Is Precipitation Driven by the Atlantic Multidecadal Oscillation Influencing Land Use and L and Cover in Florida Carly Muir University of Florida
Abstract C limatic variability and potential anthropogenic im pacts have caused precipitation to undergo many changes in the past century. A recent statewide study of from 1981 to 2012 (Tsai et al.) suggested that vegetation has become greener, or less dry, in the winter and drier in the summer. If so, this is in keeping with predictions of increased winter precipitation in the southeast US, under conditions of climate warming. Winter (November April) and summer (May October) are created from the NOAA derived monthly precipitation data for each of seven climate divisions over the period 1981 2012 After conducting simple regression analys e s it is apparent that winter rainfall has actually decreased while summer rainfall has increased during this time period Although tests of residuals indicate normality and serial independence, simple plots of seasonal rainfalls suggest that an abrupt change occurred in the last half of the 1990s coincident with a change in the state of the Atlantic Multi decadal Oscillation (AMO) Application of the hypergeometric distribution to the number of years returning above median seasonal rainfall totals before and after the late1990s shift in the AMO statistically confirm this visual impression. The AMO is an extremely important influence on water resources of the state and does not currently explain the proposed increases in winter precipitation from climate change scenarios but suggests that factors, other than a simple correlation with seasonal rainfall are producing the observed patte rns of changes in vegetation greenness
Introduction Climate change is expected to alter rainfall patterns both regionally and globally. In turn precipitation is one of the major controls on vegetation (Karl and Knight 1998), thus any changes in precipitation climatology is likely to result in a change of land cover and land use (Cleland et al. 2007 ). Florida has been growing rapidly over the past 70 years and currently ranks as the fourth most populace state in the US. Pop ulation of Florida has already passed 19 million, and is projected to continue (US Census); putting many at risk from adverse effects of climate change. In order to properly manage land use and land cover, scientists, land use planners and policy makers mu st understand the reason for changes in landscape. Tourism and agriculture constitute the two mainstays of the economy. The second la rgest tourist attraction following theme parks beauty. Changes in land cover and land use could th erefore have considerable impacts upon the state economy. Vegetation acts as a surface to the atmosphere and in moderating the energy balance of the surface. Changes in vegetation could therefore induce furthe r changes in local, regional and even hemispheric climate ( Marshall et al. 2004 ). Finally, vegetation acts as a good short term store of the atmospheric greenhouse gas, carbon dioxide. An understanding of fluctuations in natural factors which may drive changes in vegetation cover is therefore of considerable economic and environmental interest to the state. With this in mi nd, recent research (Tsai et al. under review) investigated changes in Florida veget ation since 1981 through remote sensing techniques and the Normalized Difference Vegetation Index (NDVI) Analyse s of the monthly s atellite derived index show ed an increase in vegetation green during the winter months and a decrease during the summer (f igure 1) Land use and land cover change can now be analyzed using remote sensing, and has created a new method of studying human impact on the Earth. Examining satellite imagery is critical in understanding the role that land use and land cover change have on the environment ( Lambin et al. 2001). Monitoring alterations of land cover has become valuable to predicting changes in regional climate. However, it is still unclear whether th e
observed change s are due to anthropogeni c factors like urbanization, changes in agricultural and de and re forestation, or a result of climatic variability and or change Figure 1 Values of directional persistence metric, D, for (a) January to (l) December 1982 2006 from Tsai et al., ( under revie w ) Directional persistence is calculated based on comparisons of NDVI observation in a month to the observation in the same some in the base year 1982 Positive scores (greening) are shown in hues of green and negative scores (declining greenness) are in red. In this study monthly precipitation data from the National Oceanic and Atmospheric Administration (NOAA) climate divisions within the state are analyze d over the same time period as the available satellite imagery to determine whether long term changes in rainfall patterns may be responsible for the broad temporal (summer/winter) and spatial patterns observed in changes of NDVI
