the high rate of soil erosion found many of the subsistence agricultural lands of th e country. Increasing loss of fertility may have contributed to farmers looking for alternate s ources of income. Secondly, economic pressures such as falling trends in grain prices and income from alternate sources such as remittances and off-farm work (Hecht 2005; Kloos ter 2003) may have shaped farmers decisions to stop farming the less fertile soils in favor of other more pr ofitable sources of income (Chapter 3). Although it is possible to map these factors in detail for more reduced regions, they were not within the `scope of this study. The distance to previously existing forests was also a factor that promoted forest regeneration. Lands closer to prev iously existing forests were more likely to regain forest cover than those further from them. The probability of reforestation decreased 0.002 for every meter of distance from the nearest existing forest patch. This could be attributed to the presence of seed
107 sources and dispersing agents (Holl 1999, Zimmerman et al 2000). In this manner pastures or abandoned fields close to forests would be more likely to receive germplasm from nearby trees rather than develop into savannas. Access to large surface water sour ces had an inverse effect on reforestation than it did on deforestation. Lands with steeper routes to surface water were more likely to be left to regenerate forests on both sub-periods. The possibility of irrigation of lands with easier access to water sources is likely to influence th e decisions of farmers to cultivate more intensively with fewer fallow periods. Also, the data supports the idea th at lands with poorer access to irrigation tend to be abandoned more easily than those that can remain under year-round production through irrigation. It was expected that places with lower rainfall (which is sometimes a limiting factor to agriculture in the country) would be more lik ely to be abandoned while places with higher rainfall would be more likely to be turned into or kept under ag ricultural use. However, the data shows evidence of the opposite. Average rainfall maps (SNET, 2005) shows that places with higher average rainfall were significantly more likely to regain forest cover. Among the factors that behaved as expected was the distance to roads. Areas with higher accessibility were less likely to regain forest co ver than those with lower access. The probability of reforestation increased logarithmically as di stance from roads increased. Similarly protected areas were a significant factor in increasing forest cover recovery. Contrary to what was expected elevation ha d a negative effect on re forestation. Most of the forest cover regeneration that occurred happened in lowlands while the land with no forest cover in the higher grounds tended to remain unfor ested or under shifting cultivation during the study period. Similarly, forest regeneration over the whole st udy period had an unexpected
108 positive relationship with fire frequency. It would be important to study the composition of these forests, whether they are a composed of fire resistant vegetation such as Curatella americana and Byrsonyma crassifolia or if they are they are becomi ng slowly degraded. Studies in the Amazon basin (Cochrane and Shulze 1999) have s hown that many times the effects of fires do not result in direct destruction of forest, but rather in slow degradation process by which tree regeneration is eliminated and adult trees become more vulnerable to pests and disease. On the other hand paleobotanical evidence suggests that vegetation in El Salvador has long been adapted to the presence of fires si nce prehistory (Dull 2008) which w ould mean that fire is a part of the local ecology. The case may be that, un like the vegetation studied in the Amazon, Lowland vegetation in the Lempa basin is comp osed of fire resistant species adapted to conditions where fire is actually pa rt of the ecosystem. It is important to mention that the fire data used spanned only the end of the second su b-period so it is not possible to obtain any concrete conclusions from this data. More in depth studies should be done to explain this phenomenon as well as the nature of the vegetation in question. Stable Forest Cover: Permanent Presence or Absence of Forest Cover Throughout the Study Period. Many of the variables related to both forestation or defore station also had significant rela tionships with the probability of stable fo rest cover (Table 4-4). The probability of land remaining under forest cover with no change over the study period had a positive relation to elevation. The probability increased by 0.002 fo r every meter above sea level. The same, however, is true for the probability of land rema ining without forest cover where the probability of unforested land remaining so increased 0.001 pe r meter above sea level. This means that although new forests were less likely to appear at higher elevations, at th e same time, the existing forests were not likely to be removed.
109 This points to more stability in the dynamics of land use in the highlands of the watershed. Less people tend to live in the high lands which occupy a very small percentage of the land in the region. The cooler climate allows for permanen t orchards of highland crops which cannot be grown elsewhere. Paired with more constant ra infall, this secure income may account for the stability of these regions in terms of forest cover. Another factor that may have started to affect the stability of forest cover in these regions is tourism, which has started to thrive based on the attraction of highland pine-oak forests and cooler climates. Nationally protected areas, although covering less than 2% of the study area, had a positive impact in preserving forest cover. La nds under protection were significantly less likely to loose their forest cover and unforested la nds within these areas had a lower chance of remaining without forest cover throughout the period. Coffee was confirmed to be one of the main reasons that extensive forest cover has remained in the basin. Coffee forests, particul arly midland and highla nd coffee located away from population centers, were very likely to re main under constant forest cover throughout the period than other types of forest. Similarl y, land not under forest cover in 1979 had less probability of remaining unforested if located in coffee producing areas. Limited accessibility was also a strong factor in preserving forest cover. Places further from the roads and with lower secondary road dens ity were more likely to remain forested than those with better access. Also, proximity to main cities significantly decreased the probabilities of forest cover remaining stable throughout th e study period. Built surfaces are one of the most stable land covers usually it is very hard for an urban area to return to forest once it has been built.
110 Slope of the terrain had no significant relation to conserva tion. Although theory indicates that steeper slopes would tend to remain more forest ed than flatter terrains, this was not the case for the Lempa basin. Nevertheless, percentage slope definitely had a sign ificant inverse relation to the probability of land remaining unforested. He nce we can conclude that while steep areas are not necessarily helping to conserve forests, farm lands and pastures in st eeper slopes are they are more likely to be abandoned or used in a le ss intensive manner than areas with lower slope. The data also reveals that non-forested areas located near existing forests tended to have a lower probability of remaining unforested throughout the study period. This may indicate that not all the abandoned fields tend to return to fore sts or do so at the same, and nearby forests may play an important role as seed sources in accelerating the process of forest regeneration. Finally, Land Use Capability had an inversel y proportional relation with stable forest cover. This means that land with poor potentia l for agriculture had a greater probability of retaining a stable land cover over the study period than land with hi gher potential. This trend is more marked in the case of forest cover than non-forest cover. This means that there was a higher potential for poor la nds with forest to remain forest ed than the potential for poor-non forested lands to remain under non-forest use. Spatial Distribution of Forest Cover Change Probabilities When we apply the logistic m odels to map out the probabilities they represent we can examine the spatial distribution of the probabilitie s of deforestation and re forestation (Figures 41and 4-2). We find that, although th e factors driving both phenomena are similar, they reflect differently on their spatia l probability distribution. Spatial projection of probabilities gives us a representation that is hard to visualize by simply examining the models themselves. For exam ple, it can be noted that the probabilities of forest regeneration are distribut ed in a more dispersed patter n than the probabilities of
111 deforestation. A possible explanatio n for this is that the process of forest cover removal is an active process in which humans sy stematically change forest cover. While reforestation, as it is happening in the Lempa basin, is a passive phenomenon. The decision to abandon a field is not done precisely for inducing forest cover regenerati on, instead this results a side effect of other priorities in the landowners livelihoods. Although the factors affecting both phenomena are similar, th e difference in the strength of the relationship these variables have with land cover chan ge decisions results in different spatial patterns between forestat ion and deforestation probabi lities. The probabilities of deforestation show a more sy stematic pattern around roads and urban centers, while the probabilities of deforestation show a more scattered pattern, with less influence from accessibility factors. It is worth mentioning that proba bility maps do not represent the actual land covers existing in that particular spot, only the probability of a particular cha nge occurring if land cover conditions exist to allow it. For example, even if the conditi ons of a certain area crea te high probabilities of deforestation, deforestation is only possible if forest cover is pres ent in that area. Applications of Spatial Logistic Models to Scen ario Building Based on these models, and assuming conditions for the years 2003-2020 will be similar to those existing during the period of 1990-2003, it is possible to mode l the probabilities of deforestation and reforestation in the Lempa Basin. One important as pect to take into account is the new Northern Transversal Road which is st arting to be built this year as part of the Millenium Fund Initiative of the United States Government. The models for deforestation and reforestation for the study period were applied using the conditions present in 2003 as a starting point. One scenario lacked the transversal ro ad while the other included it. The images
112 subtraction of the probability ma ps that resulted show the area s where the probabilities change for each variable (Figures 4-3, 4-4). The changes predicted by the models indicate that in the case of the road being built the probabilities of deforestation would increase in that region while the probabilities of forest regeneration would decrease. It is important to point out that an increase or decrease in the probabilities of forest cover change does not nece ssarily imply it is going to happen. In order to make a prediction a threshold probability of change must be set. In this case a pr obability of more than 0.50 was chosen as the default threshol d for a change to happen, also, if the current land cover was also taken into a ccount there cannot be deforestation unless there is forest cover at the beginning of the period). The estimated cha nges in forest cover for these two scenarios are presented in Figure 4-5. Scenario 2, which is the most likely one sin ce the Northern Transversal road is already underway. This scenario predicts the road will ca use a direct decrease in forest cover of 652 ha, and will prevent 1,754 ha from reverting back to fo rests with relation to the scenario where the road is not built. However, both scenarios predict that the overa ll tendency of the following decade will be towards a net increase of forest cover of about 6 -7 %. This means there will be a slow down in the forest regeneration trend observed over the st udy period (Figure 4-6) but still an overall increase. This is in accordance to forest transitio n theory which says forest cover of countries will grow and then stabilize at a new level after the transition ha s happened (Mather 1992; Foster 1992; Rudel 2005; Foster and Rozenwe ig 2003; Perz 2007) The reasons for this in El Salvador may lie in the shortage of tillable land and the reduction in subsistence agriculture due to the integration of a larger part of th e population into a market economy.
113 These two scenarios are simply an illustration of the uses of this technique. Concretely they are both very limited in scope. They predic t possible futures based on the extrapolation of conditions present at 2003 (land values, crop values, mi gratory tendencies). Social and economic conditions that are likely to change are not bein g taken into account for this study since it is not within its scope to do so. It is possible to do this by studying the effects changes on such conditions have over the coefficients of the variables in the model. Applications for these types of models in the conservati on field include planning new protected areas in places were forest is lik ely to be under less pressure; concentrating conservation efforts where they are likely to be more needed, estimating environmental impacts of development projects. Conclusions High slopes incre ase the probability of forest regeneration and decrease the probability of deforestation. Elevation, unlike expected, had an inverse relation with the probability of forest regeneration, this phenomenon happened mostly in the lowlands. An exception to this are the coffee growing areas, with excepti on of areas close to urban centers, higher elevations promoted forest conservation and regeneration, and decreased chances of forest removal. The proximity to roads is also a factor that significantly in creases the chance of forest being removed and reduces th e chance of regene ration happening. Lands with lower agricultural potential we re most likely to be abandoned, probably in favor of other more profitable activities. Lands with the highest potential for agriculture were not found to be more prone to deforestation. One possible explanation is that lands with high agricultural potential ha ve already been cleared of fo rest cover, since there was a significant relation between high agricultur al potential and the probability of land remaining without forest cover throughout the study period. Areas where topography allows easy access to surface water sources ha d a higher likelihood of deforestation and tended remain under agricultura l use than in other areas. Coffee has had a significant impact on the preservation of tree cover in the region probably, because of its high prof itability, especially at higher elevations. As the results show, higher elevation in coffee growing regions has a direct relationship with forest cover conservation and regeneration, and an inverse relation with defo restation. An exception to this tendency are coffee plantations located near urban centers. Here land values for urbanization exceed the land values of coffee farming and forest removal is more likely.
114 Fire was found to have no significant relation with forest cover removal. In fact it seemed to have a positive relation with forest cove r conservation and even regeneration. A possible explanation of this is that a lot of this forest cover is an asso ciation of fire resistant species such as Curatella ameircana which grows in fire prone regi ons and is an indicator of low soil fertility and high aluminum toxicity. Th ese conditions are probably related to low fertility of these lands. Proximity to existing forests is a highly signi ficant factor in fore st cover regeneration. Although further studies would be needed to confirm this for the Lempa River basin, literature indicates that at smaller scales forest regeneration is increased in the proximity to forests due to the dispersal of seeds from the forest. Proximity to urban centers had a strong invers e logarithmic relation with the probability of deforestation. Large fragments of forest (mainly coffee agrofore sts) have been cleared in the surroundings of major urban centers, mos tly located in the central highlands as the urban sprawl happens. National protection of forested areas covered less than 2% of the region, however these initiatives seem to be very successful since de forestation is significantly reduced in these areas, the probabilities of reforestation and of forest cover bei ng stable throughout the period also increase significantly. Rainfall has a significant effect on the probabi lities of reforestation and deforestation, however more detailed analyses need to be conducted to describe the nature of this influence. Future topics of research that arise from th is study are: the detailed effect of climate variation on land use decisions; localized in fluence of topography in LU decision making by farmers; relation between topography, elevat ion, environmental, and land values with urban expansion; relation between soil nutri ent content on land us e change decisions. Logistic regressions can be used as a tool to evaluate altern ative scenarios of forest cover change. The creation of probability maps can serve as a useful tool in evaluating the impacts of large-scale development proj ects as well as watershed management.
