Title: Characterization and spatial variability of a phosphate minesoil in Central Florida /
Full Citation
Permanent Link: http://ufdc.ufl.edu/UF00099337/00001
 Material Information
Title: Characterization and spatial variability of a phosphate minesoil in Central Florida /
Physical Description: xiii, 177 leaves : ill. ; 28 cm.
Language: English
Creator: Gensheimer, Gregory J., 1956-
Publication Date: 1985
Copyright Date: 1985
Subject: Soils -- Florida -- Polk County   ( lcsh )
Reclamation of land -- Florida -- Polk County   ( lcsh )
Phosphate mines and mining -- Florida -- Polk County   ( lcsh )
Soil Science thesis Ph. D
Dissertations, Academic -- Soil Science -- UF
Genre: bibliography   ( marcgt )
non-fiction   ( marcgt )
Thesis: Thesis (Ph. D.)--University of Florida, 1985.
Bibliography: Includes bibliographical references (leaves 170-176).
Statement of Responsibility: by Gregory J. Gensheimer.
General Note: Typescript.
General Note: Vita.
 Record Information
Bibliographic ID: UF00099337
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: alephbibnum - 000505086
oclc - 22751886
notis - ACS5240


This item has the following downloads:

characterization00gens ( PDF )

Full Text








I wish to express my appreciation to my committee for

their guidance in the research and preparation of this

dissertation, especially Dr. Randy Brown who did not know

what he was getting into when he met me. He held up quite

well even after being dragged around all areas of phosphate

mine reclamation research from Mulberry to Cape Canaveral,

Florida, and from radon gas evolution, through soil

potentials, to geostatistics and spatial variability.

Special thanks must also go to two non-committee

members. Thanks are extended to Jacob (Jaap) Bos for

helping me, in the heat of the summer, to survey some

minesoil types and to set out my sampling plan. He helped

keep the sampling ordeal tolerable by being able to joke and

smile even after a day of 980, humid heat. Thanks also go

to Ron Jessup, for inadvertently teaching me geostatistics

and helping me to "flesh out" my dissertation. His

questions were always the same: "What are you trying to do

today?" "Why?" or "What does that mean?" and "What happens

if you do . ?" My answer was always the same too; a

shrug of my shoulders and an "I dunno." Then I'd go find


AGRICO, Inc., graciously provided access to literally

thousands of acres of mined lands so that I might choose a

"good" site. Mr. Dale Carson, Reclamation Director, and Mr.

Don Morrow, General Manager Mining Operations, gave time

from their busy days to show and describe different types of

mine lands.

Thanks must go to the Soil Science Department for

providing the opportunities and of course the money that

kept me from running around on the coal mines of Hazard and

Harlan, Kentucky, "totin" my .557 or what could have been

much worse, becoming a--"Bulldog." The Soil

Characterization Program in particular provided financial

assistance, travel expenses and the necessary laboratory

equipment, space, and materials. I was able to meet people

from all over the world: Nepal, Southeast Asia, China,

Europe, several African and Latin American countries,

Australia, California, and even North Miami Beach. I was

also able to meet a midwestern lady, Jan Allison, my best

friend over the last 5 years.

Finally, but certainly not the least, thanks must go to

my family for providing the loving and learning environment

I grew up in. When I think back, I am probably here today

because I had the W.B. Encyclopedias to read whenever I had

the urge.



ACKNOWLEDGMENTS ......................................... ii

LIST OF TABLES.............................................. vi

LIST OF FIGURES............................................ vii

ABSTRACT .................................................xi


LITERATURE REVIEW........................................ .. 4

Geology............................................... .. 4
Phosphate Mining......................................10
Soil Classification and Variability...................20
Geostatistics......................................... 24
Semi-variograms.................................... 27
Kriging.......................................... 52

MATERIALS AND METHODS..................................... 40

Site Selection........................................ 40
Laboratory Analysis....................................51

RESULTS AND DISCUSSION................................... 6

Morphological Characteristics of the Minesoils........65
Specific Morphological Characteristics................66
Minesoil Characteristics.............................69
Geostatistics .........................................80
Contour Mapping.................................101
Comparison of Kriged and Non-Kriged Contour
Maps........................................... 130
Applicability of Standard Soil Analytical
Procedures for Minesoil............................. 137








BIOGRAPHICAL SKETCH...................................... 177


Table Page

1 Statistical moments of surface soil (25 cm)

2 Means of subsurface soil parameters by
position and as combined means...................75

5 Means of surface soil parameters by rows
in the medium grid............................... 78

4 Semi-variogram equation types and variables
for each soil parameter using all sample
points ...........................................81

5 Goodness-of-fit values for each soil
characteristic subset............................98

6 Correlations of Kriged versus measured
values for 11 points............................ 125


Figure Page

1 Counties of the Land Pebble Phosphate
District of Florida and study site
locations (A and B) in southwest Polk County......5

2 Most commonly used theoretical
semi-variograms .............................. .....29

3 Schematic diagram of typical semi-variogram
curves; Situation A exhibits nesting,
Situation B pure nugget..........................51

4 Schematic of profile locations at the
Payne Creek area site............................42

5 Schematic showing locations of three
nested grids covering the field..................44

6 Soil survey of the study site (inside
dashed lines) and surrounding area from
the 1927 Polk County Soil Survey (Soil
Survey Staff, 1927).............................. 46

7 Photograph of sampling tube (7.5 cm in diameter)
and sample depicting the variability of color
and texture in undisturbed samples immediately
after being removed from sampling tube...........50

8 Schematic of three nested grids showing
where lines were drawn separating 204 values
into east, west, and south populations for
goodness-of-fit tests............................59

9 Examples of hypothetical layer sequences in
profiles A and B of phosphate minesoils...........65

10 Photograph of profile 2, Payne Creek area,
showing layering in the lower half and lack
of layering in upper half due to effects of
earthmoving after mining..........................67

11 Plan view of study site prior to reclamation
and associated sample point locations............70

12 Non-Kriged contour map of depth (cm) to
sand tailings on the medium grid (350 = Row A,
300 = Row F) ..................................... 72

13 Cation exchange capacity direction-independent
semi-variogram calculated from 126 values
(medium grid) (modified from Bos et al., 1934)
and from all 204 values ..........................84

14 Elevation direction-independent
semi-variogram calculated from 126 values
(medium grid) (modified from Bos et al., 1984)
and from all 155 values ..........................85

15 KCl-pH direction-independent semi-variogram
calculated from 126 values (medium grid)
(modified from Bos et al., 1984) and from all
234 values........................................89

16 H20-pH direction-independent
semi-variogram calculated from 126 values
(medium grid) and from all 204 values.............90

17 Extractable Mg direction-independent, northwest
to southeast, and northeast to southwest
semi-variograms calculated from all 204 values...94

18 Extractable acidity direction-independent,
northwest to southeast, and northeast to
southwest semi-variograms calculated from all
204 values........................................95

19 Extractable K direction-dependent semi-
variograms calculated from all 204 values........99

20 Organic C direction-dependent semi-variograms
calculated from all 204 values..................100

21 Kriged organic C, large grid (increment
is 3.035%)......................................104

22 Kriged extractable acidity, large grid
(increment is 0.60 meq/100 g)...................105

23 Kriged H20-pH, large grid (increment is

24 Kriged extractable Mg, large grid
(increment is 0.4 meq/100 g)....................107

25 Error of Kriged sand, large grid (increment
is 0.063%) ......................................109


26 Error of Kriged extractable K,
large grid (increment is .0007 meq/100 g).......110

27 Error of Kriged extractable Mg,
large grid (increment is 0.07 meq/100 g)........ 111

28 Error of Kriged organic C, large grid
(increment is .0015%)............................112

29 Error of Kriged extractable acidity, large
grid (increment is 0.075 meq/100 g)............. 115

30 Error of Kriged elevation, large grid
(increment is 0.25 ft).......................... 114

31 Error of Kriged sand, small map (increment
is 0.033%) .......................................115

32 Error of Kriged extractable K,
small map (increment is .00037 meq/100 g).......116

33 Error of Kriged extractable Mg,
small map (increment is 0.07 meq/100 g).........117

34 Error of Kriged organic C, small map
(increment is .0024%) ...........................118

35 Error of Kriged extractable acidity, small
map (increment is 0.07 meq/100 g)...............119

36 Error of Kriged elevation, small map
(increment is 0.078 ft)..........................120

37 Kriged sand, large grid (increment is

38 Kriged CEC, small map (increment is
0.5 meq/100 g) ..................................128

39 Kriged extractable Mg, small map
(increment is 0.4 meq/100 g)....................129

40 Kriged extractable K, small map
(increment is 0.005 meq/100 g)..................131

41 Kriged extractable acidity, small map
(increment is 0.4 meq/100 g)....................132

42 Kriged KC1-pH, small map (increment is 0.2).....153

43 Non-Kriged sand, large grid (increment is
3.5%).......................................... .. 135

44 Non-Kriged organic C, large grid
(increment is 0.105%)........................... 156

B-1 Sand direction-independent semi-variogram
calculated from all 204 points..................152

B-2 Organic C direction-independent
semi-variogram calculated from all 204 points...153

B-3 Extractable K direction-independent
semi-variogram calculated from all 204 points...154

C-1 Kriged KC1-pH, large grid (increment is 0.25)...156

C-2 Error of Kriged KCl-pH, large grid
(increment is 0.035)............................ 157

C-3 Error of Kriged KC1-pH, small map
(increment is 0.03)............................. 158

C-4 Kriged CEC, large grid (increment is
0.5 meq/100 g) ..................................159

C-5 Error of Kriged CEC, large grid (increment
is 0.06 meq/100 g) .............................. 160

C-6 Error of Kriged CEC, small map (increment
is 0 .05) ........................................161

C-7 Kriged elevation, large grid (increment is
1.25 ft) ........................................162

C-8 Kriged elevation, small map (increment is
0.3 ft) ...................................... 163

C-9 Error of Kriged H 0-pH, large grid
(increment is 0.045) .............................164

C-13 Kriged H20-pH, small map (increment is 0.2).....165

C-11 Error of Kriged HO9-pH, small map
(increment is 0.05)............................. 166

C-12 Kriged extractable K, large grid
(increment is 0.007 meq/100 g)..................167

C-13 Kriged organic C, small map (increment
is 0.015%) ......................................168

C-14 Kriged sand, small map (increment is 0.75%).....169

Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy




May, 1985

Chairman: V.W. Carlisle
Major Department: Soil Science

As of 1978, 20% of Polk County, Florida, was owned by

phosphate mining companies. Seventy percent of that land

was either being mined or had been mined. Mine lands exist

predominantly in agricultural areas, so characterization of

minesoils will be vital in planning for future agricultural

uses. The present soil characterization system is based

primarily on types of minesoil "parent materials," including

overburden (spoil), sand tailings, clay, or mixes of these


Initial reconnaissance of spoil indicated that conven-

tional soil characterization techniques might not be ade-

quate to assess differences in spoil types. Therefore, this

study was designed to assess both minesoil characteristics

and spatial variability of the spoil. Samples were taken

from two grids and two transects, all nested. Points were

spaced 50 x 200 m, 10 x 10 m and 1 m apart.