C limate change is one potential cause of changes in precipitation regime. Several models suggest that there will be an increase in winter rainfall and a decrease in summer rainfall over majority of the Northern hemisphere ( Climate Impact in the Southeast EPA ) and in particular the southeaster n United States Observed changes in NDVI may be a manifestation of the early onset of this effect. Low frequency (over decades) climatic variability arising from long term changes in interactions between oceans and the atmosphere above them, could also have similar impacts. One such climatic factor tha t has been shown to be extremely influential in determining precipitation in the Southeast United States is t he Atlantic Multidecadal Oscillation (AMO). The AMO represents a relatively small change of about 0 .4 C in North Atlantic sea surface temperatures that occurs over several decades (Enfield et al., 2001 ). Although its range is small, its spatial extent is large and it therefore represents tremendously large changes in the total energy stored in and released from the ocean influencing weather patterns throughout the eastern US. The AMO has been linked to dr ought patterns throughout North America as well as hurricane severity ( Johns and Lee 2012 ) and has considerable effect on water resources. Between warm and cool phases estimate d inflows to Lake Okeechobee vary by as much as 40% ( Enfield et al. 2001) While most of North America receives less rain during AMO warm phases, a study using fl ow data from several rivers in south central Florida shows that parts of the state actually exhibit the opposite relationship to the AMO (Kelly et al. 2008 ) The precipitation regime c hanges across Florida from more continental in the north (winter and summer seasonal peaks) to distinctly seasonal (summer maximum) sub tropical in the south In 1995 the AMO entered a warm phase (Koch Rose, Mitsova Boneva & Root, 2011) meaning that have changed within the time of the study, and that the nature of that change may have been felt differently across the state This change to a warm phase has also been posited as a potential cause of the observed temporal and spatial patterns of greening (Tsai et al., under review) A second well established control on Florida precipitation is El Ni o Southern Oscillation (ENSO) Wa rm phases of ENSO tend to yield increased precipitation totals in the state, while
cold phases encourage drier conditions (Douglas & Englehart, 1981; Ropelewski & Halpert, 1987; Kahya & Dracup, 1993), particularly during the winter. This thesis proceeds w ith a brief review of precipitation climatology of Florida, and more particularly of the seven climate divisions, within which NOAA considers climate to be fairly homogeneous. This is followed by a detail ed explanation of the statistical methodologies employed and a description of the results. Finally the significance of the results is discussed with regard to the motivation for the research and broad conclusions are drawn. Study Area & Purpose This study was conducted using monthly precipitation da ta from the seven NOAA divisions of Florida (figure 2) within each of which climate is generally considered to be reasonably homogenous. The divisions also appear prominently in the remote sensing research conducted by Tsai et al. (under review).
Figu re 2. The six mainland climate divisions of Florida employed in the study. Throughout the text the Northwest division is ref erred to as climate division 1, North as climate division 2, North Central as division 3, South Central as division 4, Everglades a nd SW Coast as division 5, and Lower East Coast as division 6. The seventh division corresponds to the Florida Keys. After T sai et al. (under review) NOAA provides mean monthly data within a division for each year, based on historic records from 1895 to 2 012 ( http://www1.ncdc.noaa.gov/pub/data/cirs/drd964x.pcp.txt ). Although there might be some question about the representativeness of the early records given the limited number of observing stations, the figures are based on many stations during the period of interest of this study. Positive and negative AMO phases were established using years determined by a study published by the State University System of Florida ( Koch Rose, Mit sova Boneva and Root 2011) P recipitation climatology v aries across the state. Figure 3 indicates that maximum annual to tals (often above 60 inches, or 1500mm) occur in the western panhandle. A second area of elevated annual rains (54 inches, 1300mm) is found along the southeastern coastal strip. A broad zone of slightly lower totals (under 50 inches, 1250 mm) runs from the northeast to the southwest of the
state. Figure 3. Mean annual precipitation totals (inches) across Florida.