115 Figure. 4-1 Probability map for reforestation according to the l ogistic model for the period 19792003 in the Lempa River Basin of El Salvador.
116 Figure 4-2 Probability map for deforestation acco rding to the logistic model for the period 19792003 in the Lempa River Basin of El Salvador.
117 Figure 4-3 Changes in the probabili ty of forest regeneration acco rding to two possible scenarios for the period 2003-2027. Only areas with no fo rest cover in 2003 are included in the map.
118 Figure 4-4 Changes in the probabili ty of deforestation according to two possible scenarios for the period 2003-2027. Only areas with forest in 2003 are included in the map.
119 7,452 20,147 12,695 6,800 21,901 15,101 5,000 10,000 15,000 20,000 25,000 deforestationreforestationNet increment in forest cover Type of changeHectares Scenario 2 Scenario 1 Figure 4-5 Estimated changes in forest cover for the period 2003-2027 accor ding to two different scenarios.
120 0 10 20 30 40 50 60 1970198019902000201020202030 YearPercentage forest cover Figure 4-6. predicted trend of forest cover in the Lempa river basin according to the proposed scenarios.
121 Table 4-1 Results for forest change models Type of change Period under study Nagelkerke r2 Chi Square Overall prediction accuracy Deforestation 1979-1990 0.21936.21* 70.3% Deforestation 1990-2003 0.171009.1* 77.9% Deforestation 1979-2003 0.17573.9* 84.4% Reforestation 1979-1990 0.091380.3* 78% Reforestation 1990-2003 0.232940.3* 71% Reforestation 1979-2003 0.193315.56* 69% Permanent forest cover1979-2003 0.385941.1* 92.2% Permanent non-forest cover 1979-2003 0.407883.2* 75.1% *Highly significant (P < 0.00001) Table 4-2 values for the variables in the deforestation models Variable 1979-1990 1990-2003 1979-2003 Protection -6.34E-01-1.10E+00-9.01E-01 Slope 9.56E-03-1.44E-02-8.32E-03 Access to surface water sources Not significant-6.78E-06-3.81E-06 Average rainfall Not significant-8.27E-04-4.15E-04 Elevation in coffee growing areas -2.40E-03Not significant-2.53E-03 LN of distance to major urban centers -1.60E-01-1.76E-01-2.74E-01 LN of distance to main roads -7.26E-02-4.52E-02-9.22E-02 Elevation Not significant -4.50E-04Not significant Fire frequency for 1996-2003 Not appli cableNot significantNot significant Highland coffee Not significant-2.46E+00Not significant Midland coffee Not significant-2.911757Not significant Lowland coffee Not significa nt-1.41E+00Not significant Constant 2.19E+00 3.41E+003.09E+00
122 Table 4-3. values for the variables in the reforestation models Variable 1979-1990 1990-2003 1979-2003 Protection 0.390 0.450 0.630 Elevation 0.001 -0.001 -0.001 Agricultural Potential 0.044 0.145 0.136 Slope Not significant 0.023 0.023 Distance to existing forests -0.003 -0.002 -0.002 Access to surface water s ources 6.1 E-06 7.4745E-06 8.86E-06 Rainfall Not significant 0.001 0.0005 Elevation in coffee growing areas 0.001 0.001 0.0006 LN of the distance to main urban centers 0.286 0.347 0.34 LN of distance to main roads 0.041 0.067 0.09526577 Constant -4.153 -6.121 -5.70 Table 4-4. values for the variables in the logistic models regarding stable forest cover. Variable Permanent forest cover Permanent absence of forest cover Protection 0.617 -0.551 Elevation 0.002 .001 Land Use Capability -0.061 -0.091 LN distance to main urban centers 0.260 -0.204 LN distance to main roads 0.066 -2.65 E-05 Distance to existing forests (1979) Not Applicable 0.0008 Slope Not significant -0.02 Access to surface water sources 7.1193E-06 -7.96 E-06 Elevation in coffee growing areas 0.004 -0.0004 Constant -6.679 1.217
123 CHAPTER 5 INFLUENCE OF PRECIPITATION VARIATI ON ON T HE SPATIAL DISTRIBUTION OF FOREST COVER REGENERATION IN THE LEMPA RIVER BASIN OF EL SALVADOR. Introduction One of the m ajor risks posed by various clim ate change scenarios is the increase in variability of precipitation regimes (McCarthy et al 2001, IPCC 2001). Variability is particularly important for farmers in developing countries w ho depend on rain-fed crops and have little or no access to meteorological forecasts. Uncertainty in th e timing of rainfalls can lead to crop losses either by early or late planting, or by planting crops which are vul nerable to pests and diseases brought about by extremes in temperature or humidity (Wilken 1987). The impacts of climate change often affect the production base of soci ety; the repercussions of which can be so important as to determine major changes in soci eties and land use. For example, changes in precipitation levels appear to have greatly cont ributed to the collapse of the Maya Civilization (Webster et al. 2007; Hodell et al 2001; Curtis 1998) and the fa ll of several dynasties in China (Zhang et al. 2008). According to Zhang, in thos e times the major drivers of climate changes were natural cycles, however, evidence points to man-driven climate change to be the major driver in the present. Food insecurity brought about by changes in c limate often leads to changes in livelihood strategies. For example, Ziervogel et al. (2006) me ntion; searching for o ff-farm labor, migration, renting or selling lands, as strategies undertaken by Mexican farmers faced with such conditions. Changes in livelihoods at a household scale directly change the patterns of land use that can be observed a larger landscape scal e (Geoghegan et al. 2001). The allo cation of labor into off-farm activities often means field abandonment which, in many cases manifest as an increment in forest cover in the region (Klooster 2003).
124 The resurgence of forests in developing countri es sometimes is part of the process called forest transition (Mather 1992; Drake 1993; Grainger 1995; Perz et al. 2007). Forest transition theory holds that in developing countries conditions such as industrialization, globalization of markets, increase in per capit a income, modernization, migra tion, and other factors tend to reduce the demand for farmlands, and encourage th e regeneration of forests (Mather et al. 1999; Foster and Rosenzweig 2003; Aide and Grau 2004; ). Although an overarc hing theory of forest transitions has not been found (Perz 2007) and fact ors that trigger them can vary from region to region, literature indicates that, in general, the process occurs when alternative incomes to subsistence farming become a widespread option. (Klooster 2003 ; Grau et al. 2003; Foster and Rosenzweig 2003). Recent studies indicate that El Salvador is one such place (Hecht 2005, 2007; Chapter 2). Here conditions of industrialization, immigration, decline in grain pr ices, globalization, and armed conflict (Hecht 2005) indi cate a significant increment in overall forest cover. Although climate has been mentioned as a fact or in land cover change (Ziervogel 2006), there few studies that have addressed the quant itative relationship climate and forest cover regeneration. This study seeks to determine what effect climate has on the decision to abandon land and let it regenerate into forests. The spatial scale of the analysis is th e Lempa River basin which covers 9868.45 km2, approximately 47.6% of the El Salvadors total area (Figure 5-1). Average annual precipitation within the watershed ranges from 1141 to 2886 mm (SNET 1998). Elevations vary from 0 to 2678 meters above sea level (NASA 2000). The Lemp a basin provides El Salvador with 33% of its electrical power (Lozano & Cottin 2002) and supplies about 66% of the countrys population with drinking water.
125 The study examines forest cover changes between 1979 and 2003. Monthly rainfall records kept by El Salvadors Meteorological service (now SNET) from 1971-1999 were used to reconstruct The main objective of this study is to determ ine what is the relationship, if any, between interannual and spatial vari ability in precipitation and the process of fore st cover regeneration.. Mapping Forest Cover Change in the Lempa River Basin Forest cover m aps for the Lempa River ba sin for the years 1979 and 2003 provided the bases of this analysis. Previous work divided the study area into forest and non-forest pixels (60 x 60 m). Since the spatial scale of the precipitation data was much larger, the land cover data was converted to a 1000 x 1000 m resolution using the de nsity tool in Arc GIS, and the number of forest pixels (60m x 60m) in a 1 km2 was counted during 1979 and 2003. The result was a forest pixel density map for each year, in which each 1 km2 cell was attributed the percentage value of the area classified as forest within it. The two resulting maps were subtracted ( 2003-1979) to determine the magnitude and direction of change forest cover in that peri od. A positive value indicates increased forest cover while a negative one indicates a decrease. For the purposes of an alysis, this information was converted into binary form where 1 represented an increase greater than 23% in forest cover and 0 represented increases lower than 23%, no change or negative changes in forest cover. The figure of 23% was chosen since th is is the average forest area increase detected throughout the period (Chapter 2). Reconstructing Monthly Precipitation Di strib ution Through Linear Regression. General Precipitation Regime in El Salvador General Precipitation Re gime of El Salvador
126 The pattern of precipitation of El Salvador is typical of the Pacific coast of Central America (Figure 5-2), consisting of a rainy and a dry season. The fo rmer, starts in May and ends around November, results mainly from humidity drawn from the Pacific Ocean. Although the ITCZ does not migrate as far north as El Salvad or (13 14 degrees North), its influence can be detected in the humidity brought by the westerly winds (Hastenrath 1988), This phenomenon is also thought to be responsible for temporales or humid fronts that cause a uniform, constant precipitation over the Pacific coasts of Ce ntral America and last for several days (Hastenrath,1991). Tropical storms in the Atlantic generally do not affect El Salvador directly, instead their effect is felt as humid air from th e Pacific is drawn across the isthmus towards these cyclones resulting in periods of constant rain fall throughout the country (Portig 1965; Waylen and Harrison 2002). Precipitation usually a drops during the months of July and August in what is called the canicula or midsummer drought (MSD) (Magaa et al 1999). This drought is brought about by the intensification of the trade winds over the Caribbean, particularly the Caribbean low-level jet, which tend to temporarily subdue the effects of the ITCZ. (Webster 1992) The MSDs effect is closely linke d to a jet that passe s through one of the major gaps in the Central American Cordillera and over the Gulf of Fonseca (Chelton et al. 1999). This pathway permits the trade winds to jet through from the eastern part of the c ountry. The resultant offshore winds over the Gulf of Fonseca (and al so the Gulfs of Tehua ntepec, Papagayo and Panama) is to produce upwellings of cold water which create stability in the atmosphere and reduces convectional activity (Webster 1992). Th is wind gap also allows the passage of the northern cold fronts during the boreal winter to pass thorough causing a reduction in ocean
127 temperatures and evaporation (Chelton et al. 1999), which, coupled with the receding of the ITCZ, causes a pronounced dry season from December to March. The regional climate is also affected by th e El Nio Southern Oscillation phenomenon (ENSO). Generally El Nio (warm phase ENSO) mani fests itself as an intensification of the Caribbean trade winds in the boreal summer pr eceding the arrival of the cooler waters in November and December off the coast of Peru. On the average, El Nio years bring higher levels of precipitation during the beginning of the rain y season. However it also increases the intensity of the MSD and an overall reduction in annual precip itation in the country dur ing the later part of the rainy season (Figure 5-3). La Nia (cold ph ase ENSO) on the other hand, weakens the trade winds and increases the freque ncy of tropical cyclones in th e Caribbean (Hastenrath 1988) thereby increasing overall yearly precipitation. La Nia also affects the te mporal distribution of precipitation, lowering precipitation levels dur ing the beginning of the rainy season and increasing it from August to November. These effects can be seen in the average monthly precipitation for the region for the years 1971-19 99 according to SNET (2005) historical records (Figure 5-3). A major factor influencing th e spatial distributio n of precipitation is topography. The central mountain range is the fi rst obstacle that is encountered by humid winds coming in from the Pacific. The effect of th ese mountains a typical orograph ic pattern in which humidity precipitates as the air cools dow n as it rises over the mountains (use the references from the class). There is a corresponding rain-shadow effe ct that reduces the precipitation on the leeward side of the range and the inner valley. As the air rises above the northern mountain range precipitation increases again as the remaining humidity precipitates at higher elevations. (Hastenrath 1988, Daly et al. 1993, Marqines et al. 2003).