Average values of the surface 25 cm were H20-pH 6.6,

KCl-pH 5.6, sand 85%, organic carbon (OC) 0.5%, cation ex-

change capacity (CEC) 10 meq/100 g, extractable K 3.096 meq/

100 g, extractable Mg 2.1 meq/10 g, and extractable acidity

4.7 meq/100 g. All values except extractable K appeared to

depend on position with respect to spoil islands. Average

subsurface values were H20-pH 5.9, KC1-pH 5.6, sand 84%, OC

0.36%, CEC 9.8 meq/100 g, extractable K 0.05 meq/100 g,

extractable Mg 2.4 meq/100 g, and extractable acidity 3.8

meq/100 g.

Direction-dependent and direction-independent semi-

variograms were calculated for each parameter. Except for

elevation, all exhibited a nugget variance and sill. The

nuggets of sand, CEC, OC, extractable acidity, and K were

greater than 40% of the respective sill values. Organic

carbon had the shortest range (90 m) and extractable K

exhibited the longest (400 m).

The combination of long ranges, large nuggets, and

placement of sampling grids contributed to imprecision of

parameter contour mapping over the field. Kriged and non-

Kriged maps of sand and OC were compared. Both types of

maps were relatively inaccurate; neither map was superior to

the other. Increasing the density of Kriged points from the

same number of measured points does not result in more

accurate contour mapping. It is recommended that a sampling

scheme have relatively equal numbers of points spaced at

varying distances apart, spread evenly over the study site

to maximize understanding of spatial variability.



The soils of Polk County, Florida, were originally

surveyed in the mid 1920's with the county report published

in 1927. The county is currently being resurveyed because

of changes that have taken place in the last 50 or so

years. Most significant is the increase of knowledge about

soils leading to changes in soil series. Some of the old

series have since been recorrelated and renamed.

Another significant change is the amount of mine land

that exists today. As of 1978, 20% of Polk Co. was owned by

phosphate mining companies. Fourteen percent of Polk Co.

was either being mined or already had been mined (USEPA,

1978b). The majority of the remainder of the county will

probably be mined in the near future.

Since the mine land of Polk Co. is located

predominantly in an agricultural area, characterization of

the minesoils will be vital in planning for future

agricultural uses. The mining companies are already using

the land for pasture, citrus, and short rotation pulpwood

plantations. Additional land could be put back into either

agricultural or forest use as more land is reclaimed and

sold back to private citizens.

Currently, the U.S.D.A. Soil Conservation Service (SCS)

is mapping minesoils as part of the Polk County Soil

Survey. The SCS system is based primarily on the types of

"parent materials," including overburden (OB), sand tailings

(ST), clay, or mixes of these materials in which the

minesoils are developing. Two examples of minesoil map

units in Polk County are Arents, which encompass the OB

soils, and Psamments, which are the ST soils (personal

communication, Richard Ford, Soil Survey Party Leader, USDA

Soil Conservation Service, Bartow, FL, May, 1982).

Since the Psamments' variability is limited to

differences in particle size and the quantity and mineralogy

of phosphate and trace minerals, the mapping of this unit

should be adequate. Overburden, on the other hand, is much

more variable since it is a mixture of all soil and

substratum materials above the mineral ore. Presence or

absence of such materials as dolomite, leach zone material

(upper part of the Bone Valley Formation leached of most of

the Ca-phosphates), clay layers, argillic horizons, spodic

horizons, or even organic matter from swamps, bayheads, or

cypress domes will affect the minesoil characteristics.

Soil maps and legends that do not reflect the possible

differences in minesoils developed on OB will not be very

valuable to managers of that land type. After many decades

of mining, the OB mine land type is very extensive in Polk

County. As the acres of land reclaimed to this type are

increasing, the importance of a complete classification

scheme also increases.

One objective of this study was to characterize some

common chemical and physical properties of the minesoils

developing in OB. The OB in an OB-capped ST setting was

chosen since this setting is one of the most common types of

reclaimed mine land being built today (J.D. Carson, Director

of Reclamation, AGRICO, Mulberry, FL, May, 1982). To relate

the properties of minesoils to native soils, standard

procedures were to be used. Another major objective was to

study spatial variability of soil parameters so that future

data would be collected in a manner that would be more

representative of the minesoil parameters in the field.



The Land Pebble Phosphate District includes seven

Florida counties: Hillsborough, Polk, Hardee, Manatee,

Sarasota, DeSoto, and Charlotte (Fig. 1). The Hawthorn

group of geologic formations contains the principal

phosphate bearing formations in this area. These formations

dip gently to the south and southeast.

Miocene sediments in Florida are characterized by

authigenic dolomite and phosphorite mixed with a flood of

terrigenous sediments which transgressed across the

peninsula as the Gulf Trough ceased to exist. The Gulf

Trough separated the continental landmass and its

terrigenous sediments from the Eocene-Oligocene limestone

banks. The Hawthorn group is defined as the phosphatic

portion of the Miocene-Pliocene sediments and has been

divided into three lithologic formations: (1) Miocene

Arcadia, (ii) Miocene Noralyn, and (iii) Pliocene Bone

Valley Formation. The open marine Arcadia Formation

consists dominantly of dolomite mixed with primary

allochemical phosphorite and subordinate amounts of

terrigenous sediments. The Arcadia Formation contains the

vast phosphorite resources of the future. The Noralyn


0 50 km

Fig. 1. Counties of the Land Pebble Phosphate District of
Florida and study site locations (A and B) in
southwest Polk County.

Formation consists of shallow-water, coastal-marine,

terrigenous sands and clays mixed with primary orthochemical

and allochemical phosphorite. The bulk of phosphorite is

presently being mined from the Noralyn Formation. The Bone

Valley Formation is a thin and local unit of fluvial,

estuarine, and coastal-marine sediments composed of terri-

genous sands and clays, abundant shell material, and

reworked lithochemical phosphorite. Only locally does the

Bone Valley Formation contribute a major portion of the P

being mined (Riggs, 1979a).

Riggs (1979b) claimed that the formation of lithologic

units with such significant amounts of P is considered

abnormal. In addition, he noted that the mechanism for

phosphorite formation is probably much more complicated than

the simple chemical precipitation of apatites from a

nutrient-rich, upwelling, current system. He noted that, in

order for the phosphorite to accumulate, a shallow-water,

coastal environment is essential during the time of primary

phosphorite sedimentation. There must be a series of

structural arches or highs providing a shoaling environment

and adjacent basins and embayments. An appropriate topo-

graphy must exist both to produce the phosphorite and to

allow for its accumulation.

The reasons for phosphorite formation are not generally

agreed upon. Riggs (1979b) reviewed the literature and

brought forth two plausible reasons. He suggested that the

associations of phosphate with dolomite and the Mg-rich

minerals such as montmorillonite, palygorskite, and

sepiolite, and with high levels of fluorine and other trace

minerals, were caused by volcanic ashfalls of significant

duration. The ash would have supplied the Mg and trace

elements and the P would have resulted from the decaying

marine organisms poisoned by the ash.

Riggs (1979b) then introduced a more probable reason,

saying that phosphorite formation appears to occur during

periods of changing tectonism, leading to the supercharging

(supersaturation) of the regional chemical systems of bottom

waters. The cold, supercharged, almost toxic bottom waters

upwell into shallow shelf environments and across the

structural highs. The P accumulates as precipitates of

loose, colloidal, microcrystalline mud in suspension that

Riggs calls microsphorite plus biologically produced teeth,

bones, shells, and mollusk kidney stones.

The orthoohemical microsphorite mud plus other

sediments, along with dolomite and biological particles,

responds to local energy conditions and biological processes

within the environment. Small animals ingest the muds,

sediments, dolomites, and other micro-particles and excrete

them in pellet form. In low energy environments, the muds

settle, indurate, and break up producing intraclasts. Under

higher energy situations, some of the mud is physically

aggregated, producing oolites or pseudo-oolites. The

phosphate gravels, sands, and clays are transported along

and off of the shoals, diluted by sediments and chemicals

(limestone and/or dolomite), and deposited in basins where

they accumulate.

A typical stratigraphic section, from a phosphate

miner's viewpoint and from the surface down, consists of

topsoil or surface soil, residual sand mantle called the

overburden (OB), leach zone, matrix zone, bedclay zone, and

bedrock zone (USEPA, 1978c).

The residual sand mantle varies from about 1.5 to 9.1

meters in thickness due to differential primary

sedimentation, weathering, and reworking of materials. Sand

and altered clays kaolinitee altered from montmorillonite),

predominate, although montmorillonite, apatite, and

vivianite are also present (Altschuler et al., 1964).

Altschuler et al. (1964) stated that the nature of the

OB may be the most striking geomorphic and stratigraphic

consequence of the supergene groundwater alteration. Voids

in the sand are created by volume losses due to clay

transformation and/or leaching, loss of the swelling

property (montmorillonite), and clay translocation. The

result of such clay movement is the presence of cutans

lining the floors of cavities, coating fractures, and

forming minor clay hardpans throughout the weathered zone.

the leach zone is part of the phosphorite which has

been modified by weathering processes, varying from 0.3 to

3.5 m in thickness. Characteristically, it is vesicular,

and may be friable or indurated. Quartz sand is cemented

and indurated by the secondary minerals wavellite,

crandallite, and millisite. Discontinuous hardpans located

between the leach zone and the matrix are composed of the

various movable materials leached from above by acid ground

waters and precipitated in the upper matrix zone due to the

effect of ground water neutralized by basic calcium

phosphates (USEPA, 1978c).

The matrix zone, ranging from 0 to 15.2 m in thickness,

consists of an unconsolidated mix of phosphate pellets and

granules (concentrate) and cobbles and boulders (pebbles) of

phosphatized limestone, quartz, sand, silt, and clay (USEPA,

1973c). Various compositions of carbonate substituted

apatite exist between the end members, fluorapatite, and

hydroxyapatite. Apatites exist along with montmorillonite,

quartz, chert, and calcite. Although many cations commonly

substitute in the apatites, the most important one but not

most abundant is uranium. It occurs in concentrations of

about 0.01% (McKelvey, 1955).