Figure 4. Mean monthly precipitation totals (mm) 1895 2012 in the seven NOAA climatic divisions of Florida. Vertical dotted lines delimit those months considered to comprise winter and summer seasons in the research. The northern and panhandle divisions receive both winter frontal precipitation and convection rains (figure 4) as two distinct peaks. However, the fronts generally weaken before reaching South Florida (Jordan 1984) and their contribution diminishes progressively southwards producing a very pronounced si ngle, summer peak in south Florida. The purpose of this study is to assess how seasonal precipitation and climatic variability may explain the change in land cover. Figure 4 shows that the use of calendar years would divide the winter rainy season be tween two years, particularly in the northern climate divisions. Therefore the winter season is considered to start in November of one calendar year and proceed through to April of the next, while summer runs from May to October. Figure 4 shows these div isions to be applicable throughout the state. Methods Simple linear r egre ssion is used to detect any long term chan ges in precipitation over time. S easonal precipitation totals (in mm) for each year are derived by aggregating the six monthly totals within Summer (May October), and Winter (November April) month s Seasonal totals are also expressed in terms of the percentage of the annual (winter plus summ er) precipitation that occurred in ea ch season. Given the complementary nature of summer and winter percentages, analyses are only carried out on summer percentages, with the opposite behavior inferred for winter ones. The precipitation variable, P, constitutes the dependent variable, y, an d the year, t, the independent variable x, in a simple linear relationship: Where three of these variables, in each of the seven climate divisions over the entire period of record, from 1895 to 2012 (table 1)
Div 1 Div 2 Div 3 Div 4 Div 5 Div 6 Div 7 Summer + + + Winter + + + + + + + %Summer Table 1. Signs and significance levels of the slopes of the fitted simple linear Regression, applied to the seasonal precipitation variables, 1895 2012. Shaded cells represent a significance level of .05. For the time period corresponding with the satellite imagery data was used from 1981 2012. Statistical tests of significance are carried out on both the estimate values of slope and the derived correlation coefficient. This reduction in temporal scale allowed a much better examination of how the potential linkage between precipitation and land c over change as seen in the satellite imagery. The assumption of normally distributed data was tested using the Kolmogorov Smirnoff test, and confirmed the assumption. Serial independence was tested by calculating standard deviation for residuals, and exam ining correlation between each year and the preceding year. The final assumption for the linear regression was homoscedasticity. An F test was employed to confirm a constant variance throughout the data and the results confirmed validity of the assumptio n. The hypergeometric distribution is used to test the notion that seasonal rainfall totals may have changed (increased or decreased) since the late 1990s when the AMO shifted into its current warm phase. It is ideal as it requires no assumptions about the distribution of the seasonal rainfall totals and permits calculation of the probability of observing a particular value of successes or failures, without replacement in the sample (Burt et al. 2009). The long run (1981 2012) median seasonal rainfalls are calculated f or each climate division. Data are then split into two periods, 1981 1995 represents the AMO cool phase, and 2000 2012 corresponds with an AMO warm phase. Within each period a success is defined as a seasonal precipitation
total that exce eds the long run median and a failure represented a total below median rainfall. The successes were summed for each season during both time periods, and the probability of getting the observed value simply by chance was calculated W here p(x) is the probability of sampling x successes/failure (x above/below seasonal long term median precipitation). N is the size of the population (1981 2012), n is the sample size (15 in the cool phase of AMO and 13 in the warm p ha se), and k is the total number of observations with the desired property ( above or below median seasonal totals, N/2) in the available record. The null hypothesis throughout is that the observed number of seasonal precipitation totals falling above (below) the median in the phases of AMO are not significantly different from that expected at random at the 0.10 and 0.05 significance level. The hypergeometric test can also be employed in a similar fashion to determine whether seasonal precipitation totals consistently fall above or below their long term medians (Ropelewski & Ha lpert, 1987; Grimm et al., 2000; Mason & Goddard, 2001; Waylen et al., 2012) Using the Florida State University COAPS classification, years are divided into one of three classes, El Nio (Warm phase El Ni o Southern Oscillation (ENSO) ), La Nia (Cold pha se ENSO) and neutral ( http://coaps.fsu.edu/jma ). Results Simple Linear Regression Regression performed over the period 1895 2012 only yielded three significant value s two in division one, and the other in summer of division 6 Running the regression for only the years of satellite imagery ( 1981 2012 ) yielded many more significant values, although levels of significance did vary between the variables and divisions ( Table 2 ). Winter precipitation shows cons istent decreases in all climates divisions, which are significant except in division 4. Summer rainfalls appear to be generally increasing over this
period, except in division 1. However only in division 6 was the outcome significant. The overall impress ion of a statewide increase in summer rains and decrease in winter totals is reinforced by results for the percentage of annual precipitation falling in the summer, which increased significantly in all divisions, apart from division 1 I t is interesting to note that although the trend in both winter and summer totals has been downward in division 1, the western panhandle, the decline of winter rainfall must be greater than that of summer, as the trend in summer rainfall percentages is positive. Div 1 Div 2 Div 3 Div 4 Div 5 Div 6 Div 7 Summer + + + + + + Winter %Summer + + + + + + + Table 2 Signs and significance levels of the slopes of the fitted simple linear Regression, applied to the seasonal precipitation variables, 1981 2012. Light gray shading indicates those values significant at the 0.1 levels and dark gray those at 0.05. Hypergeometric Test Simple plots of seasonal precipitation (figures 5 and 6) show changes following 1995, which coincides with a change from an AMO cool phase to a warm phase.