128 Linear regressions as a tool for reconstruct ing historical precipitation distributions Although historical point data give us an idea of the average monthly precipitation in the region, they do not indicate how th is precipitation is distributed in space. In order to reconstruct historical distribution of rainfall over the basin, li near regression models were used to extrapolate point data into continuous surface data. The Sistema Nacional de Estudios Territori ales (SNET 2005) database consists of 17 rainfall stations contained w ithin the basin and 4 outside th e basin (Figure 5-1). Each one possesses monthly precipitation data from 1971 to 1999 with varying proportions of missing data. In total sample of 4874 m onths (after excluding missing data) is used for the study. One case was defined as the total precipitation recorded at one station within the basin for a particular month. Of the 4874 cases, 338 were randomly selected and intentionally left out of the modeling process for future testing of the accuracy of these models. The remaining 4536 cases were used for recons truction of the precip itation for each month of each year (228 models in total) using multivar iate linear regression (Rogerson 2001). In order to estimate historical rainfall throughout the st udy period as an input for the land cover change models in the second part. Once calibrated, the models were validated using the 338 unused cases. And the modeled results compared to observed precipitation. The r2 values of the predicted data were compared to those of the mo dels to confirm the predictions from the models were able to predict the variation in external data as well as the predicted it in the data used to build them. Temporal and geographic variables were used as independent variables in the modelbuilding process:. Geographic variables, such as elevation and distance to the coastline came from digital maps. The temporal variables such as ENSO phase and precipitation data from outside the basin served to capture interannual variability in the data.
129 Geographic variables Elevation: elevation is the main t opographic variables associated with local precipitation (Hastenrath 1988, Waylen et al. 1996, Marquine z et al. 2003, Lloyd 2004). Data from the NASA shuttle topography mission was used for this purpose. X-Y coordinates Five variables related to location were tested as possible predictors of precipitation. X and Y coordinates (m eters UTM Zone 16, WGS 84), X*Y, X2*Y, X*Y2 and X2 and Y2. The theory behind using these variables is to detect large scale climatic influences such as variation in the ITCZ, influe nces of the trade winds coming in through the eastern part of the basin, rainfall from hurricanes a nd tropical storms coming from th e Pacific, or dry winds coming from the north during winter. Distance to the coast. Since a large proportion of the rain fall originates over the Pacific Ocean and moves inland from there, the distance from that source was also tested as independent variables for each model. Geographic zones. The watershed was divided into ge ographical zones according to the influence geomorphology in the region has on pr ecipitation (Figure 5-4) These areas were basically the leeward slopes of the central mountain range, the inner valley, the windward slopes of the northern mountain range, and the coastal lowlands. These areas were used as dummy variables during the regressions. Temporal variables. The phases of ENSO are responsib le for a large part of the in terannual variation in monthly precip itation. In order to include these in the re gression process, a temporal mask was used when conducting the regressions for each station, marking them as El Nio or La Nia years according
130 to the Japan Meteorological Agency (JMA 2003) According to the classification, this study period included 7 La Nia years, 7 El Nio years, and 15 normal years. Additionally, to quantify the in trannual variation of preci pitation not captured by the dummy variable (variations in intensity of ENSO cycles, lag effects from past years, stochastic events, etc.), the data from the four rain gauge s outside the basin (Figure 5-5) were used as a proxy of the actual intensity of rainfall for each month at different years. The idea of using this data is to capture the temporal variation that originates from in terannual changes in the regional climate in a continuous rather than discrete mann er. Since the data from all four stations had a high degree of correlation, a pr incipal component analysis was run (Rogerson 2001) to extract the main trends in the data. Three principal co mponents for each month (over the 29 year period) were tested as independent vari ables to predict monthly precipitation within the watershed. Each one of these components represents a completely independent trend in variation of interannual precipitation data. Combination variables During the modeling process all the above vari ables were also combined in ways that would be able to capture more of the varia tion in the precipitati on data. For example, elevation*inner valleys, El Nio*eleva tion, La Nia*leeward slopes, etc. The general methodology for creating the monthly models was to first conduct a stepwise linear regression using SPSS Inc. (1999). Once co mpleted, the most significant variables from the stepwise regressions were combined to cr eate additional models besides the ones obtained through that process. Once a suitable model wa s found for each month, it was applied for that month in all the years within the study period.
131 All variables used were tested for normality and multicolinearity. No variables presenting a Variance Inflation Factor greater than 5 or a tolerance below 1-r2 (Rogerson 2001) were included to prevent effects of multicol inearity within the models. The result of this process was a set of 12 spatially explicit models used to produce 348 monthly precipitation maps, 12 monthly maps of th e distribution of coeffi cient of variation in total precipitation, 12 monthly average precipit ation maps, 24 minimum and maximum monthly precipitation maps, 24 average monthly precipit ation maps during both phases of ENSO, and 29 yearly total rainfall maps. All these were used as inputs for predicting land cover change. 3.5 Results of linear regressions for the reconstr uction of spatial distribution of historical precipitation. The statistics of the resultant models that indicate a definite relationship between precipitation and the geophysical and tem poral variables employed (Table 5-1). The r2values of the predictions of the test data set indicate that most of the models were efficient in predicting precipitation of these case s (Table 5-1). An exception to this were the models for the months of December-February. These models failed to accurately predict the variation in precipitation of the te st data and presented very low r2values The reason for this probably lies in that the precipitation in these mo nths is minimal, and the variation that does occur is probably not due to any of the variable s used to construct thes e models. One model that was considerably low was that of August. This model captured 0.30 of th e variation found in the original dataset, but explained onl y 0.20 of the variation in the test data. One explanation for this may be that August is a month when stochastic phenomena such as hurricanes in the Caribbean sometimes affect El Salvador increasing precipitation. This is likely to cause a large proportion of unexplained variation in the data that had no relation to the variables used in this study.
132 From examining the models there are four dist inct periods in the sturdy area according to the time of the year. The first general pattern is that of the dry season (December April) (Figure 5-5, Table 5-1). The models for this time period i ndicate that there is ve ry little variation in observed monthly precipitation data. Taking into account the limitations of the models for December, January, and February (the height of the dry season), we can sa y that the distribution of the scant precipitation that falls during the dry season tends to be driven by location and elevation. The models show that the dist ribution of rainfall during the dry season is greatly affected by ENSO (Figure 5-5). Normally pr ecipitation at this time is more or less uniform, ranging from 50-100 mm over most of the basin with sligh tly higher (100-150 mm) on the northwestern highlands. La Nia on average is the driest for this period since total ra infall over most of the watershed tends to be below 50 mm with he exce ption of the north-western highlands. El Nio years tend to have slightly higher precipitations (50-100 mm) with drier areas concentrating in the central Lempa valley. This distribution of rainfall du ring this period can mark the differences in the timing of field burning and preparation for pulse crops whic h may lead to variations in productivity from year to year. The next period with a distinct precipitation pattern is the early ra iny season (MayJune) (Figure 5-6). During these months, the pre-canicul a period, the models indicate that the main driving factor is the humidity brought in from th e Pacific Ocean. This importance can be seen in the relation that variables such as elevation and rain shadow effects have on precipitation as the humid air collides with the central and northern mountains (Table 5-1). Elevation significantly increases precipitation in this period (0.06 mm/m in May, 0.06 mm/m in June during normal
133 years, and 0.09 mm/ m in June dur ing El Nio years) as humid air collides against the mountains when it enters from the ocean. Both the leew ard slopes and inner valley also tend to get significantly less precipitation (-56.9 8 and -37.67 mm respectively) than the other regions as a large part of the moisture is precipitated on the windward slopes. The models show the distribu tion of precipitation changes a ccording to the ENSO cycles. La Nia manifests as a slow start in the ra iny season, with lower than average precipitation during these months over the greater part of th e basin (Figures 5-3, 56). El Nio on the other hand brings slightly higher th an average rainfall over the regi on. The highest precipitation for these months happens in the northern highla nds, while the lower pr ecipitation tends to concentrate around the Central Lempa valley. The third distinct precipitation period is the m onth of July when the intensification of the trade winds creates the MSD, this results in a d ecrease in precipitation (Figures 5-4, 5-6). This intensification manifests in the models mainly as an inverse relation be tween precipitation and the distance to the coast (-0.008 mm/ m). As the trade winds blow from the Caribbean through the Gulf of Fonseca wind gap, the humidity from the ITCZ seems to concentrate itself mostly near the coasts. The MDS marks a reversal in the effects of EN SO cycles so far. The higher than average rainfall brought by El Nio during the dry and early rainy season drops drastically (Figure 5-4), creating areas of very low preci pitation, particularly in the central and lower Lempa valley (Figure 5-7) and areas with higher than normal precipitation in the central highlands. la Nia on the other hand, mitigates the eff ects of the MDS and maintains an overall higher and widespread distribution of precipitati on throughout the basin.
134 The last precipitation trend obs erved was that of the second ha lf of the rainy season, after the MDS. For the months of August, September, October and November, the models and historical records indicate that La Nia tends to intensify precipitation (Figures 5-4, 5-8; Table 51). This overall intensification is concentrated mostly towards the eastern part of the country. The effects of El Nio manifest as a slight in crease in precipitation in the eastern Lempa valley but a marked decrease in rainfall towards the western valley. Distance to the coastline as well as elevation play key roles on determining the distribution of rainfall in the basin during th is period. Unlike in the early rainy season and the MDS, in this later period geographic location plays a very sign ificant role in precip itation. The coefficients show precipitation tends to be higher in the north and north eastern regions of the basin. Predicting Land Cover Change Based on Rainfall Distribution Logistic Regressions for Testin g th e Probability of Reforestati on during the Study Period. Based on the hypothesis that climate variability brings about food insecurity and consequently changes in land use patterns, a stud y of the reconstructed precipitation distributions and the increase of forest c over in certain regions of th e Lempa basin was undertaken. Precipitation maps resulting from the models, Land Use Capability index (Klingebiel and Montgomery 1961) were used to examine the rela tion between the probability of increases in forest cover to climatic and edaphic factors. The Land Use Capability Index classifies land according to its agricultural potential. In this st udies a selection of areas within the basin in categories V-VII (low agricultural potential) in Klingebiel and Montgomerys Land Use Capability map was used to create an analysis layer to be combined with precipitation data. The reason for using this variable is because soil fe rtility interacts with climate in agricultural decisions. Very often places with similar precipitation will have very distinct agricultural
135 potential according to the soils, hence it was thou ght necessary to integrate this data along with the precipitation variables. A random sample of 10% (n = 2566) of the fo rest cover change and dependent variables maps was taken to relate the land cover data to the corresponding climatic data. The independent variables were related to the bi nary variable of in crease greater than 23 % forest cover through logistic regressions (Menard 2002) (Equation 5-1). Due to th e large amount of possible explanatory variables and the complex relations be tween them, stepwise regression was used as a preliminary sorting mechanism. Once non-signif icant variables were di scarded, the remaining ones were tested in different combination of models different to th e ones from the stepwise regression to find an optimal model that expl ained the effects of precipitation on land cover change. Equation 5-1. Logistic equation. P = e(ax1+bx2+cx3++C) 1 + e(ax1+bx2+cx3++C) Where: P = Probability of a 1 km2 area gaining more than 10% forest cover e = logarithmic constant C = Regression constant a,b,c.. = Regression coefficients xn = independent variables Results for Logistic Models of Fore st cover Ch ange Based on Climate The logistic regressions indicat e that the probability of forest cover increasing more than 23% can be accurately predic ted using climatic data (Table 5-2). With a pseudo r2 of 0.30 and an overall prediction accuracy of 63.68%, the logistic model indicates land cover change decisions for the study period were significantly influenced by precipitation and its interaction with soil fertility.