A bedclay zone, the uppermost part of the Arcadia

formation, is derived from the movement of clays out of the

OB, leach zone, and matrix. It underlies the matrix

discontinuously. It is a water-saturated, plastic sandy

clay, grading into dolomite below. Usually, it has a higher

uranium content than the matrix and is sometimes mined if

the P205 content is economically significant. Common

minerals include attapulgite, montmorillonite, quartz, and

dolomite (Altschuler et al., 1964).

Below the bedclay zone is the bedrock zone, which

comprises the presently noneconomic parts of the Arcadia

Formation. It is composed of a fine-grained sandy and marly

dolomite, and dolomitic sand and marls, all of which are

sparsely phosphatic (USEPA, 1978c).

Phosphate Mining

Large scale surface mining of phosphate in the Land

Pebble Phosphate District started in 1888. The value of the

rock mined in 1892 had already surpassed $1 million (Wang et

al., 1974). Production of phosphate rock in Florida peaked

in 1933 at 43.0 million metric tons (Florida Phosphate

Council, 1985), declined to 42.8 million metric tons in 1981

and to 30.0 million metric tons in 1952. This decline was

due to a combination of the following factors: high

interest rates, low crop prices, adequate soil P levels,

increased foreign competition, and strength of the dollar

(Stowasser, 1931).

In the U.S., the majority of phosphate is used for

agricultural purposes, mainly as ordinary and triple

superphosphate. Other uses include leavening agents, water

softening and cleansing agents, plasticizers, insecticides,

military smoke screens and incendiary bombs, fluorescent

lights, television tubes, animal feed supplements,

beverages, ceramics, catalyst and oil refining agents,

photography, and dental and silicate cements (Ruhlman,


The phosphate mining industry is economically important

to Florida. Each job in the phosphate industry generates

6.2 other jobs. For each dollar of income the phosphate

industry generates, 3.4 dollars of other income are

generated. And for each dollar of increased phosphate

industry activity, 5.8 dollars are generated in other

economic activity (USEPA, 1978a).

Another economically important fact is that the

phosphate companies own large amounts of land. Phosphate

companies own 2255 km2 or 14.1% of the seven county Land

Pebble Phosphate District. Of this land, 20% is currently

being surface mined or quarried, or exists as pibs. The

rest of the land exists as herbaceous range land, crop or

pasture land, orchards and groves, forested wetlands, and

evergreen forest land. The most common land use is as

herbaceous range land (USEPA, 1978b).

The most common use of land not owned by phosphate

companies is range land 31.3%, followed by crop and pasture

land 22.2%, wetlands 12.5%, and urban or built up land

9.3%. DeSoto and Hardee Counties have mostly agricultural

land. Sarasota and Charlotte Counties have significant

amounts of urban or built up land. Manatee County is built

up along the coast but underdeveloped elsewhere.

Hillsborough County is one half urban and one half

agricultural. Polk County has the most diverse land use.

The northwestern corner, north of Interstate 4, is primarily

agricultural, the land between Interstate 4 and State Route

oJ mostly urban or built up, and the land south of State

Route 50 is split between agricultural and mine use (USEPA,

1978b) (Fig. 1).

Phosphate mining in the area, while disruptive in the

short run, is not expected to affect land use drastically in

the long run. Although the area occupied by mining

activities is expected to rise 15.5% between 1975 and 2055,

it is expected that most of the land will be returned to

premining uses (i.e., mainly agriculture or range). A 7%

decline is expected in agricultural land and another 7%

decline in range land. These anticipated declines are not

attributable to mining but to the 44% expected increase in

urban or built upland (USEPA, 1978e).

Surface mining for phosphate began in the District when

the deposits were first found. As draglines became more

efficient and more rugged, ore could be extracted from

beneath deeper and deeper OB.

Currently, electric-powered, walking draglines with

bucket capacities of 27 to 38 m3 and boom lengths of 69 to

84 m are used to mine phosphate. The land is cleared of

trees and brush, swamps are drained and muck or peat is

removed in preparation for the start of mining. The

dragline "walks" adjacent and parallel to the mining face in

increments of about 23 to 38 m, moving the OB and extracting

the ore. This procedure is repeated for a distance of 0.1

to 1.6 km before the dragline moves away from the mining

face and starts its walk back, creating a new mining face as

it goes (USEPA, 1978d).

The matrix is dumped by the dragline into a pit,

slurried, and hydraulically pumped to the beneficiation

plant. Beneficiation encompasses the processes involved in

separating economically valuable phosphate from the inert

sand and the clay slimes which are part of the matrix. The

processes include washing, milling (hammermill), screening,

clarifying, separating, and floating.

In the first step, the coarse phosphate rock is

separated from clay, sand, and fine phosphate via washing,

screening, and milling. The product is rock phosphate

having particle diameters between 1.5 and 19 mm. Material

smaller than 1.5 mm is washed again, removing waste clay

slimes consisting of particles smaller than 0.1 mm. The

remaining material is now ready for the flotation

processes. One process is for materials greater than 0.5 mm

in size and the other process is for materials less than 0.5

mm. During these processes, sand tailings (ST) and more

clay slimes are removed. Since the whole sequence of

processes involves the use of large quantities of water, the

waste products are pumped to disposal areas as they are

removed from the ore (USEPA, 1978d).

The 03 (sometimes called spoil after mining) is a mix

of topsoil, subsoil, and all substrata including leach zone

materials, above the matrix. The OB remains at the mining

site, piled in long parallel rows 46 to 76 m apart. This

pattern results from the geometry of the dragline's boom

length and the arc through which the dragline swings. After

mining is complete, groundwater pumping stops and the water

table rises, causing long lakes to form between spoil rows.

Reclamation of these mined landscapes can be

accomplished in several ways. The simplest method is simply

to grade the spoil rows, forming shorelines around the

finger-like lakes. Other methods involve filling of the

valleys between the spoil rows with beneficiation waste

products. Sand or a mix of sand and clay can be used as

fill, raising the average level of the land. The spoil

"islands" that remain are flattened and spread out over the

other materials. The mining can also be planned so that the

spoil rows act as dikes and the whole area may be turned

into a clay settling pond. Once the clays have dried

sufficiently, the area can be capped with sand or OB to

improve stability, though not to the point where

construction is feasible.

Revegetation of mined lands to forage species has been

the primary practice in the District because forage types

can be chosen to match the different spoil characteristics

and because quick growing grasses hinder erosion. Reforest-

ation and row cropping of these areas have been minimal.

Although Wright (1930) states that citrus has done poorly on

mine sites due to the infertile and drougthy nature of the

spoils, the real cause of poor growth may be lack of proper

management (personal communication, Don Morrow, General

Manager, Mining Operations, AGRICO, Mulberry, Fl, May,

1932). After several reclamation failures, AGRICO now

realizes that they must first become acquainted with and

then follow prescribed management procedures for any new

reclamation endeavor such as citrus.

Detailed information on the characteristics of the

three types of materials resulting from mining and

beneficiation of phosphate rock is limited. General

information is common. Sand tailings have low water holding

capacity, little organic matter, and minimal fertility

(Mislevy and Blue, 1981; Hawkins, 1985). Properties of ST

resemble properties of soils commonly underlying sand pine-

scrub plant communities (EcoImpact, 1931).

Dried phosphatic clay materials range in clay content

from 27 to 85%. Although they have excellent nutrient and

moisture retention properties, they are prone to water-

logging and thus very difficult to cultivate (Hawkins,


Hawkins (1985) stated that spoil typically has better

natural fertility and better moisture holding capacity than

native soils due to slightly higher contents of clay, Ca,

Mg, and K. He noted that because of the higher clay

content, particularly when organic matter was absent, the

spoils became very hard and cloddy when dry and very plastic

and slick when wet. Phosphatic clay soils have excellent

nutrient and moisture retention properties but are also very

difficult to cultivate and prone to waterlogging. Sand

tailing soils, on the other hand, have very low nutrient and

moisture retention properties but are easy to cultivate.

Hawkins based his descriptions of the three types of

minesoils on particle size determination, pH, Ca, Mg, P, and

K analyses for 12 samples from reclaimed mine lands 5 to 50

years old.

The land areas of spoil, ST, and clay from phosphate

mining activities are of sufficient size to be routinely

delineated by the Soil Conservation Service (SCS). The SCS

is currently conducting a soil survey in Polk County. Soil

map units being used for the minesoils include three

materials alone and in combination. Spoils are called

"Arents" in the Polk County soil survey legend; ST soils are

called "Psamments"; clay soils are called "Slickens" if

still watery and "Slickens, dewatered," if dry. An example

of a map unit name given to a combination of materials is

"Arents, clayey substratum," which describes spoil capping

dried clay on an old clay settling pond (personal

communication, Richard Ford, Soil Survey Party Leader, USDA

Soil Conservation Service, Bartow, FL, May, 1932).

Some of the insights gained in coal mine reclamation

research can be useful in interpreting data trends in

phosphate mine research. Coal mine reclamation research was

started earlier than phosphate mine research, possibly

because the problems are much more visible in and around a

coal surface mine on the side of a hill than in a phosphate

mine in relatively flat terrain. Acid mine drainage

devastated entire watersheds, turning all the stream beds a

bright orange color while highwalls and lines of

deforestation ringed the mountains. Recent research has

moved away from looking at revegetation problems and has

instead emphasized the chemical, physical and micro-

biological properties of the minesoils, applicable

analytical procedures, variability of minesoil properties,

and the comparison of minesoil properties with native soil


Ciolkosz et al. (1933) studied 25 minesoil pedons

located predominantly in Western Pennsylvania. They found

the minesoils to be deep and well drained, and to have

subsoil rock fragment contents of >70%. Surface or near

surface horizons had from 40-60% rock fragments and

predominantly medium textures (loam, silt loam, clay

loam). High rock fragment content made these soils drought

and reduced their effective cation exchange capacity (CEC)

to a low level. Three-fourths of the minesoils had very low

to low pH values. Many had salt contents at a level which

restricted plant growth.

Plass and Capp (1974) found that in general coal mine

spoil was deficient in nutrients, had an unfavorable

moisture regime, was acid, and contained excessive salts or

toxic substances. Pedersen et al. (1983) found that

minesoils normally held less water at comparable tensions

than native soil.

Berg (1978) noted that some sampling procedures and

soil tests have rather severe limitations when applied to

minesoils, especially because some users may be unaware of

those limitations or give little consideration to them.