Figure 5. Time series of summer season precipitation totals in the seven climate divisions 1981 2012 and the seasonal value of the AMO.
Figure 6. Time series of wint er season precipitation totals in the seven climate divisions 1981 2012 and the seasonal value of the AMO. Application of the hypergeometric test produced significant results in all divisions except division 7 (the Keys), and majority of these values appear in the winter season (figure 7). During the AMO cool phase from 1981 to 1995, a substantial number of winter seasons experienced above long term median rainfall While in the warm phase from 2000 to 2012 several divisions showed a significantly higher number of winter seasons during which precip itation totals that were below the median. Although there were t wo significant values during summer periods the AMO did not show any discrete trends with the rainfall
Figure 7. Results of the applicat ion of the hypergeometric test to winter and summer seven climatological divisions. Despite the well established link between winter precipitation in particular and ENSO, The hypergeometric test reveal ed no significant associations with ENSO. The small numbers of years
classified within the warm and cold phase groups during the study period greatly reduced the chances of detecting any such associations. Discussion and Conclusion This study set out to determine whether seasonal patterns of change in precipitation, as wide figures, could be the causal mechanism for the observed f the satellite based NDVI. In general, the results strongly and consistently refute the hypothesis that increased winter rainfalls, arising from climate change and low frequency variability is the cause of increased greenness of winter. The fitting of li near trends to the thirty year time series of seasonal rainfall totals indicates increases in summer rainfall totals and decreases in winter totals across the state. Even in the western panhandle where both summer and winter totals appear to have declined the percentage of annual precipitation falling in the summer has increased. Although no conclusive inferences could be drawn about the influence of ENSO, it is quite clear that changes in seasonal precipitation totals are coincident with the change in t he phase of the AMO. The research provides a means to analyze possible relationships between observed land cover cover will continue to change. It is necess ary to study not only trends in rainfall but also the impact of low frequency oscillations like the AMO and ENSO to manage and assess land cover and economic and environmental value. Forecasts of future climate predict that the Southeastern US will experience an increase in winter precipitation (Misra et al 2011), however there appear to be few signs of such changes in seasonal precipitation totals over the pa st 30 years within the state. climate variability and the uniqueness of its climate, particularly in its peninsular portions Several studies have concl uded that parts of Florida, especially the peninsula, exhibit inverse effects to the AMO (Kelly et al. 2008 ). Climate variabi l ity has shown to greatly influence water resources of Florida, which also emphasizes the need to accurately discern its impact to the highly populated state (Enfield et al. 2001 ).
Using data grouped into divisions that share homog enous climate conditions enabled a detailed analysis for the entire state of Florida (Tsai et al. under review ) Simple l inear r egression performed using seasonal rainfall totals show that in most areas of Florida winter rainfall has actually decreased since 1981. This implies that precipitation is not the cause of increased green vegetation in Florida, as seen in the satellite images. Plots of seasonal rainfall illustrate a decrease following the year 1995, which corresponds to a change in the AMO. There is a consistent decrease in the number of months that experienced median winter rainfall totals since the year 2000. Although the AMO and p recipitation data do not give an explanation to the greening vegetation, this analysis provides direction to further research regarding the land cover change in Florida.