136 The coefficients in the table indicate the ma gnitude and direction each variable has on the odds of dependent variable occurring. The sign of the coefficient indicates whether the relation is direct or inversely proportional to the odds ratio of forest increase happening. The exponent of the coefficient indicates the pr oportional change in the pr obability of reforestation happening, values above 1 indicate a positive relation while values below 1 indicate a negative relation with the de pendent variable. The results of this study indicate that at a 1 km2 scale, increments of more than 23% for the study period in forest cover be predicted w ith a 78.2% accuracy based solely on precipitation data. The effects of edafo-clim atic associations related to la nd cover change combine with socioeconomic and geophysical conditions to shap e human decisions to abandon or clear forests for cultivation. Although the forest regeneration model tended to underestimate the probability of reforestation with regards the actual changes re corded during the period, it accurately predicted the distribution of regeneration in the eastern, lower and cen tral Lempa valley. However, prediction accuracy was weaker in the north and northwestern regions. It may be that climate is not be playing an important part in forest cover changes in these areas. The northern regions tend to have abundant, constant rainfa ll, yet the effects of the 1980s armed conflict, accessibility and soil fertility in these areas tends to limit the agricultural potential for those lands (Chapters 3 and 4) and may be playing a more impor tant role than climate in land us e. Another possibility is that the precipitation data for the nor thern highlands are limited to one representative station and the precipitation variations there may not have been captured accurately by the linear models. No significant relation was found between rec onstructed precipitation for the dry season months (December-April). Although the models for most of this season did not have a good
137 predictive power, it is likely that the small am ounts of precipitation dur ing this season do not affect forest cover regenerati on very much. The effects of c limate on land abandonment in the region are mostly related to its role subsis tance crop agriculture (Klooster 2003; Hecht 2005; Chapter 3). During these months, subsistence agriculture is at a stand still due to lack of sufficient rain such small variations in rainfall at this moment are not likely to affect land use decisions. Mean rainfall during the months of May and Ju ne had a significant negative effect on the probabilities of forest regenera tion. Mean rainfall sign ificantly decreased the probability of reforestation by a factor of 0.99 during May and 0.96 during June for every millimeter of rainfall during these months. This means the areas that get most abundant rainfall in the early rainy season are more likely to remain under agricultural use as opposed to those that remain dryer. Abundant rainfall in this month may allow farm ers to plant early making these lands more productive than those were rains starts later. For the month of July during El Nio years, precipitation also had a negative effect the probability of reforestation. July is strongly associ ated with the MSD (Figure 4-4) in El Salvador, data indicates that El Nio tends accentuate this regional drought (Figure 5-4) which often leads to crop failure. The probability of reforesta tion decreased by a factor of 0.96 for every mm of rain, meaning that regions with lower precip itations in July duri ng El Nio years had significantly higher probabil ities of reverting to forests than areas where the drought was not as strong. Interestingly, places with more variati on in precipitation (highe r coefficient of inter annual variation or CV) were less likely to regain forest cover than those with more stable rainfall. This could mean that it is not places wh ere the MSD is more variable in intensity that
138 are prone to abandonment, but ra ther those where it causes consis tently low preci pitation during this month, particularly during El Nio years. The opposite is true for the month of August. Areas where the coeffici ent of variation was high during for this month tended to be 3.06 times more likely (for every percentage point of CV) to regain forest cover than those with lower CVs. August is one of the months when the effects of El Nio and La Nia are most contrasting (Figure 5-4) During this month La Nina causes abnormally high precipitation while El Nio has the opposite effect. This volatile climate may cause uncertainty as to crop selection, a nd bring about drought, pe sts and disease makes agriculture harder in these areas. While the MSD is a predictable change in inteannual rainfall, the strong variations during Augus t are directly linked to ENSO phases which makes them less predictable for traditional farming. This variability between extremely high rainfall from year to year may be the cause for abandonment. These eff ects are mostly felt in the central and eastern Lempa valley (Figure 5-9) In El Salvadors traditional agricultural cycl e, September and October are the months when the first maize crop has been harvested and a sec ond crop of maize and/or beans is planted. Areas that can sustain these multiple crops are more productive than areas where only one harvest is possible or where the second harvest has to be substituted for sorghum or less profitable crops. Beans however are susceptible to fungal diseases Acua (1976), and require lower precipitation than maize. During these months, the effects of La Nia increase precipita tion in certain regions (Figures 5-4, 5-9). This may explain why areas with higher pr ecipitation in October during La Nia years were more likely (with an increase factor of 1.31 times per mm) to be abandoned than those with lower precipitation.
139 The effects of La Nia phenomenon are usually felt in El Salvador during the months of October and November (Figure 5-5) Generally this is manifested by long temporales or frontal precipitation. Precipitation during the month of November had a strong direct relationship with the probability of forest cover increasing in an ar ea. Places where mean precipitation during normal or El Nio years was unusually high had mo re chance for abandonment than places where precipitation was more moderate This direct relation between high precipitation and field abandonment is probably indicative of areas wh ere it is too humid fo r bean cultivation. Finally, the regression models indicate that places with lo w agricultural potential were more likely to be abandoned if the mean year ly precipitation was highe r. The probabilities of forest regeneration happening increases 1.001 ti mes for every millimeter of rainfall they received. The combination of low fertility and high rainfall may decrease agricultural productivity making these lands less attractive for cultivation Discussion The effects associated w ith of La Nia seem to be the most influential climatic factor over forest cover in the region. The first effect of this ENSO phase is felt in the beginning of the rainy season. La Nia year tend to be dryer than El Nio and normal years and the model indicates this is a condition significantly related to increases in forest cover. Dry starts to the rainy season my cause uncertainty in field burni ng and planting schedules for traditional farmers. It may also shorten the growing period for the whole agricultural cycle. The effects of La Nia are felt even mo re during the later rainy season (August November). The model indicates that high variation, particularly abnormally high precipitation, is a main driver of land abandonment in the region and this is usually a ssociated with La Nia. The excess of rainfall during this time may cau se flooding, pests and disease that amount to
140 serious crop losses. The probability of forest regeneration is even higher in places where El Nio creates the opposite effect to La Nia creating areas of high interannual variation. This variability creates uncertainty in planting se asons and crop selection in traditional farming systems. Agriculturalists are prob ably more likely to choose off farm work or migration in these places where risk of agricultural loss is higher. The central and eas tern Lempa valley are some of the areas that are affected by these increases in rainfa ll (Figure 5-1). El Nio also had its effects on forest cover a lthough not strong as La Nia. One such effect is the intensification of the MSD which the mode l indicates have higher probabilities of forest cover recovery. One such area is the central Lempa valley (Figure 5-1). Although climate is not the sole reason for la nd abandonment evidence points to it as a key factor influencing land use decisi ons in the region. In the light of climate change it would be important to look further into the mechanisms of how climatic variation can be taken into account in as a factor in regi onal planning and decision making. Traditional rain-fed agriculture in El Salva dor is highly dependent on constant, predictable rainfall. This system is well adapted to intra-annual variations in precipitation as well as consistent interannual changes such as the MSD. The choice of crops, planting techniques and timing is adapted to the typical pattern of a marked dry season and a very wet season interrupted by a midsummer drought. However, it is inter-annual variability that seems harder to cope with. For one part there is more uncer tainty regarding the dates for field preparation (burning) and planting dates. Burning or planti ng too soon or too late may result in lower yields due to plant mortality or pests (Wilken 1987). Loss of agricultura l profitability can happ en in areas where the effects of the MSD are felt stronger, and where high variation during the late rainy season can cause crop loss due to flooding, pests and disease. The long term effect of increased risks in
141 agriculture production may be an important factor for farmers to consider other options such as off-farm labor or migration which result in land abandonment. The role of climate in forest transitions is a complex one, yet its effects appear to be significant and should be taken into account by decision makers in the country. The results support the concept of decreasing food insecurity with climate variability/change. The abandonment of farmlands, particularly those with le ss fertile soils and vulner ability to variations in precipitation has many repercussions on Salvad orian society. The effective loss of subsistence farming as a survival strategy is bound to in crease migration towards the US and to urban centers. This means uncontrolled growth in urban population and problems of crime, overpopulation, unemployment, public health issu es, demand for drinking water and electricity, etc. The expansion of cities also causes deforest ation in the fringes of these damaging valuable aquifers that sustain urban populations. On the other hand, the increase in forest cover may have positive effects on the countrys ecology. Increased infiltration, atte nuation of flooding and reducti on of sedimentation in dams are just some of the ecosystem services thes e new forests are providing. Although little is known about the floristic composition of new forests, they do provide habitat for El Salvadors wildlife. Other opportunities they may offer are their incorporation to carbon sequestration markets, a system already operating successfully in neighbo ring countries such as Costa Rica and Mexico. The future professional management of these forests as sources of valuable tropical woods and even ecotourism are other options that the country s decision makers should consider in order to adapt and benefit from the forest transition process. Adapting and coping with climate change it is a challenge for developing countries like El Salvador. Although the effects of precipitation variability can be a cause of relevant social
142 changes such as migration, urbanization, and food in security, there is little small countries can do to bring about immediate change. Instead it is important that local gove rnments find ways of adapting to and mitigating the adverse effects br ought about climate change. The rise in forest cover presents new opportunities for conserva tion, production of ecosystems services and forestry, and it will be up to decision makes to take adequate advantage of these changes. Conclusion In El Salvad ors Lempa River basin the abandonment of traditional agriculture in favor of other types of livelihoods has lead to an increase in secondary forest cover. Unfavorable climatic conditions can accentuate the effect of social and economic pressures that lead to field abandonment. In this study there is evidence that low precipitation le vels at the start of the rainy season, the intensification of the MSD by El Nio, higher than average precipitation caused by La Nia during the late rainy season as well as high interannual variat ions caused by ENSO, had a significant effect on the probability of forest cover increase in the study region. According to the results of this study, climate is strongly related to forest regeneration in the central and eastern Lempa valley, less so in the northern and western regions.
143 Figure 5-1. Location of study area: Lempa River basin, El Salvador.
144 0 50 100 150 200 250 300 350 400 450 janfebmaraprmayjunjulaugsepoctnovdec MonthAverage yearly precipitation (mm) Figure 5-2. Average monthly precipitation and standard deviati on in El Salvador based on rainfall data from 32 rain gauges for the years 1979 2003.
145 0 50 100 150 200 250 300 350 400 janfebmaraprmayjunjulaugsepoctnovdec MonthAverage precipitation (mm) La Nia El Nio Normal years Figure 5-3. Differences in rainfa ll distribution for El Nio, La Ni a and normal years in the Lempa River basin over the period 1971-2005.
146 Figure 5-4. Location of rain ga uges and geographically differe ntiated rainfall areas in El Salvador.
147 Figure 5-5. Distribution of mean total prec ipitation in the Lempa basin during the dr y season (December April) for the years 19711999 according to precipitation models.
148 Figure 5-6. Distribution of mean total precipitation in the Lempa basin during th e beginning of the rainy season (May June) f or the years 1971-1999 according to precipitation models.
149 Figure 5-7. Distribution of mean total prec ipitation in the Lempa basin during the mid-summer drought (July) for the years 1971 -1999 according to precipitation models.
150 Figure 5-8. Distribution of mean total prec ipitation in the Lempa basin during the s econd part of the ra iny season (August-Nove mber) for the years 1971-1999 according to precipitation models.
151 Figure 5-9. Actual Vs. Predicted increments in forest cover (>23%) in the Lempa basin according to the climate based forest cov er change model.