Berg reviewed sampling, sample preparation, pH, lime

requirement, soluble salts, Na adsorption ratio, N, P, and

trace element analysis.

A more rigorous, but slightly different approach to

sampling was undertaken in coal mine areas by Sobek et al.

(1978). They emphasized pre-mining OB analysis in addition

to sampling of minesoils. They explored what they

considered to be all the important field and laboratory

methods in a step-by-step fashion. They recommended

characterization of OB rock strata before mining begins, so

that mining and reclamation could be planned to segregate

the different materials and selectively place the neutral,

easily weatherable 08 materials near the rooting zone and

bury the potentially toxic (acid) materials beneath the

rooting zone.

Apparently, the large quantities of minesoil

characteristic data plus the increasing acreage of mine

lands influenced the creation of a minesoil classification

scheme. With 121,500 ha of highly disturbed land in West

Virginia, the National Cooperative Soil Survey recognized a

need for more meaningful mapping units. They provisionally

approved a system for classifying minesoils developed in

West Virginia (J.C. Sencindiver, 1975. Ph.D. dissertation,

West Virginia University, Morgantown, WV, as cited by Smith

and Sobek, 1978).

The minesoil mapping units were based on rock types,

coarse fragments (presence of splintery edges and disorder

in direction of long axis), texture, pH at 25 cm in the

profile, and dominant profile mineralogy. Smith and Sobek

also set standards for minesoil suitability classes: (i)

suitable for multiple uses; (ii) suitable for rural, urban,

recreational, or industrial building sites and grounds;

(iii) suitable for extensive recreation and access (hiking,

camping, hunting, etc.); (iv) suitable for production of

forest products; (v) suitable for pasture, hay, or other

crops not requiring plowing; and (vi) suitable for

intensive agriculture.

Although Ciolkosz et al. (1935) characterized 25

minesoils, they did not use the mapping units developed by

Sencindiver. Instead, they used the term Minesoil and

modified it with texture classes or classes based on the

amount of coarse fragments.

Indorante and Jansen (1981), working on soil surface

mines in the Midwest, stated that variation of some native

soil properties such as bulk density, pH, organic carbon,

cation exchange capacity, and particle size distribution

were similar to the variation of the properties in a

minesoil. Schafer (1979) studied native and minesoils in

Montana, finding that native soils were less variable than

minesoils on a local scale (O to 10 m spacing) but much more

variable on a landscape scale (greater than 500 m

spacing). Unlike minesoils, variation of natural soils was

highly correlated with landscape features and soil forming

processes at the larger scale.

Soil Classification and Variability

Soil systematics, whether natural or artificial, is

largely a matter of matching like specimens, profiles, or

sites, and distinguishing among unlike ones (Webster,

1975). Webster said that the main concern of the users of a

classification system is that variance or a similar measure

of diversity is, on the average, smaller among members of a

class than among members of different classes.

The general purpose classification of soils is based on

easily observable and, therefore, mainly morphological

attributes (Webster and Butler, 1976). Perhaps the most

widespread use of soil classification schemes is for soil

survey and mapping. The soil surveyor's classes are usually

defined for a few distinguishable properties that are also

manageable by the users in the hope that variation in other

properties is similarly restricted (Webster and Butler,

1976). In order to choose appropriate soil characteristics

to use in the classification scheme, surveyors must consider

the needs of the users. These needs may include information

such as pH, available nutrients, and soil-water character-

istics for agronomic interests, and shear strength, plastic

limits, and expansion coefficients for engineering

interests. It is, however, impossible to satisfy the needs

of all users because groupings of some particular soil

characteristics are incompatible with the classification

scheme. That is, variability of some soil attributes is

larger within classes than between classes.

Even if the classification scheme is adequately based

on the properties manageable by interested users, it is

still useless if the graphical representation of soils

information does not represent accurately the character-

istics of the soils in the field. The soil survey map is

the graphical representation with corresponding descriptions

of observations and measurements from pits and boreholes.

Even though the observations and measurements are point

samples, it is hoped that mapped soil classes have values

similar to the observations and measurements but different

from those of other classes (Burgess and Webster, 1980).

Burgess and Webster noted that the distribution of

properties is displayed by assigning the typical value of

that property within its class to individual parcels on the

map. The majority of values are assigned to areas that have

not been sampled, so these values instead must be

predicted. The predicted values may be different from the

actual values.

The validity of soil survey maps depends on the

accuracy and precision with which the samples represent the

characteristics of soils in the field. Studies of soil

variability show that accuracy and precision of soil mapping

are not perfect (Reynolds, 1975; Campbell, 1977, 1978;

Mausbach et al., 1980; Lanyon and Hall, 1981; Russo and

Bresler, 1931a, 1931b; Sisson and Wierenga, 1981; Edmonds et

al., 1982; Lascano and Bavel, 1982; Cassel, 1983).

Proper soil sampling techniques are paramount to

determining variability of soil properties. Cline (1944)

stated that sampling error was commonly much greater than

analytical error. He said that the accuracy of chemical

analysis in defining field soil characteristics depends on

the degree to which (i) the gross sample accurately

represents the soil from which it was taken, (ii) no

changes affecting the results occur in the sample prior to

analysis, (iii) the subsample analyzed accurately

represents the gross sample, and (iv) the analysis

determines the true value of the character under test.

Precision is increased if several small samples from the

whole are used rather than one larger one. Also, since a

sample represents values over some limited area, precision

is maintained if the sampling scheme stays within the areal

limits (Cline, 1944).

Reynolds (Ph.D. Dissertation, University of Bristol, as

cited by Reynolds, 1975) studied variability of soils along

slopes and found that increasing the number of samples to

reach a desired level of precision is not always feasible.

To estimate the mean of pH, soil depth, soil-water, and

organic matter populations with 1% precision would require

as many as 689 samples. If a precision of 0.2% were chosen,

as many as 17,227 samples would have to be taken.

Considering the number of samples Reynolds found necessary

for precise soil sampling, it is not surprising that most

often the number of samples taken to characterize a soil

series is found to be inadequate (Ike and Clutter, 1968).

Reynolds' findings were based on his idea of how soil

variability should or could be studied. More variable soils

required more samples to reach a given precision than less

variable soils. This type of research has limited value in

solving problems associated with validity of soil maps. In

studying variability of forest soils, Ike and Clutter (1968)

found that even though estimates of population mean values

were accurate, the sample provided little information as to

the nature and pattern of variability of the properties


Instead of doing research that would help to reduce

variability of soil property measurements, workers have

begun to study the variability itself. Edmonds et al.

(1982) studied short-range and long-range variability in

soils. They found that extreme short-range variability

within delineations of natural landscape elements or pedons

caused long-range variations to seem erratic. Mausbach et

al. (1930) studied the variability of measured properties in

morphologically matched pedons and found that a transect

sampling scheme is practical in assessing trends in areal

distribution of a property. Campbell (1977, 1978) studied

variability of soil properties across soil boundaries. He

defined three variability models: (i) completely random

values with no coherent pattern to their distribution,

(ii) gradual variation of measurements without distinct

boundaries (i.e., the contact between two soil types can be

represented by a trend surface), and (iii) distinct,

uniform soil regions separated by abrupt boundaries such as

are implied by soil maps. He also suggested that other

workers' problems with sampling studies were caused by

failure to consider sample spacing or arrangement.


In the 1970's, soil scientists began using

geostatistics to study the variability associated with

spacing and geometry of sample points. Geostatistics, or

statistical methods applied to geological ore reserve

estimation, evolved with the need for increased precision of

estimating quality or quantity of ore. In an ore reserve

evaluation, homogeneous regions are determined and values

are assigned on the basis of judgement and fact (Popoff,

1966). The facts are determined from exploratory data, spot

sampling, production data, or data from other parts of the

same deposit.

In addition to geostatistics, some of the methods used

to predict areal values from point sample values include

(i) method of isolines, (ii) method of triangles, (iii)

method of polygons, and (iv) distance weighting.

Rutledge (1976) listed three fundamental objections to

these conventional methods:

(i) The procedures for assigning values (to "chunks"

of ore body) are arbitrary and without a sound

theoretical basis; the methods are a function of

geometry of samples rather than a function of the

quality of the ore.

(ii) The procedures can be biased and there is no way

of ensuring against bias.

(iii) Estimation procedures either do not include a

method of determining the precision of the

estimate, or allow precision to be determined


In addition to the above, two more objections are given to

the use of distance weighting methods. First, Rutledge

(1976) said that using arbitrary distance functions "merely

formalizes the mystical principle of gradual change" (1976,

p. 300). Second, this method assumes that samples are

random, which may not be the case. Criticism of these

procedures has also been put forth by Delfiner and Delhomme

(1975), Clark (1979a), Royle (1979), and Burgess and Webster


Royle (1979) tempered his criticism by admitting that

sometimes the above methods are accurate estimators. He

noted that variance of a population is made up of both a

random component and a spatial component. The methods work

well when the random component is small due to the smaller

chance of choosing a sample whose value is far from the

mean. As the random component increases relative to the

spatial component, the methods become less and less

useful. Like Royle, Delfiner and Delnomme (1975) stated

that the procedures were optimum only in limited

situations. Their main objection was that the procedures

could not be used to filter the random factor from the

sample interrelationships. The estimated surface would

still pass through the sample points even if highly erratic,

giving an unrealistic map.

Krige (1976) stated that when he first tried to

estimate gold ore reserves, he found it impossible using

common statistical methods. He said that without

mathematical statistics or geostatistics it was impossible

to find an underlying logical pattern. He studied the

pattern of distribution of 500,000 values from 24 gold

mines, including gold, uranium, and pyrite contents. His

approach was not based on any theories but on practical

observations and experimentation.

In writing about the history of geostatistics, Krige

(1976) stated that Matheron is responsible for the

development of "Regionalized Variable Theory" (RVT) and the

use of semi-variograms. Royle defined regionalized

variables as, "those variables whose values are related in

some way to their positions" (1979, p. 92). Huijbregts

(1975) noted that, even though the variables seemed

unpredictable and highly erratic, their behavior is not

completely random. He said that neighboring points are

related by a complex set of correlations which he called the

"structure." In order to analyze properly any spatial

phenomena, one ". . must be able to extract from the

apparent disorder of available data the major structural

characteristics of the phenomena and a measure of the

correlation between values at neighboring points throughout

space" (Huijbregts, 1975, p. 38).