Works Cited Burt, James, Gerald Barber, and David Rigby. Elementary Statistics fo r Geographers Third edition New York: Guilford Press, 2009. 275 276. Print. Chase, T. N., Pielke Sr, R. A., Kittel, T. G. F., Nemani, R. R., & Running, S. W. (2000). Simulated impacts of historical land cover changes on global climate in northern winter. Climate Dynamics, 16 (2), 93 105. doi: 10.1007/s003820050007 Cleland, E. B., Chuine, I., Menzel, A., Mooney, H., & Schwartz, M. (2007). Shifting plant phenology in response to global change Trends in Ecology and Evolution 22(7). Retrieved from https://www.usanpn.org/files/ShiftPlantPhenoGlobalChange_2007.pdf (CARA) Consotorium for Atlantic Regional Assessm ent, (2006). How does climate change impact land use ? Retrieved from website: http://www.cara.psu.edu/land/lu primer/luprimer11.asp Douglas, A. V., & Englehart, P. J. (1981). On a statistical relationship between autumn rainfall in the central equatorial Pacific and subsequen t winter precipitation in Florida. Monthly Weather Review, 109(11), 2377 2382. Enfield, D., Mestas Nunez, A., & Trimble, P. (2001). The A tlantic M ul tidecadal O scillation and its relation to rainfall and river flows in the continental U.S. Geophysical Research Letters 28 (10), 2077 2080. Fig 3. Oregon Spatial Climate Analysis Service. Florida Average Annual Precipitation [Web Map]. Retrieved from http://www.priweb.org/ed/earthcomm/environment /cc/florida/florida_avg_annual_precip.html Grimm, A. M., Barros, V. R., & Doyle, M. E. (2000). Climate variability in southern South America associated with El Nio and La Nia events. Journal of Climate 13 (1), 35 58. Johns, E., & Lee, T. National Oceanic and Atmospheric Association, (n.d.). Weather and climate stro ngly influence salinity,water quality and circulations of south florida coastal waters and bay Retrieved from website: http://www.aoml.noaa.gov/outreach/floridaseagrant/ pdf_files/TropicalConnections_WeatherAndClimateInfleunceCircuation_JohnsLee.pdf Jord an, C. L. (1984). Water R esources Atlas of Florida : Florida State University. Kahya, E., & Dracup, J. A. (1993). US streamflow patterns in relation to the El Nio/Southern Oscillation. Water Resources Research, 29(8), 2491 2503. Karl, T. R., & Knight, R. W. (1998). Secular trends of precipitation amount, frequency, and intensity in the United States. Bulletin of the American Meteorological S ociety, 79 (2), 231 241.
Kelly, M. H. and Gore, J. A. (2008), Florida river flow patterns and th e Atlantic multidecadal oscillation. River Res earch Appli cations 24: 598 616. doi: 10.1002/rra.1139. Retrieved from website: http://onlinelibrary.wiley.com/doi/10.1002/rra.1139/c itedby Koch Rose, M., Mitsova Boneva, D., & Root, T. State University System of Florida, (2011). Florida water management and adaptation in the face of climate change Retrieved from website: http://floridaclimate.org/docs/water_managment.pdf Lambin, E. F., Turner, B. L., Geist, H. J., Agbola, S. B., Angelsen, A., Bruce, J. W., . Folke, C. (2001). The causes of land use and land cover change: moving beyond the myths. Global E nvironmental C hange, 11 (4), 261 269. Mason, S. J., & Goddard, L. (2001). Probabilistic precipitation anomalies associated with ENSO. Bulletin of the American Meteorological Society, 82(4), 619 638. Mira, V. (2011). Climate scenarios: a F lorida centric view (Master's thesis) Retrieved from http://floridaclimate.org/docs/climate_scenario.pdf Ropelewski, C. F., & Halpert, M. S. (1987). Global and regional scale precipitation patterns associated with the El Nio/Southern Oscillati on. Monthly weather review, 115(8), 1606 1626 Southwest Florida Water Management Distr ic t, (2004). Florida river flow patterns and the multidecadal oscillation Retrieved from website: http://www.swfwmd.state.fl.us/documents/reports/riverflow_patterns.pdf Tsai, H., Southworth, J., & Waylen, P. ( under review ). Spatial persistence and temporal patterns in vegetation cover across F lorida, 1982 2006 Physical Geography United States Environmental Protection Agency. (2013). Climate impacts in the southeast Retrieved from website: http://www.epa.gov/climatechange/impacts adaptation/southeast.html Waylen, P., Keellings, D., & Qiu, Y. (2012). Climate and health in Florida: Changes in risks of annual maximum temperatures in the second half of the twenti eth century. Applied Geography, 33, 73 81.