152 Table 5-1.Linear models for mean monthly pr ecipitation for the Lempa River basin for the period 1971-1999. Month adjusted r2 standard error Validation r2 Varibles Coeficient January 0.49 9.00 0.03 Constant 0.670 external precipitation data PC1 1.783 Elevation 0.002 February 0.38 11.90 0.03 Constant 0.450 external precipitation data PC1 0.700 Elevation 0.002 March 0.18 18.77 0.24 Constant 18.184 location in x axis -4.63E-05 distance to the coast 1.05E-04 external precipitation data PC2 (0.257) external precipitation data PC3 0.409 XYY location during El Nino Years -1.09E-05 April 0.38 36.20 0.49 Constant 0.582 external precipitation data PC1 0.436 XXY location during La Nina years -1.37E-04 XYY location during El Nino years -1.58E-05 Elevation 0.017 Distance to coast 3.29E-04 May 0.30 60.98 0.30 Constant 47.603 Presence of El Nino 26.574 Elevation 0.064 Inner valleys (37.666) Lee ward mountains (56.985) external precipitation data PC2 0.274 external precipitation data PC3 (0.192) XXY location during normal years 2.73E-04
153 Table 5-1 continued. Month adjusted r2 standard error Validation r2 Varibles Coeficient June 0.40 75.97 0.41 Constant 58.735 external precipitation data PC2 0.441 distance to the coast 0.001 Elevation during El Nino yeas 0.093 Elevation during normal year 0.061 July 0.55 61.39 0.49 Constant (10,649.382) external precipitation data PC1 0.602 X location 0.002 Distance to the coast (0.008) Y location 0.007 external precipitation data PC2 0.167 August 0.30 71.26 0.20 Constant 148.516 external precipitation data PC1 0.187 external precipitation data PC2 0.211 external precipitation data PC3 0.140 Elevation during La Nia years 0.079 Elevation 0.029 September 0.36 79.20 0.35 Constant 53.244 XXY location during El Nio years 0.001 external precipitation data PC1 0.384 Elevation during normal years 0.078 northern highlands 76.594 XXY location 3.63E-04 October 0.53 57.48 0.54 Constant 11.109 El Nino (43.134) distance to the coast 3.24E-05 XXY location 0.001 external precipitation data PC1 0.439 external precipitation data PC2 0.217 Inner valleys (24.613) XXY location during normal years -3.20E-04
154 Table 5-1 continued. Month Calibration r2 standard error pf prediction Validation r2 Varibles Coeficient November 0.72 24.49 0.58 Constant 0.525 Elevation 0.015 Elevation during La Nina years 0.010 external precipitation data PC1 0.772 external precipitation data PC2 (0.123) external precipitation data PC3 0.108 December 0.25 11.61 0.001 Constant 0.525 Elevation 0.015 Elevation during La Nina Years 0.010 external precipitation data PC1 0.772 external precipitation data PC2 (0.123) external precipitation data PC3 0.108 Table 5-2. Logistic models relating the probability of a 23% increase or greater of forest cover happening in the Lempa River basin base d on precipitation, slope and land use capability data. BExp(B) Sig. Mean rainfall in May (0.01) 0.99 0.042 Mean Rainfall In June (0.04) 0.96 0.000 Mean rainfall in July during El Nino years (0.04) 0.96 0.000 Coefficient of variation during July (0.29) 0.75 0.000 Coefficient of variation during August 1.12 3.06 0.000 Mean rainfall in October during La Nina years 0.27 1.31 0.000 Mean rainfall in November during normal years (0.28) 0.76 0.000 Mean precipitation in areas of low agricultural potential 0.001 1.00 0.000 Constant 2.47 11.81 0.249
155 CHAPTER 6 GLOBALIZATION AND FOREST TRANSITI ONS IN SM ALL DEVELOPING COUNTRIES: LESSONS FROM THE CASE OF EL SALVADORS LEMPA RIVER BASIN Introduction Forest transition is a term used to describe the process by which a country or region undergoes changes in its socioeconom ic structure th at lead to an increase in forest cover from previously deforested lands. Scientists st udying this phenomenon have found several paths through which forest cover recovers. The first is the economic development path (Rudel 2005). Seminal forest transition research found that countries undergoing modern ization and industrialization tended to have increases in forest cover (Foste r 1992; Drake 1993; Garinger 1995). Th e reason for this is that as the countrys industry develops, people tend to switch from low paying farm work to higher paying jobs in industry. As this happens, marg inal lands are gradually abandoned and tend to revert back to forest land. Such is the case of United States (Foster, 1992) and France (Andre, 1998; Mather et al ., 1999) among others. The second path to forest transition mentione d by Rudell (2005) is the Forest scarcity path. In this scenario, the scarcity of forest pr oducts crates the need for the active establishment of forest plantations at a large scale, as is th e case of certain regions in India. Unlike the first path, this one is the pr oduct of a conscious acti on by the government to in crease forest cover to satisfied the populations needs. A third path to forest transitions, is thr ough globalization (Rudel 2002). This path relates increments in forest cover to the complex effects global markets have on a countrys economy. These changes affect the way land is used to m eet foreign market demands and fluctuations in prices in local agricultural pr oducts. Migration, income from rem ittances, fluctuations in global food markets are all ex ternal factors outside a co untrys control that aff ect the way people use the
156 land. Unlike the economic development path, glob alization does not necessarily involve an improvement on the general populations income. This chapter summarizes the lessons learned from the case study of the forest transition happening in El Salvadors Lempa River basin (Fi gure 1-1). El Salvador is located on the pacific coast of Central America and is the smallest, most densely populated country of the mainland Americas. The Lempa River basin covers a bout half of the countrys area (10,000 km2). The study involved the examination of satellite data from 1979-2003 in order to determine the socioeconomic, spatial and climatic factors that may have promoted an increase of 23% in forest cover over the study period (1979-2003). Caracteristics of Forest Transition s Resulting from Globaliz ation The case of El Salvadors forest transition fits partially into both the first and third paths mentioned above. The Forest Scarcity path does not apply to El Salva dor. According to the Ministry of Agricultures records, the amount of forest plantations that have been established in the past years is slightly above 2000 hectares, wh ich is very small percentage of the countrys territory. It is likely that the countrys sm all size has allowed timber from Guatemala and Honduras to relatively good prices reducing the need for the gove rnment to strongly promote local production. On the other hand El Salvadors open trade policies have allowed for its economy to become dependent of global markets. Many of the countrys main incomes depend on foreign income. Assembly line textile factories have employed thousands of low paying urban jobs causing boom towns to appear around duty free z ones. The massive migration to the United States and the subsequent income from remittances has also changed the way rural homes obtain their income.
157 Although to a certain degree there has been an improvement in the incomes of the general population, this improvement does not seem to be substantial enough to trigger a forest transition. We can compare El Sa lvador to the case nearby Cost a Rica, where a quantitative increments in forest cover have been detected (Kleinn et al. 2000) and coincides with significant increases in income, abundant employment opport unities in technology, indus try and tourism, as well as efficient conservation policies (Kull 2007). Costa Rica seem s to fit more smoothly into the economic development path. Unlike Costa Rica, poverty levels in El Salva dor have not changed very much since the 1970s (Briones et al. 2005). Although the country s economy has improved in terms of its GDP over the past decades (MINEC 2008), the distribu tion of this wealth does not seem to have improved the general income levels of the populat ion, particularly in ru ral areas (Briones et al.2005). Even if wages in urban centers have improve d slightly, the results of this study indicate that distribution of poverty is significantly and positively correlated w ith the distribution of forest regeneration (Chapter 3). Globalization has put the pressure on rural population and forced migration and changes in livelihoods w ithout necessarily a lleviating poverty. The results also confirm Rudels (2002) descri ption of the globalization path as destroying first nature and creating sec ond nature. In the Lempa basin simultaneous deforestation and reforestations where taking place during the study period. Most of the deforestation took place in the midland coffee agrogorests surrounding urba n centers and the northern pine forests. Although coffee agroforests are by no means primary forests, they represent one of the oldest tree covers left in the country. At the same time dense, second ary vegetation tended to dominate most of the areas where reforestation took place.
158 One important characteristic of this type of forest transition is that it is not an intentional or even monitored process. The regeneration of forests from globaliza tion is a byproduct of socioeconomic change, and not a conscious e nd sought by society to satisfy forest product demands. Only until recently (2008) has the governm ent el El Salvador started to knowledge the existence of some of these new forests. This means that the potentia l for managing these new forests for timber, ecosystem services or cons ervation has been overlook ed from territorial planning by the government. Forest transitions in Small Tropical Countries The case of El Salvador is an interesting illu stration of the effects of globalization on forest transitions, and can give insights into the characteristics simila r processes happening in tropical developing countries. For one part the countrys small size make it uniform enough to be studied as a unit. In larger countries such as Mexico (Klooster 2003), Brazil (perz and Skole 2003), or India (Foster and Rozensweig 2003), the drivers of forest transitions may vary from one region to the other. Internal differences in the produc tivity and land uses within these countries can make it harder to single out the reasons behi nd the forest cover dynamics. In El Salvador, particularly the Lempa River basin, this internal effect is minimized due to its small size. Small, densely populated countries used to be considered unlikely places for forest cover to regenerate (Allen and Barnes 1985; Angels en and Kaimowitz 1999). Th e Malthusian view of economies would indicate that th e scarcity of land w ould lead population to deplete its forest resources and favor intensive farming on all available land. Although this is true in many countries such as Haiti (Dolisca 2007) and Cameroon (Martens and Lambin 2000), and East Timor (Bouma and Korbyn 2004), other small, populous countries such as Puerto Rico (Grau et al. 2003) and El Salvador (Hecht 2005, chapter 2) seem to have followed a counterintuitive path
159 and actually regain forest cover. In order to understand why these countries have undergone this change, it is important to unders tand the differences in FTT regard ing small and large countries. The way people manage the landscape in these land-scarce countries is different than landrich countries. For example, th e forest transition described by (Rudel et al. 2002) in Ecuador or by (Perz and Skole 2003) in Brazil are very diffe rent than the ones happe ning in El Salvador. Land-rich countries in the trop ics are often characterized by ex tensive farming practices. The general cycle is the expansion of the agricultural frontier to establ ish crops; the depletion of the nutrients from the tropical soils; the change of land use from crops to pastures; and the eventual abandonment as farmers move to more fertile lands. This creates a hallow frontier in which old growth forests are removed and eventu ally replaced by secondary forests. In countries like El Salvador, that type of land use was likely happening around the 1600s during the colonial period (Chapt er 1). However, due to the small size of the country and its growing population during the 20th century, a hallow frontier wa s not a luxury that could be afforded. By the 1970s most of the country was already under constant farming production. It is estimated that only about 20% of the Lempa Rive r basin was under any type of tree cover. Of this forest cover about 9% was composed of sh ade coffee agroforests, and the rest corresponded mostly to protected areas or very steep, un-till able highlands on the north of the country. The intense demand for land in countries such as El Sa lvador means that forest transitions require a shift in peoples livelih ood strategy, rather than only a change in location like it can happen in land-rich countries where people can move to cl ear new lands and continue with their same livelihood strategies. Still, it is not often that de nsely populated countries through a forest transition. Countries like Haiti, for instance, have been severely deforested for many decades and show now sign of
160 forest transitions occurring (Dolis ca et al 2007). By examining th e cases of El Salvador we can see that for a country with its dense population to shift from a marginal agricultural society to forest transitions two factors are key: first, a tr iggering factor that forced people to change their livelihoods drastically, and s econd the opportunity for alternate sources of income. War and Other Triggering Fa ctors of Forest Transitions El Salvadors civil war during the 1980s seem s to have been a key turning point for the countrys forest cover for the period 1979-2003. It is the event that m arked the inflection point in the theoretical Kuznet Curve that FTT takes as a model for forest cove r change (Figure 2-2). From the early 1900s up until the 1970s the countr ys land use trend was the removal of native forests and its replacement for permanent agriculture. The land use varied according to the aptitude of the soils. Sugar cane and cotton dom inated the fertile lowlands, coffee was the preferred crop for the highlands and cattle and pu lse crops dominated the less fertile lowlands (Chapter 1). It is likely the stress of the growing populati on, the lack of fertile soils for production, and the unequal distribution of wealth played an important part in st arting the civil c onflict. From a systems ecological perspective, the balance in the system that perpetuated the social and economic pattern that defined the country for the 20th century reached the point of unsustainability in the late 1970s and created the violent disruption that was the civil war. The problem was further enhanced by the global influe nce of the Cold War, which provided the funds and weapons for the armed conflict lead ing it to become more violent. The violent social change brought by a war will often force people to change their livelihoods as they are displaced (Korf 2004). In the case of El Salvador there was a dramatic increase in out-migration from rural areas as people fled the conflict areas. Migrants were mostly subsistence farmers who were forced to adapt to urban life or migrate abroad. The influx of