Huijbregts (1975) described a variogram, 2Y(h), as the

average quadratic deviation between values, (Y), at two

points, x and x+h, of space:

2Y(h) = E([Y(x+h) Y(x)]2) (1)

Commonly, Y(h), the half- or semi-variogram, is used. The

interval, h, or the lag, has direction and therefore is a


In order to use Eq. (1), it is assumed that the

distribution of the differences in values between the pairs

of points is the same over the area of interest (i.e.,

"quasi-stationarity" exists) (Clark, 1979a). In other

words, Y(h) does not depend on the magnitude of the x values

or on the domain (area) where Y(h) is estimated.

The semi-variogram is the minimum structural tool and

the minimum statistical tool needed to make a structural

analysis. Another way of wording Equation 1 is that y(h) is

the variance of the error made when estimating Y(x+h) by

Y(x), and thus reflects the ability of Y(h) to solve any

estimation problem. Since the function Y(h) has properties

closely associated with the structural features of the

population, it can be used to quantify those features. The

basis of structural analysis is the study of the behavior of

the semi-variogram with respect to values and direction of

the vector (h) (Huijbregts, 1975).

The equation used in calculating a semi-variogram is

1 N(h) 2
y*(h) = I [Y(xi+h) Y(xi)] (2)

where y*(h) is the calculated semi-variogram,

N(h) is the number of pairs of points used in the

Y's represent the measured values in space
separated by a distance (scalar) along the
vector h.

As the histogram is to the population distribution function,

[y*(h)] is to [y(h)]. The estimate [y*(h)] of the true

semi-variogram generally increases as h increases, and it is

the pattern of this increase that is used in determination

of structure.

In practice, points are plotted on a graph and a

mathematical model is fit to them. Five models are most

commonly used (Gambolati and Volpi, 1979): (i) spherical,

(ii) exponential, and (iii) Gaussian, for curves that tend

to level out (have a sill) with increasing h, and (iv)

linear, parabolic, or root, and (v) logarithmic, for curves

that do not have a sill (Fig. 2). In these models, c is the

"nugget," the Y intercept (except for the logarithmic model



I i

a 3a

y(h)=0 h=O
y(h)=c[1.5h/a-O.5(h/a) ]+C 0 Y(h)=w + C b>a

c + C -----------
.950 + C ----------

-(h) I


h 3a
y(h)=0 2/a h=0
Y(h)= (l-e-(h/a) )+C h>0

a=1.5/ a=l




y(h)=O h/
Y(h)=o(l-e -a)+C


Y(h)= aLn(h)+C

Y(h) )a +C

h>O and 0

Fig. 2. Most commonly used theoretical semi-variograms.

.95w + C



where C is the intercept of X=1); a is the range, the value

along h where the semi-variogram levels out; the sill is the

Y(h) value where the curve levels off; and w is the

difference between the sill and nugget values. Not every

mathematical model can be used because, by theory, the

models must be conditionally positive definite (always

giving a positive value).

A few characteristics of the semi-variogram give

immediate information to the user (Fig. 3). Theoretically,

the Y intercept would always be zero. The value at one

location should equal the value of another sample taken zero

distance away. When the Y intercept is greater than zero,

the semi-variogram has a "nugget," indicating that at least

a part of the variability is random or that the sampling

scheme was too coarse to eliminate all of the positional

variability. If the intercept is such that the semi-

variogram is flat, then the sample positions were too far

apart to see any correlation, and variability is completely

random (Fig. 3, Situation A).

The value (a) along h, where the semi-variogram levels

out, is called the range or zone of influence. Pairs of

points closer together than this distance are in some way

correlated. Members of pairs farther apart are independent

from each other. The value of Y(h) at which the semi-

variogram levels out is called the sill. This value

approximates the variance of the population.


ai n




Fig. 5. Schematic diagram of typical semi-variogram
curves; Situation A exhibits nesting, Situation B
pure nugget.

If the calculated semi-variogram increases at least as

rapidly as (h)2 for large distances of h (parabolic shape),

then a trend is indicated. This condition may also be

expressed as regional drift (Journel and Huijbregts,

1978). Local drift is exhibited by a periodic rise and fall

in the semi-variogram (Clark, 1979b).

If the semi-variogram starts at cl along Y (Fig. 5,

situation a) then two ranges exist, al and a, and the semi-

variogram is said to exhibit nesting (Huijbregts, 1975). A

semi-variogram is additive for all ranges and sills that may

be present. If a subordinate range and sill are significant

then they may show up in the curve as extra inflection


Another structural feature can be seen by viewing

several semi-variograms from the same population. Because h

from the semi-variogram equation is a vector, it is

important to calculate semi-variograms in different

directions. If the resulting semi-variograms have different

ranges or sills, the material (soil) is anisotropic. That

is, the spatial correlation among points changes with

direction. The direction-independent semi-variogram might

exhibit "nesting" or inflections when its component semi-

variograms are different from each other.


An estimation procedure, called Kriging, is used to

estimate values optimally at unsampled locations. The

general equation is

Zo = 1z(x1,y1) + A2z(x2,2) + .. + Anz(x ,y) (3)

where Z = the linear sum or weighted average of
parameter Z at location (xoYo),

Z(xn,yn) = measured values of parameter Z at
locations (xn,yn),

n = coefficients or weights associated with
the data points.

This is the same type of equation used for other distance-

weighting estimators except that, in Kriging, the weights

are chosen so that error associated with the estimate is

less than that for any other linear sum. The A's depend on

the known spatial dependence as expressed by the semi-

variogram and the geometric relationships among the observed

points (distances between measured points and the point to

be estimated). Kriging is an unbiased estimator, meaning

that the average error of its estimates is zero. Whenever

the sample population is normally distributed, Kriging is

the Best (estimation variance is a minimum) Linear (uses

linear combinations of equations based on neighbor point

values) Unbiased (the average error of its estimates is

zero) Estimator (BLUE). Rendu (1978) stated that in

practice, the assumption of a normal distribution of the

sample values is not often satisfied except when the sample

mineralization (ore body) has a relatively high grade or

when the value considered has a low variability. He further

stated that judgement and past experience would then be used

to decide whether the assumption of normality could be

used. This latter statement results in much discomfort for

beginning geostatisticians.

Additional discomfort is generated by the number of

Kriging variants or alternatives that exist to overcome the

effects of non-normality or non-stationarity. Some of the

variants will be discussed below.

Punctual Kriging is simply estimation of point values

resulting in the most accurate isarithmic map that can be

made using a normally distributed set of point data.

Sometimes however, local discontinuity can obscure long

range trends in the maps. Block Kriging, the Kriging of

areas rather than points can be used to overcome this

problem resulting in smaller estimation variances and a

smoother map (Burgess and Webster, 1980b). Universal

Kriging was designed to overcome the effects of non-

stationarity. In this variant, the trend is mathematically

removed from the data before calculating the semi-

variograms. The requested number of points are Kriged and

then the trend is added back into the Kriged points.

Sometimes universal Kriging is also capable of transforming

a non-normal population into a normal population during

trend removal.

Two Kriging variants, lognormal and disjunctive

Kriging, were designed to overcome the problems due to non-

normal data. Lognormal Kriging, as the name implies, is

used when data are positively skewed. The semi-variograms

are calculated on the logtransformed data. Disjunctive

Kriging is used when any other best-fit transform is used to

convert data into a form that approximates a univariate

normal distribution. The semi-variograms are calculated on

the transformed data (Henley, 1981).

Henley (1981) found fault with the Kriging variants

that attempt to overcome problems of non-normality or non-

stationarity. He found that most authors of geostatistics

suggested that departures from stationarity were not of

practical significance since local stationarity was often

assumed. Henley stated that no general proof or statistical

test existed to determine whether such an assumption was


Henley (1981) also objected to the use of lognormal and

disjunctive Kriging. The variable to be estimated (after

transformation) is a non-linear function of the original

data. This function may, in fact, be very complicated.

Henley stated that these methods produced a sub-optimal,

non-linear, biased estimator.

Krige developed the practical use of geostatistics for

gold reserve estimation (Krige, 1976). Since then his

system has been used widely for the study of many types of

ore deposits. Journal and Huijbregts (1978) used 24

different types of deposits as examples in their book

alone. Outside of mining interests, Kriging has been used

to study hydraulic head of aquifers under the Venetian

lagoon (Gambolati and Volpi, 1979), meteorology,

pluviometry, topography (Journel and Huijbregts, 1978), and

also to study hydrogen ion concentration in bulk

precipitation (Bilonick, 1985).

In soil science, Kriging and structure determination

have been used to study spatial variability of various

chemical and physical parameters. Campbell (1978) studied

the variability of sand content and pH on adjacent mapping

units. Two grids, 200 x 80 m in size, were positioned as

close to each other as possible without crossing the map

unit boundary. Samples were taken at 10 m intervals. Mean

pH was the same for both mapping units, while sand content

was 8.5% on one (Pawnee) and 1.2% on the other

(Ladysmith). The 10 m intervals were not sufficiently close

to show any spatial correlation for pH (i.e., a pure nugget

effect was observed). Sand contents were correlated to 50 m

for the Ladysmith, but the grid on Pawnee was not large

enough for accurate estimation of sand (i.e., no range was


Vauclin et al. (1983) studied variability of sand

content and available water content. They used a 10 m

spacing with a grid measuring 70 x 40 m. They found that

available water content values were correlated up to a 40 m

range and that sand contents were correlated to 33.5 m.

Byers and Stevens (1985) studied the spatial

variability of hydraulic conductivity (K) and particle size

of an untilled fluvial sand. They used two 14.85 m

horizontal transects perpendicular to each other and one

vertical transect 4.9 m long. They used a sample spacing of

15 cm in the horizontal plane and f cm in the vertical plane

so that the variability of the small bedding structure would

not be lost. Particle size values were correlated to only

83 cm and In K values were correlated to 53 cm.

The smaller range associated with particle size

reported by Byers and Stevens (1983), relative to that of

Campbell (1978) and Vauclin et al. (1983), may be a function

of the sample spacing. Gajem et al. (1981) used four

transects sampled at 20 cm intervals, four at 200 cm, and

one at 2300 cm, to study the spatial variability of water

contents, pH, exchangeable cations, surface area, mean

diameter, and bulk density. Except for two cases, the range

of each variable increased with increasing sample spacing.

Exchangeable cation values were correlated to 1.15 m as

determined by the transects with the 20 cm spacing. On the

transects with 200 and 2000 cm spacings, the range was 20

m. Bulk density values were correlated to 2 m as measured

on the 20 cm and 200 cm spacing transects. Gajem et al.

explained the increasing range values as a function of the

larger population variance encountered with larger sample


Yost at al. (1982a) used extremely large sample spacing

(1-2 km) along transects to study the variability of several

parameters including pH, exchangeable Ca, Mg, K and Na, and

the sum of the ions. With this large spacing, they found

very large ranges, 8 to 58 km. From the results, they

suggested that soils over large areas may be grouped in

order to obtain uniform regions of soil properties suitable

for management regimes.

In a subsequent paper, Yost at al. (1932b) used the

semi-variograms to study precision of contour mapping. They

found that the maps of error associated with Kriged values

could be used to indicate where additional samples should be

taken to provide the most information. They also found that

it may not be necessary to include the trends in Kriging

equations or to remove the trend in the data prior to

analysis. Russo and Brasler (19S2) found that adding

additional points to the field at locations where maximum

Kriging errors were calculated lowers the overall error more

than if the points were added randomly.

Burgess and Webster (1930) studied variability of Na

contents and thickness of cover loam (depth to sand and

gravel) in different fields. A square grid containing 440

observations, 15.2 m apart was used for the Na study and a

spacing of 23 m in another field was used to study thickness

of cover loam (depth to sand and gravel). The Na semi-

variogram increased linearly (no range) while the cover loam

semi-variogram had a range of 101 m.

Vieira at al. (1981) studied variability of infiltra-

tion rates in a cropped field using a 55 x 159 m grid.

Samples were spaced at 5 m intervals along X, skipping the

rows potentially affected by crop roots. Spacing along Y

was 1 m. They found infiltration measurements were corre-

lated to 35 m. They also found that the normal distribution

should not lead to the conclusion that locations for field

observations should be selected randomly. The presence or

lack of spatial structure of a set of observations has no

bearing on the frequency distribution of that same set. The

two properties exist independently of each other.

Seostatistics is in its adolescence as a tool soil

scientists use to attempt to understand variability of soil

characteristics. Geostatistics has been used to study the

variability of a number of soil chemical and physical

properties, in fields and across regions. As the number of

studies of different soil properties increases, soil

scientists will be better able to compare their results with

others. Jury (1984) reviewed eight studies and found a

correlation between lag spacing and range. Jury called this

condition a measurement bias. Other measurement biases may

be found as the experience of the soil geostatisticians


One of the important aspects of soil survey is knowing

where to draw boundaries separating unlike soils. Soil

surveyors may not be able to use geostatistics on a site by

site basis but they may be able to use geostatistics to

identify the spatial behavior of soils in areas of

transition between soil types. Understanding the spatial

structure of soils in these kinds of areas may alleviate

some problems in positioning boundaries.


Site Selection

Several "overburden-capped sand tailings" sites were

chosen from AGRICO's mined-land inventory as possible study

sites, with the help of J.D. Carson (Director of Reclamation

AGRICO, Mulberry, FL, May, 1932). Carson stated that this

land type is significant because of the large acreage either

already reclaimed or yet to be reclaimed in this manner.

Characterization data for phosphate minesoils were not

available, so several reconnaissance surveys were made to

gain a preliminary understanding of soil properties

occurring in this land type.

The first profile examined was of overburden-capped

sand tailings, reclaimed to pasture. It was located in

southwest Polk County (Fig. 1, Location A), Fort Green area,

S20, T32S, R25E (Tallahassee Meridian). This profile was

described by noting distinct horizontal layering, presence

of lenses, and colors, including mottles.

Profiles of spoil that had not been remixed after being

deposited initially by the dragline were examined next. The

profiles were in an erosion gully that cut through the

spoil. This site (Fig. 1, Location B), in the Payne Creek

area, located 13 Km east of the previous site, in S15,

T32S, R21E, had been minimally reclaimed to pasture in a

land-and-lakes topography. The tops of the spoil piles were

leveled, forming long (approx. 500 m), narrow (approx. 50-

150 m) strips of level land between similarly sized and

shaped lakes.

Profiles 1 and 2, located nearer to the head of the

gully (Fig. 4), exhibited multiple layers of various

thicknesses dipping towards the adjacent lake. Profile 5,

located closer to the lake in a tributary gully, was

positioned perpendicular to 1 and 2 and was much more like

the profile at the first site.

The layers in profiles 1 and 2 were described at two

places in the profile (opposite sides of the 1 m wide face)

to account for the significant dip in the layers. Since

profile 5 did not exhibit such distinctive layers, it could

not be described the same way. There were no distinctive

features that could be measured by depth. The whole profile

was described as if it were one layer by noting presence and

field textures of lenses, texture of the soil matrix, and

colors, including mottles.

A nearby site reclaimed to pasture in 1938 was chosen

as the primary study site because it was accessible,

relatively level, and high enough above the surrounding

landscape to remain dry during the rainy season. It is

located in the next section south of the gully site, S22,

T32S, R24E (Fig. 1, Location B), in a field about 1 km east-

west and about 0.75 km north-south. The western boundary of





0 5m

Fig. 4. Schematic of profile locations at the Payne Creek
area site.

the study site was the western section line; the southern

boundary was an east-west road 380 m north of the southern

section line (Fig. 5). Topography was generally smooth,

with minor local undulations. The highest part of the field

was just northeast of the middle of the site. From here,

tne land sloped slightly to the eastern edge to a swampy

swale and slightly to the west. To the south, it sloped

sightly, then more steeply (5% slope) leaving about 150 m of

level land adjacent to the road. A small creek drained the

swale running through this low section and under the road.

The steepest part of the south-facing slope paralleled the

road from the swale on the eastern side to an area about 430

m to the west, where it then curved north. From this point,

it became less steep and separated into several terraces

(rough berms about 1 m high installed for erosion control)

spaced about 50 m apart which continued curving towards the

northwest corner.

Original revegetation plans consisted of planting

Bahiagrass (Pasoalum notatum) and Bermudagrass (Cynodon

dactylon). Hairy Indigo (Indigofera hirsuta) and

Aeschynomene (Aeschynomene americana) have naturally

invaded, forming a few very dense patches interspersed in

the grass.

The dark, organic-rich layers apparent in these mine-

soils came from the native soil profiles or wetland sedi-

ments. On the other hand, the clay and sandy clay layers

probably came largely from the clay layers above the matrix

e- J

z -4

m f4
++4++~LI44r -1+ 1

1+I + + $ + + ) > Kra

iI 0

/ I ~ J
a a

a i

%a g a


] 0)

and to a lesser degree from the non-extensive, loamy

argillic horizons of the native soils.

Native soils on this site were all poorly drained.

Another characteristic of all the soils except one (Felda)

was the presence of an organic-rich horizon in the form of a

spodic horizon or a mollic, umbric, or histic epipedon. The

soils on the site were first described (Fig. 6) (Soil Survey

Staff, 1927) as Leon fine sand (comprising 50% of the area),

Leon fine sand, loamy phase (5%), Portsmouth fine sand

(20%), Portsmouth fine sand, swamp phase (10%), St. Johns

fine sand (10%), and Blanton fine sand (5%) (Fig. 6). Since

the original soil survey, some of the soils have been re-

correlated and re-named. The St. Johns fine sand has

retained its name and is in the sandy, siliceous,

hyperthermic, Typic Haplaquods family (Soil Survey Staff,

1975). The Leon fine sand is now called Myakka fine sand, a

sandy, siliceous, hyperthermic, Aeric Haplaquod. The

Portsmouth fine sand, swamp phase is now called Felda, a

loamy, siliceous, hyperthermic, Arenic Ochraqualf. The

Blanton fine sand is now called Tavares fine sand, a

hyperthermic, uncoated Typic Quartzipsamment. The

Portsmouth fine sand was split into two series, Floridana

and Placid, depending in part on the texture. The Floridana

contains enough clay to be loamy and is classified as a

loamy, siliceous, hyparthermic, Arenic Argiaquoll. Placid

fine sand, with less clay is a sandy, siliceous,

hyperthermic, Typic Humaquept. The Leon fine sand, loamy



0 200 m

Fig. 6. Soil survey of the study site (inside dashed
lines) and surrounding area from the 1927 Polk
County Soil Survey (Soil Survey Staff, 1927).

phase, was split into four series, Wauchula, Wabasso,

Pomona, and Eau Gallie, depending on depth to spodic

horizon, depth to argillic horizon, and base status.

Wabasso fine sand, a sandy, siliceous, hyperthermic Alfic

Haplaquod, and Wauchula, a sandy, siliceous, hyperthermic

Ultic Haplaquod, are loamy within 40 cm of the surface.

Pomona sand, a sandy siliceous, hyperthermic Ultic

Haplaquod, and Eau Gallie, a sandy, siliceous, hyperthermic

Alfic Haplaquod are not loamy within 40 cm of the surface

(personal communication, Richard Ford, Soil Survey Party

Leader, USDA, Soil Conservation Service, Bartow, FL,

January, 1934).


The sampling scheme (Fig. 5), consisting of three

nested regular grids, was designed for maximum use of

geostatistics. The first sampling grid set up was

rectangular, with its long axis parallel to the run of spoil

as noted on older aerial photographs. The intent was to

assess the variability of the population while minimizing

the potential problems that might occur in sampling

perpendicular to the spoil rows. Information from the first

grid was used in designing one grid within the first grid

with points spaced closer together, and another grid with

points spaced farther apart that would fill the whole field.

The mid-sized grid, 30 m north to south by 200 m east

to west, was situated in the northeast quadrant of the

field. It consisted of six east-west (E-W) rows of 21

points each. The rows were 10 m apart and the points in

each row were also 10 m apart (Fig. 5) (Bos at al., 1984).

The largest grid, 600 m north to south by 803 m east to

west, consisted of 5 north-south (N-S) columns of 13 points

each. The columns were 203 m apart and the points in the

columns were 50 m apart (Fig. 5).

The smallest grid, comprising two connected, perpendi-

cular 10 point transects, was situated on the western edge

of the mid-sized grid (Fig. 5). The northmost point of the

north-south transect was the westmost point of the east-west

transect. Points in these transects were 1 m apart.

Samples from the mid-size grid were removed using a

hydraulic coring tube or auger, depending on the amount of

coarse fragments encountered. The coring tube was 7.5 cm in

diameter, and the auger 12.5 cm. Sample cores were taken to

2 m unless (i) the presence of a shallow water table

inhibited further sampling or (ii) ST was found at

shallower depths. Two or three cores, if necessary to

obtain sufficient volume of sample, were taken from each

grid point and mixed together. The cores at each sampling

point were taken as close together as possible to limit

variability within each sample. The cores were split into

25 cm sections whenever distinct layers were less than 10 cm

thick. Layers less than 10 cm thick would not yield enough

sample for all of the chemical and physical analyses. The

25 cm thickness was adjusted a maximum of 10 cm up or down,

if thick homogeneous layers were present, to avoid mixing of

such homogeneous layers with each other. One sample of the

31 was taken whenever possible (Bos et al., 1934). Samples

from the larger and smaller grids were taken from the

surface only, using a 7.5 cm mud auger.