161 people into the cities resulted in the growth of large urban slum s and a rise in unemployment and crime (IPCC 2001), this may be the case in simila r places undergoing forest transitions due to war. Although war is definitely not necessary in order fo r forest transitions to take place, it is an example of the kind of disruption that can trigger it in small populous countries. Less violent events can also be the triggers of forest tran sitions. For example, in the case of Puerto Rico (Thomlinson et al. 1996; Grau et al. 2004) no civ il conflict was necessary to start a forest transition. Instead the change came from a drasti c change in US policy during the 1940s after World War II, when Puerto Rico became the recipient of intense amounts of investments directed towards industr ializing its economy. The out migration and urbanization of the populat ion created the conditions that eventually lead to regeneration of forest in the country (Hecht 2005). War in this case can be said to be related partly to globalization given the influences the US and the USSR played in fueling it. The abandonment of lands caused by war did not rise from the creation of economic opportunities but rather from forced migration to escape violence. However, for these chan ges to be lasting once the conflict is over it is importa nt that there exist satisfactor y livelihood options to keep the migrants from returning to farming the marginal lands. Importance of Employment Opportuniti es in the Forest Transition Process Even if disruptions in the soci al order such as civil war was to El Salvador, can start a process of forest transition in a small dens ely populated country, th ey are not enough to perpetuate it. Certain conditions must be present for the transition to be perpetuated. Haiti for example, has been through many periods of civi l unrest, however no las ting forest transition happened. In the case of El Salvador, the conflict ended officially in 1992, since then 12 percent increase in forest cover was detected from 1990 to 2003 in the Lempa River basin. This means
162 that a great part of the population that left the co untryside did not return to reclaim their lands. The results of this study indica te that the lands abandoned tende d to be those with very low agricultural potential, steep slopes, little access to irrigation and roads. The countrys open policy to foreign investment in low-paying industry jo bs, the growth of the construction sector, the income from remittances, and the fall of grain pr ices and rural wedges seem to have made the livelihood changes from the war period endure. In small densely populated countries like El Sa lvador there is very little untouched fertile land that can motivate pioneers to move to a frontier. This is different in places like Guatemala where, after its civil conflict was ove r, many communities that had been displaced by the war had the option to return to agriculture and chose to move into the large tracts of protected land in the Peten Region removing th e forestland there (Shwartz 1995). In El Salvador, migration to the US provide d a more attractive opti on than returning to subsistence agriculture. This can be seen in the increase in the number of emigrants 1990s even after the conflict ended. The rise in urban em ployment in industry and construction during the post-war years may have also helped to refrain a portion of the population from returning to their farm plots. The results of this study also indicate that th e income from remittances has also played an important role in the Salvadorian forest tran sition process (Chapter 3). The municipalities receiving more remittances were more likely to regain forest cover than those with less. This supports the hypothesis that income options are important in perpetuating forest transition by lesseing the need for subsistence agriculture in marginal lands. Another factor of globalization th at could have affected the loss of interest in agriculture is the influx of cheaper agricultural products from abroad. A cut on agricultural subsidies by the
163 government after the civil war (Faber1993; Paige 1999; Utting 1994) may have made the cultivation of beans and maize less attractive for farmers. An influx of cheaper grains, meat and dairy products from abroad may have also pl ayed a part in discour aging production. Klooster (2003) found the northern parts of Mexico to have undergone similar processes due to the competition with low prized US maize. Climate Change and Forest Transitions One relevant factor that th is study found had significant e ffects on forest cover changes was clim ate. Ziervogel et al (2006) relate climat e change to food insecurity which can lead to changes in land use. The results of this study (C hapter 5) indicate that indeed forest cover regeneration in El Salvador may have to do with interannual variabil ity in precipitation. The irregularity in rainfall from one year to another increased the probability of land being abandoned and left to regenerate. While the imp acts on forest cover and the benefits that are brought about by this can have a positive impact on the country as an ecosystem, the abandonment of land can also be seen as signe of increasing in poverty a nd risk for farmers. Areas with high variability in precipitation, particularly those br ought about by the cold phase of El Nio Southern Oscillation seem to be more vulnerable to abandonment. The results indicate that climate may have had an impact as significant as the civil war in certain areas of the central valley in starting, and probably pe rpetuating the forest transition process. The data from El Salvador supports the th esis that climatic variability causes food insecurity and may be the cause for land abandonment and migration. Many of the climatic change scenarios (IPCC 2001) indicate that climate will likely become increasingly variable. It is possible that places dependant on rain-fed agriculture will be likely to experience forest transitions as a result of climate change. Knowing where these vulnerable areas may be a tool for
164 governments and international organizations to bette r plan for the most efficient management of the impacts of climate change. Evidence has been found that food insecurity due to climate change may increase the probability of violent conflic t (Barnett and Adger 2007). Although the results of this study cannot directly relate the civil c onflict and poverty to climatic variab ility, the fact th at all of these variables coincide spatially and are all related to land abandonment may indicate that all of these factors are connected. The occurrence of forest re generation may have bene ficial aspects in the ecological and hydrological systems of a country, but could also be viewed as sign of risk and increasing poverty resulting from globalization and climate change. Countries undergoing these changes should take into consideration these possibilities and take preventive measures to reduce food insecurity or create altern ative employment opportunities for th e inhabitants of these areas. Impact of Spatial Factors on the Distribution of Forest Regeneration The results indicate that roads and accessib ility play a key role in forest regeneration as they do on deforestation. Past research regarding the spatial probabi lities of forest cover removal (Martens and Lambin 2000; Geogheghan et al 2001, Serneels and Lambin 2001; Verburg et al 2002; Veldkamp and Fresco, 2001) indicate proximity to roads are a key factor in deforestation. Access to roads had an inversely proportional effect on the probabilities of reforestation (Chapter 4). However a comparison of the magnitude of the coefficients of logist ic regressions relating these variables indicates that the proximity to roads is more relevant to deforestation than it is to reforestation. Other factors besides accessibility are more important on determining the probabilities of reforestation. One such factor is the proximity to exis ting forest. The probability of forest cover regenerating was significantly hi gher in places in close proxim ity to existing forest patches (Chapter 4). This implies that even if the so cial and economic conditi ons are appropriate for
165 forest transitions to happen, the presence of fore st fragments as sources of seeds and dispersers plays a key role in accelerating forest regeneration. If no seed sources are available, abandoned lands were found to be more likely to remain as scrubs or savannahs instead of reverting to forests. It is very likely that the diversity of new forests will be related to the proximity to forest remnants and the diversity within these. Another factor that increases the probability of forest regeneration is low soil fertility, high slopes and poor access to irrigation. Just as in forest transitions due to economic development, transitions due to globalization manifest themse lves first in the more marginal lands. In developing countries farmers practic ing rain-fed, subsistence agricultu re are likely to be the ones with more risk of food insecuri ty and likely to abandon their lands in favor of other sources of employment. Implications of Forest Transitions Resultin g from Globaliz ation in Small Developing Countries It is of high importance for governments of countries undergoing forest transitions to monitor this process. For one part to quantify the benefits and opportunitie s these new forest may bring to its population, but also to understand the underlying social causes of these transitions and the risks these may present. The increment in forest area brings many bene fits to a countrys ecological support system. It is important for governments to acknowledge and keep track of the distri bution of these forests and the economic value they have in increasi ng water infiltration, fl ood mitigation, reduction erosion and sedimentation of dams, carbon dioxid e sinks, and wildlife habi tat. All these values often go unacknowledged and their inclusion in the economy could serve as an income for people owning lands in the marginal areas where forest transitions tend to happen.
166 Direct economic benefits can also be obtained by efficiently managing new forests. Control over the diversity and composition of these forest, management to promote the regeneration of commercially valuable speci es, as well as promoting aforestation with commercially valuable species can help make the forest transition a sour ce of income for land owners in marginal areas. Another opportunity fore st transitions pose is the integration of these into a carbon offset market, whereby land owners may receive an additional income from foreign companies for the carbon that is being fixed by these secondary forest s. Finally, conservation efforts in countries with forest transitions can greatly benefit from knowing which lands are more likely to remain unused as the forest transition co ntinues. The concentration of protection efforts in places where there is little demand for fa rmlands may help the success of conservation programs. Besides the opportunities new forest present. Countries such as El Salvador, where globalization is the main cause of forest transiti ons, should beware certain social problems that may lie behind this increase in forests. Under these conditions, increments in forest cover may be indicators of poverty, food insecurity, or l ack of employment opportunities. Knowing the location and reasons underlying land abandonment may help manage and focalize development efforts to improve living standards in the area. The monitoring and understanding of forest tran sitions associated with globalization could ideally lead to a shift towards the path of economic developm ent. Using forest transitions as a means to create economic growth through timber production, payment of environmental services, conservation and tourism, may lead to improve livelihoods and perpetuate the existence of these new forests.
167 LIST OF REFERENCES Abram s EM, Rue DJ (1988) The causes and cons equences of deforestation amond prehistoirc Maya. Hum Ecol 16(4): 377-395 Acevedo C, Andrade-Eekhoff K, Huezo M, Maga a A, Pleitez W, Schnell C (2005) Informe sobre Desarrollo Humano El Salvador 2005: Una Mirada al Nuevo Nosotros. El Impacto de las Migraciones. PNUD. San Salvador, El Sa lvador, United Nations Development Program. Acua H (1976) Manual de Cultivos y Enfermedad es Tropicales. Santa Tecla, El Salvador, Centro Nacional de Tecnologa Agropecuaria. Agarwal C, Green G, Grove J, Evans T, Sc hweik C (2002) A review and Assesment of LandUse Change Models dynamics of Space, Ti me and Human Choices. CIPEC Collaborative Report Series. Bloomington, Indiana University. Aide T, Grau H (2004) Globalization, Migratio n, and Latin American Ecosystems. Sci 305: 1915-1916. Alcorn J (1981) Huastec noncrop resource manageme nt: Implications for prehistoric rain forest management. Hum Ecol. 9(4)395-417. Alden D (1965) The Growth and Decline of I ndigo Production in Colonial Brazil: a Study in Comparative Economic History. J Econom Hist 25(1): 35-60. Allen B, Avalos-Sartorio B, Chow J, Aguilar F (2006) Tree Cover Loss in El Salvador's Shade Coffee Areas. Washington D.C., Resources for the Future. Allen J, Barnes D (1985) The Causes of Defore station in Developing Co untries. Annals of the Assoc Am Geogr 75(2): 163-184. Amaroli P (1996) Los Recursos culturales del Parque Nacional El Imposible. Salvanatura. San Salvador. Andrade-Eekhoff K, Kandel S (1998) Reporte de estudio de Factibilidad Socio-Econmico del Proecto:Programa de Desarrollo Econmico para Familias Pobres Receptoras de Remesas en El Salvador. San Salvador, El Salv ador, FEDECACES & HIVOS TRIODOS. Andre M (1998) Depopulation, land-use change a nd landscape transformation in the French Massif Central. Ambio 27(4): 351-353. Angelsen A, Kaimowitz D (1999) Rethinking the causes of deforestation: Lessons from Econom Mod. The World Bank Observ 14: 79-98. Ayalew A, Yamagishi H (2004) The application of GIS-based logistic regression for landslide susceptibility mapping in the Kakuda-Yahi ko Mountains, Central Japan. Geomorphol 65(2005): 15-31.
168 Babtista S, Rudel T (2006) A re-emerging Atlan tic forest? Urbanization, industrialization and the forest-transition in Santa Catarina, so uthern Brazil. Env Conserv 33: 195-202. Barnett and Adger (2007) Climate change, human security and violent conflict. Pol Geograph 2(6) 639-655. Basist A., Bell G, Vernon M (1994) Statis tical Relationships between Topography and Precipitation. J Clim 7: 1305-1315. Bergman J (1969) The distribution of cacao cul tivation in pre-Columbian America. Annals Associ Am Geograph 59: 85-96. Blackman A, Avalos-Sartorio B, Chow J, Aguilar F (2006). Tree Cover Loss in El Salvador's Shade Coffee Areas. Retrieved October 14, 2006. Blake M, Clark J.E, Voorhies B, Michaels G, Love M, Pye M, Demrest A, Arroyo B (1995) Radiocarbon chronology for the late Archaic and Formative periods of teh Pacific coast of southeastern Mesoamerica. Anc Mesoam 6(1995): 161-183. Bouma G, Korbyn H (2004) Change in Ve getation Cover in East Timor,1989-1999. Nat Resourc Forum. 28(1) 1-12. Brand D (1961) The early history of the range cat tle industry in northern Mexico. Agric hist 35(3): 132-129. Briones C, Castro M, Lpez O (2005) Mapa de Pobreza: Tomo I: Poltica Social y focalizacin. F. P. E. Salvador. San Salvador, FLACSO/FISDL. Browning D (1971) El Salvador Landscape a nd Society. Oxford, UK, Clarendon Press. Bullard W (1960) Maya settlement pattern in No rtheastern Peten, Guatemala. Am Antiq 25(3): 355-372. Buttrick P (1971) Forest growth in abandone d agricultural land. Sci Mon 5(1): 80-91. Canales H (1984) Balsamo de El Salvador. San Salvador, El Salv ador, Ministerio de Educacion de El Salvador. Castaneda H (1999) Estudios Generales sobre la especie Myroxylon balsamum en El Salvador. Escuela de Ingeniera Fore stal. Cartago, Instituto Tec nolgico de Costa Rica. Chazdon R (2003) Tropical forest recovery: legacies of human impact and natural disturbances. Perspectives in Plant Ecol Evol Syst 6(1): 51-77. Chelton D, Freilich M, Esbensen S (1999) Satellite Observations of the Wind Jets off the Pacific Coast of Central America. Part I: Case Studies and Statistical Characteristics. Mon Weather Rev 128: 1993-2018.