In using a sampling scheme based on set vertical

distances rather than horizon characteristics, it was known

that some potentially valuable information on variability

would be lost because of the mixing of unlike layers into

one sample. The original intent had been to use the

properties of different materials in a profile, whether the

materials were in one layer or many layers, as a criterion

for characterizing minesoils. To save some of that

information, estimates were made of volumetric quantities of

materials different from each other in color and/or texture

before drying and mixing of the sample. Abundance and sizes

of roots were also noted (Bos et al., 1934). In many cases,

the samples consisted of multiple thin (about 10 mm thick)

layers of strongly contrasting materials (Fig. 7), making

separation and analysis of each layer impossible. After

drying, the samples were pulverized with an electric

grinder, mixed, passed through a 2 mm sieve, and mixed


Specific overburden characterization data, whether pre-

or post-mining, were not available. It was determined,

therefore, that soil methods and procedures to be used

should be those commonly used on native soils in the area.

Procedures used by the Soil Characterization Laboratory,

University of Florida, were chosen for this study.

Laboratory Analysis

Samples from the medium-sized grid were analyzed

first. Bos et al. (1954) reported the procedures used on

the surface samples from the medium-sized grid. Chemical

analyses included H20-pH and KCl-pH (1:1 solution to soil),

and cation exchange capacity (CEC) calculated as the sum of

bases (Na, K, Ca, Mg) extractable by 1N NH40AC at pH 7 (Soil

Survey Staff, 1972, method 5B1) plus acidity extracted with

0.5N BaC12 TEA pH 8.2 (Soil Survey Staff, 1972, method 6H1 +

SH1a). After doing the geostatistical work on these data

(described here and by Bos et al., 1984), laboratory work


Cation exchange capacity, KCl-pH and H20-pH, as

described above, were completed on the rest of the surface

samples from the larger and smaller grids. All surface

samples from all grids were analyzed for percent organic

carbon by the acid chromate digestion method (Soil Survey

Staff, 1972, method SAI) and for particle size determination

by pipette slightly modified from the Soil Survey Staff

(1972). The modification was the nonremoval of organic

matter prior to analysis since the majority of samples

contained much less than 1% organic carbon. Finally,

subsurface samples from nine locations in the mid-sized grid

were analyzed using the same procedures as before.


Fig. 7. Photograph of sampling tube (7.5 cm in diameter)
and sample depicting the variability of color and
texture in undisturbed samples immediately after
being removed from sampling tube.

Samples of the yellow siltstone were also qualitatively

analyzed by X-ray diffraction. Pebble-size material was

ground with an agate mortar and pestle, a small amount of

the powder was sprinkled onto a glass slide, and several

drops of amyl acetate-collodion mixture were added to

enhance adherence.


Geostatistical analysis of the initial parameters

(medium grid) consisted of calculations of semi-variograms

and determination of structure (Bos et al., 1934). Two

programs in Applesoft BASIC for the Apple II computer

(written by R. Jessup, Soil Science Department, University

of Florida) were used: (i) to calculate the direction-

dependent and independent semi-variograms, and (ii) to fit

one of the models to the calculated semi-variograms. The

fitting was by visual estimate only. At this time, there

was no testing of goodness-of-fit and no Kriging.

The search for a suitable Kriging program took several

months. After many unsuccessful attempts at modifying

programs from "Mining Geostatistics" (Journel and

Huijbregts, 1978) to run at the North East Regional Data

Center (NERDC), it was determined that these programs would

not be immediately suitable for use. The next stop in the

search was to the KRIGE subroutine in the "Surface II

Graphics System" (Sampson, 1978). This particular

subroutine is not available at the NERDC. Another

commercial Kriging package could not be used due to

proprietary usage rights. The "Semi-Variogram Estimation

and Universal Kriging Program" (Skrivan and Karlinger, 1979)

was chosen because in addition to being "free of charge," it

was easily modified to run at the NERDC. This program was

used for geostatistical analysis and interpolation of 2000

points from the population of 204 values of each

characteristic. This program consists of four options

required for (i) calculating the semi-variogram, (ii)

determining drift in the population, (iii) optimizing the

semi-variogram, and (iv) Kriging, i.e., using the semi-

variogram to estimate values at unsampled locations.

Generally speaking, output from one option is either

directly input into another option or is used to modify

values input into another option. One intermediate step,

described later, was used outside of the program to enhance

the program's utility.

Before a valid semi-variogram can be calculated, the

drift, if present, must be removed. If the local population

mean varies geographically, it is said to have drift. Drift

can only be determined, however, if the semi-variogram has

already been calculated. This circular logic necessitates

trial and error, with simultaneous calculation of the semi-

variogram and its associated drift. The program makes use

of this relationship by allowing the calculation of the

semi-variogram on the residuals of the observations minus

the drift. This is an iterative process of running options

(i) (semi-variogram calculation) and (ii) (drift

determination), in sequence repeatedly. The general process

is described below.

Data, consisting of x, y, and z values (geographic

east-west and north-south and the associated parameter

value), were input to option (i), which calculated

direction-dependent and direction-independent semi-

variograms and estimated drift.

Once the semi-variograms were calculated, the next step

was to determine the structure of the semi-variograms.

Structure analysis consists of fitting an equation with

parameters A, u, and C to the calculated semi-variogram

curve. Ninety-five percent confidence intervals around the

population variances were computed for each population of

204 values (elevation had 155) to indicate which semi-

variograms exhibited pure nugget (Romano, 1977). This

interval has been identified on each semi-variogram

figure. In all cases, this interval excluded the nugget

indicating significant structure (lack of pure nugget). The

ability to fit the semi-variogram depends on the experience

of the user plus the flexibility of the semi-variogram

program. A program in Applesoft BASIC was used to fit one

of the models to the calculated semi-variograms.

The models were chosen based on shape alone. No

consideration of the physical significance of the different

model types was given to the fitting procedure. There is

little evidence to link models with particular distributions

in the soil sciences.

Testing goodness-of-fit (GOF) of semi-variograms is not

quite the same as testing GOF of other curves fit to data

points on a graph. Typically, an R-square value is used to

identify the spread of the graphed points about a curve.

The semi-variogram graphedd points), however, can best be

thought of as a guide to be used for the fitting. The

variables A, C, and a and equation types are, in fact,

juggled to fit the graph, but the GOF procedures actually

test the fit of the equation to the individual sample

values. There is no evidence to suggest that a fit of an

equation to the graph with the highest R-square will always

result in the best fit to the actual sample values. This is

the reason a visual fitting procedure was used to fit the

equations to the graphs rather than a computational

method. The GOF method used in the program is described


After calculating the semi-variogram, fitting the

proper curve, and determining that drift may exist, the next

step would have been to calculate the equation of the drift

(option ii). In fact, option (ii) was attempted but not

used for two reasons. The first reason is that, of the nine

populations, drift was only noticed in elevation. The

second reason was that the program documentation did not

indicate how to use the output to choose the proper drift

coefficients. It did indicate that trial and error was the

best method. Considering that each estimate would have to

be tested for goodness-of-fit, the costs involved outweighed

the benefit of getting a slightly better fit for the

elevation semi-variogram.

In this study, several more limitations of the Skrivan

and Karlinger program were noted but it is impossible to

determine the degree to which these limitations adversely

affected the analyses. The biggest single limitation was

expense. Of the four program options, option (iii)

(goodness-of-fit test) was the most costly to run. Option

(iii) was used to determine goodness-of-fit of the

calculated semi-variogram to the real semi-variogram. The

procedure used is commonly called jackknifingg." Each data

point was individually suppressed along with its associated

row and column. The program then computed an estimate of

that point using the semi-variogram model, drift

coefficients, and the remaining points. The program

calculated the Kriged average error (KAE) of the population,

which was the difference between the real vs estimated

values using the equation

1 n
KAE = (Z ) (4)

where n = number of points,

Zi = measured value,

Zi = Kriged value.

The program also calculated Kriged mean square error (KMSE),

using the equation

1n i

Finally, the program calculated the Kriged "reduced" mean

square error (KRMSE) using the equation


1 n
in i=

where ai = standard deviation (error) of the Kriged point,

K(o) -

m n-1
X. j..- (X.,Yj) +
i =1 ii(Xi=1

where K(o)



fi(X ,Y.)

= sill

= unknown weighting coefficient

= covariance based on semi-variance and

= variance of the measurement error

= unknown LaGrangian multiple

= drift.

The KAE should approach 0 and the KRMSE should approach

1 as the fit improves. The decision as to how close the

values should be to these ideal levels before being

considered "good" is a function of the needs of the user.

Each population of 204 points had to be split into

three approximately equal groups before entering into this

option. These new smaller populations, south, east, and

0. =

west, were developed so that each would contain some points

in the medium grid (Fig. 8). Only once were all 204 data

points entered together. Five hundred seconds of computer

time were not enough for the calculations to be completed.

The high cost of each run reduced the number of times

different equations or different parameters could be tested.

The problem of cost was most noticeable where a proper

equation cannot be found or where drift is present. Each

time drift or the equation parameters are changed, option

(iii) had to be run to determine goodness-of-fit. Nine

soil variables were of importance in this study. The

maximum number of runs per parameter was limited to five.

Equation parameters A, C, and w were changed modestly to

minimize KAE and to get KRMSE to approach 1 for each of the

south, east, and west populations.

Periodicity exhibited in the semi-variograms remained

unaccounted for because of another program limitation, the

lack of flexibility in choice of models. Skrivan and

Karlinger incorporated the most commonly used models in

their program but did not add a model that may have resulted

in a much better fit for the semi-variograms modeled here.

Journel and Huijbregts (1978) described a "hole effect

model" as one incorporating a sine or cosine function to

model semi-variograms that are demonstrably cyclic. This

model was not included as one of the choices in this



C. I

-o 0

Q )
Oc 0
+ + -+ 4-

) o



z Z

0 0


0 0





Another problem in the program was noted in the Kriging

option which led to undesirable adjustments in structure

determination. Although the program allowed direction-

independent and direction-dependent semi-variograms to be

computed, it did not allow for their combined use during

Kriging when anisotropy was found to be present. The

program accounts for the varying distances between the point

to be Kriged and each of its neighboring points through the

use of the assigned semi-variogram equation and para-

meters. The program does not have the capability, however,

to use direction-dependent semi-variogram equations and

parameters that would depend on the relative locations

(directions) of the Kriged points and their neighbors.