169 Chevelier F (1963) Land and society in colonial America. Berkeley, Un iversity of California Press. CIAT-PNUMA (1998) Mapa Digital: Cobertura vegetal y Uso de suelos El Salvador. Centro Internacional de Agricultura (CIAT) and Programa de las N aciones Unidas para el Medio Ambiente (PNUMA). San Salvador, El Salvador. Cline-Cole R, Main H, Nichol C (1990) On fuelwood consumption, population dynamics and deforestation in Africa. Wo rld Develop 18(4): 513-527. CNR (Centro Nacional de Registro s) (2006) Registro de sitio Arqueologicos de El Salvador. El Salvador: Personal consultati on of CNR's data base. CNR. (1972). Censo Nacional de Poblacin 1972. Sa n Salvador, Centro Na cional de Registros. CNR. (1992). Censo Nacional de Poblacin 1992. Sa n Salvador, Centro Naci onal de Registros. Cochrane M, Shulze M (1999) Fire as a Recurrent Event in Tropical Forests of the Eastern Amazon: Effects on Forest Structure, Bioma ss, and Species Composition. Biotropica 31(1): 2-16. Cook S, Borah W (1979) Indian food production and consuption in Central Mexico before and after the conquest ( 1500-1650). Essays in population hi story, Mexico and California. Berkeley, University of California Press. 3. Crone A, Baum R, Lidke D, Sather D, Brad ley L, Tall A (2001) Landslides Induced by Hurricane Mithc in El Salvador-An Inventory a nd Descriptions of Selected Features. Denver Colorado, USGS. Cropper M, Griffiths C, Muthukumara M (1999) Roads, Population Pressures, and Deforestation in Thailand, 1976-1989. Land Econom 75(1): 58-73. Cuellar N, Rosa H (1999) Los Servicios Ambientales del Agro: El caso del cafe de sombra en El Salvador. 10/23/2006. Curtis J, Brenner M, Hodell D, Balser R, Islebe G, H Hooghimstra (1998) A multi-proxy study of Holocene environmental change in the Maya Lowlands of Peten, Guatemala. Journal of Paleolimnol 19: 139-159. Daly C, Neilson R, Phillips D (1993) A Statistical -Topographi c Model for Mappinng Climatological Precipitation over Mountainous Terrain. J App Meteorol 33: 140-158. Daniels A (2006) Incorporating domain knowledge and spatial relationships into land cover classifications: a rule-based approach. Internat J Remote Sens 27(14): 2949-2975. Diskin M (1996). Distilled conclusions: The di sappearance of the agrarian question in El Salvador. Lat Am Res Rev 3(2): 111-126.
170 Dolisca F, McDaniel J, Teeter L, Jolly C (2007) Land Tenure, Population Pressure, and Deforestation in Haiti: The case of Fore t des Pins Reserve. J For Econ 13:277-289 Drake WD (1993) Towards building a theory of population-environment dynamics: A family of transitions. In: Ness GD, Drake WD, Brechin SR (eds) Population-environment dynamics. Ann Arbor, University of Michigan Press: 305-355. Dull R (2007) Unpacking El Salvador's ecologica l predicament: Theoretical templates and longview ecologies. Glob Environ Change 18(2): 319-329. Ehrhardt-Martinez K (1998) Social Determinats of Deforestati on in Developing Countries: A Cross-National Study. Social Forces 77(2): 567-586. Escobar J (1965) El ail en la economia de El Salvador. Economa Salvadorea. Instituto de Estudios Economicos. San Salvador Universidad de El Salvador. ESRI (2006) ArcMap Version 9.1[GIS software]. Redlands, CA, Environ Syst Res Inst. Faber D (1993) Environment under fire: Imperialism and the ecological crisis in Central America. NY, Mon Rev. FAO (2000) Forest Resource Assesment, 2000. Ro me, Food and Agricultural Organization of the United Nations. FAO (2001) State of the World's Forests. Ro me, Food and Agricultura l Organization of the United Nations. Ferraz G, Russel G, Stouffer P, Bierregaard R, Pr imm S, Lovejoy, T (2003) Rates of species lost from Amazonian forest fragments. Proc Nat Acad Sci Fimanow M, Ariza E (2001) Biodive rsity of Flora and Fauna in Shaded Coffee Systems. Lima, Peru, International Centre for Research in Ag roforestry, Latin American Regional Office. Finegan B (1996) Pattern and process in neotropical secondary rain forests: the first 100 years of succession. Tree 11(3): 119-124. Foster A, Rosenzweig M (2003) Economic Growth and the Rise of Forests. Quart J Econom May(2003). Foster D (1992) Land-use history (1730-1990) and vegetation dynamics in central New England, USA. J Ecol 80: 753-772. Fowler W (1988) La poblacin nativa de El Salv ador al momento de la conquista Espaola. Mesoamericana 15: 79-116. FUSADES (1998) Rural development study. San Salvador, El Salvador, FUSADES (Fundacin Salvadorea para el Desarrollo Econmico y Social).
171 Gallina S, Mandujano D, Gonzalez-Romero A ( 1996) Conservation of mammalian biodiversity in coffee plantations of Central Ver acruz, Mexico. Agrofor Syst 33: 13-27. Geoghegan J, Cortina Villar S, Klepseis P, Ma cario P, Ogneva-Himmelberger Y, Chowdhury R, Turner B, Vance C (2001) Modeling tropi cal deforestation in the southern Yucatn peninsular region: comparing survey and sate llite data. Agric, Ecos yst & Environ 85(1-3): 25-46. Gmez I, Garca M, de Larios S (2005) La Ge stin Territorial Particip ativa Hacia la Busqueda de Medios de Vida Rurales Sostenibles: El Caso de la Manc omunidad la Montaona. Documentos de Trabajo. San Salvador, PRISMA. Grainger A (1995) The forest tr ansition: An alternative appr oach. Area 27(3): 242-251. Grau R, Arturi M, Brown A, Aceolaza P (1997) Floristic and structural patterns along a chronosequence of forest succession in Argentin ean subtropical montane forests. For Ecol Manag 95: 161-171. Grau R, Mitchel A, Zimmerman J, Thomlinson J, Helmer E, Zou X (2003) The ecological consequences of socioeconomic change and land -use in post-agricultural Puerto Rico. BioSci 53(12): 1159-1168. Guariguata M, Ostertag, R (2001) Neotropical secondary forest su ccesion: changes in structural and functional characteristics. For Ecol Manag 148: 185-206. Hampshire R (1989) El Salvador. In: Cambell DG, Hammond HD (eds) Floristic Inventory of Tropical Countries: The Status of Plant Syst ematics, Collections and Vegetation, plus Recommendations for the Future. New York, New York Botanical Garden. Hastenrath S (1963) Mapa preliminar de la pr oduccin anual de lluvia en El Salvador (Map). Servicio Meteorologico Nacional. San Salvador. Hastenrath S (1988) Climate a nd Circulation in the Tropics. Reidel, Atmospheric Sciences Library. Hecht S, Gomes I, Cuellar N, Rosa H (2000) Human or Natural World Develop 34(2): 308-323. Hecht S, Gmez I, Cuellar N, H Rosa (2005) Globalization, Forest Resurgence, and Environmental Politics in El Salvador. World Dev 34(2): 308-323. Hecht S, Saatchi S (2007) Globalization and Fore st Resurgence: Changes in Forest Cover in El Salvador. BioSci 57(8): 663-672. Heckenberger M, Kuikuro A, Kuikuro U, Russell C, Schmidt M, Fausto C, Franchetto B (2003) Amazonia 1492: Pristine Forest or Cultu ral Parkland? Sci 301 (5640): 1710-1714. Hodell D, Brenner M, Curtis J, Guilderson T (2 001) Solar Forcing of Drought Frequency in the Maya Lowlands. Sci 292: 1367-1370.
172 Holdridge L (1975) Life Zones Map for El Salvad or. San Salvador, El Sa lvador, Ministerio de Agricultura de El Salvador. Holdridge L (1978) Ecologa basada en zonas de vida. San Jos, Costa Rica, Instituto Interamericano de Ciencias Agrcolas. Holl K (1999) Factors Limiting Tropical Rain Fore st Regeneration in Abandoned Pasture: Seed Rain, Sees Germination, Microclimat e and Soil. Biotrop 31(2): 229-242. Hubaceck K, Vazquez J (2002) The Economic s of Land Use Change. Interim Report. Laxenburg, Int Inst App Syst Anal. Hurtado A, Hill K (1989) Experimental studies of tool efficiency among Machiguenga women and Implications for root-digging foragers. J of Anthr Res 45(2): 207-217. IPCC 2001 Climate Change 2001: synthesis report A contribution of Working Groups I. II and III to the Third Assesment Report of the iNtergovernmental Panel on Climate Change. Atson RT and the Core Writing Team eds. Ca mbridge University Press, Cambridge. Janzen D, & P.S. Martin (1982) Neotropical An achronisms: fruits the Gomphotheres ate. Sci 215(4528): 19-27. Jensen J (2005) Introductory Digital Image Pro cessing. Upper Saddle River, NJ., Prentice Hall. Jibson W, Crone A, Harp E, Baum R., Major J, Pullinger C, Escobar E, Martinez M, Smith M (2004) Landslides triggered by the 13 Ja nuary and 13 Febuary 2001 earthquakes in El Salvador. In: Rose W, Lopez D, Carr M, Major J (eds) Natural Hazards in El Salvador. San Salvador, El Salvador, Geol Soc Am. JMA (2003) ENSO index According to the JMA SSTA (1868present). Center for OceanicAtmospheric Prediction Studies (COAPS) in 2005. http://coaps.fsu.edu/jma.shtml Kaimowitz D, Faune A. 2003. Contras and Comandantes: Armed Movements and Forest Conservation in Nicaraguas Bosawas Biosphe re Reserve.. In: Price SH (Ed) War and Tropical Forests: Conservation in Areas of Ar med Conflict. The Haworth Press, New York. Kandel S (2002). Migraciones, medioambiente y pobreza rural en El Salvador. Documentos de Trabajo. San Salvador, PRISMA. Kleinn C, Corrales L, Morales D (2000) Forest Area in Costa Rica: A Comparative Study of Tropical Forest Cover Estimates Over Time. Environ Mon Assessm 73:17-40 Klingebiel A, Montgomery PH (1961). Land capability cla ssification. Agriculture handbook no. 210. Washington D.C., USDA. Klooster D (2003) Forest Transitions in Mexi co: Institutions and Forests in a Globalized Countryside. Prof Geogr 55(2): 227-237.
173 Komar O (2003) Priority conservation areas for birds in El Salvador. Anim Conserv 5: 173-183. Korf B (2002) War Livelihoods and Vulnerabi liy in Sri Lanka. De velop Chng 35(2):275-295. Kull CA, Ibrahim CK, Meredith TC (2007) Tropi cal Forest Transitions and Globalization: Neoliberalism, Migration, Touris m, and International Conserva tion Agendas. Soc Nat Resour 20: 723-737. Lamelas MT, Marinoni O, Hoppe A, de la Riva J (2007) Doline probabil ity map using logistic regression and GIS technology in central Ebro Basin (Spain ). Environ Geogr 54: 963-977. Lentz D, Ramirez-Sosa C (2003) Cern Plant Resources: Abndance and Diversity. Before the Volacano Erupted: The Ancient Cern Village in Central Amercia. P. Sheets. Austin, TX, University of Texas Press. Lentz D L, Beaudry-Corbett P, Reyna de Aguila r M, Kaplan R L (1996) Food stuffs, forests, fields and shelter: A Paleobotanical analysis of vessel contents from the Ceren site, El Salvador. Latin American Antiquity 7: 247-262. Lloyd CD (2005) Assesing the effect of integrating elevation data into th e estimation of monthly precipitation in Great Britain. Journal of Hydrology 308: 128-150. Love M (1999) Ideology, material culture and daily practice in Pre-Classic Mesoamerica: a Pacific Coast perspective. Social Patterns in Pre-Classic Mesoamerica. D. C. Grove, & R. Joyce. Washinghton D.C., Dumbarton Oaks. Lovell WG (1992) Heavy shadows and black ni ght: Disease and depopulation in Colonial Spanish America. Ann Asso c Am Geogr 82(3): 426-443. Lozano F, Cottin JF (2002) Electricicacion co n Base en recursos de Energia Renovable. PNUD/GEF. San Salvador, PNUD/GEF: 5. Lucas R, Rowlands A, Brown A, Keyworth S, Bunting P (2007) Rule-based classification of multi-temporal satellite im agery for habitat and agricultural land cover mapping. J Photogramm & Remote Sens 62: 165-185. Lucas RM, Honzk M, do Amaral I, Curran PJ, Foody GM (2002) Forest regeneration on abandoned clearances in central Amazonia. Intern J Remote Sens 23(5): 965-988. Lungo M, Baires S, Andrade-Eekhoff K (1997) Migracion Internacional y Desarrollo Tomo I y II. San Salvador, El Salvador, FUNDE. MAG (1998) Mapa Digital de Cantidades Normales Anuales de Lluvia (mm) El Salvador 1998. San Salvador, Direccin General de Economa Ag ropecuaria and Minister io de Agricultura y Ganadera. MAG (2005) Registro de plant aciones forestales. San Salvador Direccin General Forestal, Ministerio de Agricultura. San Salvador El Salvador.
174 Magaa V, Amador JA, Medina S (1999) Th e midsummer drought over Mexico and Central America. J Clim 12(6): 1577-1588. MAG-DGRNR-PAES-IICA (1997) Memoria de Taller de consulta sobre el manejo de incentivos para promover actividades de conservacin de suelos y agroforestera,, San Salvador, El Salvador, Ministerio de Agri cultura de El Salvador. Mahler S (2000) Migration and Transnational Issues: Recent Trends and Prospects for 2020., Institute of Iberoamerican Studies, Inte r-American Dialogue, Florida International University. Maldenoff DJ, Sickley TA, Wydeven AP (1999) Predicting Gray Wolf Landscape Recolonization: Logistic Regre ssion Models Vs. New Field Data Ecol Applic 9 (1): 37-44 MARN (1998) Mapa pedologico de El Salvador Ministerio de Medio Ambiente. San Salvador El Salvador. Marquinez J, Lastra J, Garcia P (2003) Es timation models for preci pitation in mountainous regions: the use of GIS and multivariate analysis. J Hydrol 270: 1-11. Martens B, Lambin EF (2000) Land-Cover-Cha nge Trajectories in Southern Cameroon. Ann Assoc Am Geogr 90(3): 467-494. Mather AS (1992) The forest transition. Area 24(4): 367-379. Mather AS, Needle CL, Fairbairn J (1998) The human drivers of global land cover change: the case of forests. Hydrol Process 12: 1983-1994. Mather, A. S., Fairbairn, J., and Needle, C.L. (1999). The Course and Drivers of the Forest Transition: the case of France. J Rural Stud 15(1): 65-90. Mathieu JR, Meyer DA (1997) Comparing Axe H eads of Stone, Bronze, and Steel: Studies in Experimental Archeology. J Field Archeol 24(3): 333-351. McCArthy JJ, Canziani OF, Leary NA, Gokke n DJ, White KS (2001) Climae Change 2001: Impcts, Adaptations, and Vulnerability. Camb ridge University Press. Cambridge, UK. McGrail K, Marks B, Ene E, Holmes C (1995) FRAGSTATS. Amherst, MA, University of Massachusetts. PRISMA (2002) National Level Environmental An alysis: Report to the World Wildlife Fund. San Salvador, PRISMA. Menard S (2002) Applied Logistic Regresin An lisis. Sage publicati ons. Thousand Oaks, CA. Mendez E, Bacon C (2005) Medios de vida y conservacin de la biodiversidad arbrea: las experiencias de las cooperativas cafetaleras en El Salvador y Nicaragua. Revista de Agroecologia: 27-30.
175 MINEC (2008) Desempeo Macroeconomico: PIB. Ministerio de Economia de El Salvador. http://www.elsalvadorcomp ite.gob.sv/Principal.html Morisey R J (1951) The northward expansi on of cattle ranching in New Spain, 1550-1600. Agricult Hist 25(3): 115-121. NASA (2000) Shuttle Radar Topography Mission Data. NASA (2005) MODIS rapidfire data Rapidfire Response System. http://rapidfire.sci.gsfc.nasa.gov/ Neff H, Pearsall DM, Jones JG, Arroyo B, Collins SK, Friedel DE (2006) Early Maya adaptive patterns: Mid-Late Holocene pa leoenvironmental evidence from Pacific Guatemala. Lat Am Antiq 17(3): 287-315 Paige J (1999) Coffee and Power. Cambridge, Harvard University Press. Perz S, Skole DL (2003) Secondary forest expans ion in the Brazilian Amazon and the refinement of forest transition theory. So ciety Nat Resourc 16: 277-294. Perz S (2007) Grand Theory and Context-Specifi city in the Study of Forest Dynamics: Forest Transition Theory and Other Directio ns. Profess Geogr 59(1): 105-114. Peterson G, Cumming G, Carpente r S (2003) Scenario Planning: a Tool for Conservation in and Uncertain World. Conserva tion Biology 17(2): 358-366. CCP (2007) Base de datos CCP. from http://ccp.ucr.ac.cr /bvp/docum entos.htm Centro Centroamericano de Poblacin. Pohl ME, Pope KO, von Nagy C (2002) Olm ec Origins of Mesoamerican Writing. Sci 298(5600): 1984-1987. Portig WH (1965) Central American Rain fall. Geographical Review 55(1): 68-90. PRISMA (2002) National Level Environmental An alysis. Report to the World Wildlife Fund. San Salvador, PRISMA (Programa de Investig acin sobre Desarrollo y Medio Ambiente). PROCAFE (2004) Census of Coffee Growers Sa n Salvador, PROCAFE (Fundacin Salvadorea para Investigaciones del Caf. Rodrguez A (1999) Migracin, Sociedad y cultura en Nueva Concepcin: Una Revisin desde la Etnogrfica. In: Lungo M, Kandel S (eds) Migracin Internacional y Actitudes Frente al Trabajo en reas Rurales. San Salvador, FUNDE. Rogerson PA (2001) Statistical Methods for Geography. Thousand Oaks, SAGE Publications. Rojas-Rabiela T (1985) La tec nologa agrcola mesoamerican a en el siglo XVI.. In: RojasRabiela T (ed) Historia de la agricultura poca prehispnica-siglo XVI. Mxico D. F. Instituto Nacional de An tropologa e Historia.
176 Rudel T (2002) Paths of Destruction and Regene ration: Globalization and Forests in the Tropics. Rur Soc 67(4):622-636. Rudel, T. K., Bates, D., and Machinguiashi, R. (2002). A Tropical Forest Transition? Agricultural Change, Out-migration, and Sec ondary Forests in the Ecuatorian Amazon. Ann Assoc Am Geogr 92(1): 87-102. Rudel TK, Coomes OT, Moran E, Achard F, Arild A, Xu J, Lambine E (2005) Forest transitions: towards a global understanding of land use ch ange. Global Environ Chang 15: 23-31. Sader SA, Ahl D, Liou W (1995) Accuracy of Rule-Based Methods for Forest Wetland Classification in Maine. Remo te Sens Environ 53: 133-144. Schellas J, Greenberg R (1996) Fo rest Patches in Tropical Landscap es. Island Press, Washington DC. Seligson, MA (1995) Thirty years of Transformati on in the Agrarian Structure of El Salvador, 1961-1991. Lat Am Res Rev 30(3): 43-74. Serneels S, Lambin EF (2001) Proximate causes of land use change in Narok district Kenya: a spatial statistical model. Agri cu, Ecosyst Environ 85: 65-82. Sheets P, Woodward M (2003) Cultivating Biodiver sity: Milpas, Gardens and the Classic Period Landscape. In: Sheets P (ed) Before the Vol cano Erupted: The Ancient Ceren Village in Central America. Austin, TX, University of Texas Press. Shwartz, N (1995) Colonization, Development and De forestation in Peten, Northern Guatemala. In: Painter M, Durham W eds. The social cau ses of environmental destruction in Latin America. University of Michigan Press. Ann Arbor, Michigan. Skole D, Tucker C (1993) Tropi cal Deforestation and Habitat Fragmentation in the Amazon: Satellite Data from 1978 to 1988. Sci 260(5116): 1905-1910. Sloan S (2008) Reforestation amidst deforesta tion: Simultaneity and su ccession. Global Environ Chang 18: 425-441. Smith R (1959). Indigo Production a nd Trade in Colonial Guatemal a. Hisp Hist Rev 39(2): 181211. SNET (2005) Historical precipita tion database. Sistema Nacional de Estudios Territoriales, Ministerio de Medio Ambiente San Salvador, El Salvador. SNET (2006) Cronologa de sismos destructivos en El Salvador. Accesed in 2007 from http://www.snet.gob.sv/ver/sismologia/registro/estadisticas/ Southworth J,Tucker K (2001) The influen ce of Accessibility, Local Institutions and Socioeconom ic Factors on Forest Cover Change in the Mountains of Western Honduras. Mt Res Develop 21(3): 286-283.
177 Spellerberg IF (1996) Plantation fo rests protect biodivers ityToo much of a generalization to be true. New Zeland Forest. 39:19-22. SPSS Inc. (1999) SPSS 11.0. Chicago, Il. Staaland H, Holand O, Nellemann C, Smith M (1998) Time scale for forest regrowth: Abandoned grazing and agricultural areas in s outhern Norway. Ambio 27(6): 456-460. Stratton RA, Gaesley WK (1949) Reconnaissance geology and vertebrate paleontology of El Salvador, Central America. Bull Geolog Soc Am 60(1949): 1731-1754. Super JC (1988) Food, conquest, and colonizatio n in sixteenth-century Spanish America. Albuquerque, University of New Mexico Press. Terborgh J (1999) Requiem for Natu re. Washington, Island Press. UN (1993) Informe de la Comision de la Verda d. United Nations. San Salvador, El Salvador. http://www.nacionesunidas.org.sv/nues.htm Utting P (1994) Trees, people and power. London, Earthscan. Veldkamp A, Lambin EF (2001) Predicting La nd-use change. Agric, Ecosyst Environ 2001(4): 1-6. Veldkamp A, Fresco LO (2001). CLUE-CR: an integrated scale model to simulate land use change scenarios in Costa Rica. Ecol Model 91: 231-248. Verburg PH, Soepboer W, Veldkamp A, Limpia da R, Espaldon V, Mastura S (2002) Modeling the Spatial Dynamics of Regional Land Use: The CLUE-S Model. Environ Manag 30(3): 391-405. Waylen PR, Caviedes, CN, Quesada Quesada, ME (1996) Interannual Variability of Monthly Precipitation in Costa Rica. J Clim 9: 2606-2613. Waylen PR, Quesada ME, Caviedes CN (1996) Vari ability of annual precip itation in Costa Rica, and the Southern Oscilla tion. J Clim 16: 173-195. Webster, J. W., Brook, G.A., Railsback, L.B., Cheng, H., Edwards, R.L., Alexander, C., & Reeder, P.P. (2007). Stalagmite evidence from Belize indicating sgnificant droughts at the time of thePreclassic Abandonment, the Maya Hiatus, and the Classic Maya Collapse. Paleogeogr, Paleoclim, Paleoecol 250(2007): 1-17. Wilken G (1987) Good Farmers. Traditional Agricultural Resource Management in Mexico and Central America. Berkeley, CA., University of California Press. Wijdeven SM, Kuzee ME (2000) Seed Availability as a Limiting Factor in Forest Recovery in Costa Rica. Restor Ecol 8(4): 414-424.
178 Whitmore TM, Turer BL (1992) Landscapes of Cu ltivation in Mesoamerica on the Eve of the Conquest. Annals Associ Am Geograph 82(3): 402-425. Zhang, P., Cheng, H., Edwards, R., Chen, F., Wa ng, Y., Yang, X., Liu, J., Wang, X., Liu, J., an, C., Dai, Z., Zhou, J., Zhang, D.,Jia, L., & K. Johnson. (2008). A Test of Climate, Sun, and Culture Relationships from an 1810-Year Chinese Cave Record. Sci 322: 940-942. Ziervogel G, Nyong A, Osman B, Conde C, Sergi o, C, Downing T (2006) Climate Variability and Change: Implications for Household Food Security. AIACC Working Papers. Washington D.C., Assesments of Impacts and Adaptations to Climate Change (AIACC). Zilberg E, Lungo M (1999) It's made them l azy?: Youth, Migration and Labor Identities. Transforming El Salvador: Migration, Societ y and Culture. M. K. Lungo, S. San Salvador, FUNDE. Zimmerman J, Pascarella J, Ai de M (2000) Barriers to Forest Regeneration in an Abandoned Pasture in Puerto Rico. Re stor Ecol 8(4): 350-360.
179 BIOGRAPHICAL SKETCH Hector Castaneda was born in El Salvador in 1975. He studied his bachelors in forestry at the In tituto Tecnologico de Costa Rica. After co mpleting his degree he worked for two years in the non-profit sector of El Salvador. He was ch ief of the Scientific Section of La Laguna Botanical Gardens. Later went on to study hi s masters in Interdisciplinary Ecology with emphasis in Tropical Conservation and Development at the University of Florida. His masters research consisted working with indigenous peop le in Costa Rica and evaluating the importance the forest has as a source of wild foods to th em. After completing this degree he continued on to his Ph.D also in Interdisciplinary Ecology but with a focus on geographic information systems and remote sensing applications to natural resource management. During this time he worked as a consultant for the Wild Life Conservation Society in Peten, Guatemala in the mapping of vegetation and archeological sites of Mirador-Rio Azul Nationa l park as well as on a fire risk model for the Peten Region. Hector has also taught and coordinate d for three years the Ethnobiology course as a part of the Duke study abroad program in Costa Rica. He has also worked as a consultant in the private forestry sector of El Salvador in the establishing nurseries and plantations for national and foreign investors.