Because of this, only the direction-independent semi-

variograms were modeled for use in the Kriging option. With

much more time and effort, it may have been possible to

determine the structure of each direction-dependent semi-

variogram and arrange the locations of points to be Kriged

to maximize the use of the direction dependence.

Another area of inflexibility in the Kriging option was

the inability to determine an optimum neighborhood size. A

neighborhood is the area of influence around the point to be

Kriged. It is important to be able to pick the size of the

neighborhood around the point to be Kriged so that a

localized mean can be calculated rather than a general

mean. Since the whole population was used in the estimation

of each point, the population mean resulted when the

distance between the point to be Kriged and its neighbors

was greater than the range of the semi-variogram.

Once KAE and KRMSE values were acceptably close to 0

and 1, respectively, the equation and the variables, data,

drift coefficients (if needed), and x, y pairs of

coordinates were input to option (iv) which calculated the

Kriged z value for each x, y pair and its associated

variance. Option (iv) was run twice for each parameter,

once witn 2030 xy pairs covering the whole field and again

with 500 xy pairs covering a small part of the field

including the northwest part of the medium-sized grid.

These z values and associated errors (standard deviations)

were then input to the SURFACE II mapping program to produce

isoline maps of the different values and the associated


All populations were tested for normality (Univariate

procedure) (SAS, 1982) to identify whether the Kriged

estimates would be the best linear unbiased estimators.

Transforming non-normal populations to normal was not

attempted for two reasons. First, Rendu (1978) stated that

the assumption of normality was not often satisfied and that

the decision of whether a population can be assumed to be

normal is based on past experience and judgement. Since

this study was the first attempt at using geostatistics on

these data, there were no answers to the question, "How

close is close (to normal)?" Secondly, Henley (1981) stated

that Kriging transformed data resulted in sub-optimal,


non-linear, biased estimators. The decision to use the non-

normal populations was based on a compromise of accepting

the sub-optimal estimates from the Kriging of non-normal

data versus finding a program to handle disjunctive Kriging

and still risk sub-optimal results.


Morphological Characteristics of the Minesoils

Differences existed between minesoils and native

soils. Horizontal layers or horizons as exhibited by native

soil profiles often were less common in minesoils than in

native soils. The exception was noted on older minesoils

(10-15 years) when a darker surface layer indicated the

beginning of A horizon formation (Appendix A, profile 1,

Payne Creek Site). Typically, the layers in a minesoil

profile were uneven in thickness, had abrupt boundaries, and

contained discontinuous inclusions of material quite

different from the matrix in texture and/or color. The

chemical, physical, and morphological characteristics of the

minesoil result from the OB (native soil and substrata)

properties, the sequence in which the OB materials were

excavated and dumped, the way in which the OB fell out of

the dragline bucket and came to rest, and the effects of

subsequent earth moving during reclamation.

As the dragline operator digs through the OB, he moves

the material to the previously mined cut, building long,

parallel rows of spoil piles that are uneven in height. The

morphology of the minesoil, particularly the layering, is

largely a result of the dumping action.

If the minesoil profile is located in the middle of the

pile, the layers, may be horizontal (Fig. 9, Situation A),

or they may dip diagonally across the profile (Fig. 9,

situation B). If the profile intersects an edge of the

spoil pile, perpendicular to the right edge of (B),

horizontal layering will be noted. If the profile face were

stripped away, through (B) from left to right, the layers

would seem to be closer to the surface.

Another morphological form commonly seen, the inclu-

sions of discontinous bodies within layers, seemed to result

from material rolling out of the bucket as it tipped and

then continuing down the sides of the spoil piles. In

profile, these bodies were circular, and in three dimensions

they were more or less spherical. In all the profiles

examined, these spherical inclusions were finer in texture

than the surrounding matrix or contained enough organic

matter to be darker than 3.5 in both value and chroma when

moist. The higher clay content and/or organic matter con-

tent apparently afforded sufficient cohesion to these bodies

so that they did not break up as they fell from the bucket.

The sequence of layers in a profile depends on the

sequence in which the OB was dug, moved, and dumped. Except

for this layering and occasional incipient A horizon

formation, there seemed to be no vertical trend in minesoil

properties. The presence of any particular layer in one

part of the horizon seemed not to have a genetic relation-

snip to any other layer. Adjoining layers frequently were




0 0



0 vl


quite different from each other. For example, a sandy clay

layer might be found with a layer of sand above and an

organic-rich layer below. The boundaries between the layers

were abrupt and often very convoluted due to mixing during

digging and dumping.

Specific Morphological Characteristics

Profiles 1 and 2 were somewhat alike but different from

both profile 3 and the profile observed at the Fort Green

site (Appendix A). The latter two profiles were somewhat

similar. The difference involved the presence or absence of

distinct layers. Profiles 1 and 2 exhibited layering, and

the others did not.

Profile 3 and the Fort Green profile indicated a more

complete mixing of the spoil. The Fort Green profile was

located in an area where spoil had been spread over ST.

Likewise, profile 3 was located near the sloping edge of the

landmass, and represented material that had been pushed off

of the center of the spoil piles during leveling proce-

dures. Apparently, earthmoving operations subsequent to the

initial mining caused a more complete mixing of spoil

materials and resulted in the destruction of distinct

layering. Spherical inclusions, however, were as common in

the reworked material as in the top of profile 2. It was

not known whether the spheres remained intact through the

reworking or if new spheres were created as the more

cohesive layers were once again moved. Profile 2 (Fig. 10)

Fig. 10. Photograph of profile 2, Payne Creek area, showing
layering in the lower half and lack of layering in
upper half due to effects of earthmoving after

showed evidence of layering in the lower part; the upper

part appeared to have been reworked during reclamation.

Although the gross morphology of the minesoil depends

largely on mining and reclamation techniques, many physical

and chemical minesoil properties depend also on the physical

and chemical properties of the OB (native soil and

substrata). Pre-mining 03 characteristic data for this

study site were not available. Data generalized for the

whole area were available but were used only with caution.

The pre-mining OB consisted generally of a thick sand

mantle over a layer of clay (Altschuler et al., 1964). The

OB also included the noneconomic part of the matrix (i.e.,

that part called the leach zone). Presence or absence of

this leach zone material and of dolomite was a major

determinant of the character of the spoil (post-mining

OB). If substantial amounts of dolomite occurred in the

spoil, then the minesoil pH was near neutral. If dolomite

was absent, while significant amounts of leach zone

materials were present, vegetative productivity may be

hindered due to low pH and/or aluminum toxicity. A site 19

Km east of the study area did not have dolomite in the spoil

but did have enough leach zone material to induce aluminum

toxicity symptoms in an Eucalyptus (Eucalyptus grandis)

plantation (personal communication, C.W. Comer, School of

Forest Resources and Conservation, IFAS, Gainesville, FL,

November, 1933).

Because of the great diversity of colors, textures, and

morphology, initial characterization of minesoils was

difficult. It was obvious in the initial reconnaissance

that characterization could not be too rigorous. The

problems involved the presence, absence, amounts, locations,

or combinations of thin, distinct layers and the profile-

spoil pile relationships already described.

Measuring layer thickness and distance from the surface

did not yield useful information because presence, amounts,

and locations of different materials in the profile appeared

random. The descriptions of relative quantities of

different materials per core sample retained, at least

qualitatively, the variability lost in sample preparation

(mixing, grinding, etc.). Field notes indicated the

morphological differences between samples taken from a spoil

island and samples taken from spoil on top of tailings.

Comparison of field notes with laboratory observations and

results showed that presence of weathered yellow siltstone

was associated with a textural change from loamy sand to

loam, and increased H20-pH and CEC. Field-lab comparisons

also indicated that even though the majority of a sample was

sandy in texture, the presence of clay lenses and spherical

bodies of finer textured materials would cause texture of

prepared samples to be finer than sand.

Minesoil Characteristics

A map of the sampling scheme was superimposed on an

aerial photo (Fig. 11). This aerial photo was too old







HS rI)
C, 0

+ co


Z; 0
o1 IL
w -'

C- C

H a




(1979) to show the relationships of spoil and ST as they

existed before final reclamation. The photo showed the site

during ST filling, with the process only partially

complete. The figure did show roughly, however, the

relationship between sampling points and spoil islands. It

was evident that the long rows of spoil of uneven height

were already partially covered with ST, forming spoil

islands. It was impossible to identify precisely all the

spoil islands or determine how much of the spoil remained

above the ST at the end of the filling stage.

Based on the measurements of depth to spoil-capped ST

taken in the medium-sized grid, it appears that the spoil

ridge underlying the eastern 2/3 of the medium grid largely

was buried by the ST prior to leveling. The ST was closer

to the surface on the northern half (Fig. 12) and deeper to

the south.

On the southeastern edge of the medium grid, it was

common to find a thin (3 to 30 cm) layer of ST somewhere in

the profile. These layers usually contained more clay

(field textures ranging from sand to sandy loam) than was

found in ST elsewhere. This characteristic of higher clay

content in ST just above spoil was observed elsewhere on the

study site. The extra clay in the ST just above the spoil

may have resulted from either the filtering of the clay from

the first slurry water that passed by at the beginning of

pumping or may have come from spoil sloughing into the ST as

the ST slurry passed by the spoil piles.

o Ln

0 '0

-~ ,. I
1I E

cw, D
l 'I I 0






1 4-
~ CO


\ :/Ig- $I )

ul ~ .0
r! ~e -4-

The spoil island on the western part of the medium grid

was not completely buried during the final stages of ST

filling. In fact, this spoil island may have been the

source of the spoil cover to the north in the medium grid.

The results of the sampling scheme in this grid showed that

there were chemical and physical differences in spoil

characteristics depending on geographic position relative to

the spoil islands, as will be discussed later.

The surface layer of the minesoil in this field was

generally very slightly acid (Table 1). The H20-pH

typically was about 6.6, but ranged from pH 4.8 to 8. pH

values rather than H ion concentrations were used for

statistical analyses (Shiue and Chin, 1957). Organic carbon

(OC) content averaged 0.50%, ranging from 0.17 to 1.22%.

Average cation exchange capacity (CEC) was 10.0 meq/100 g,

ranging from 5.2 to 22.2 meq/100 g. Extractable acidity

averaged 4.7 meq/100 g and ranged from 0.74 to 10.59 meq/100

g. Texture ranged from sandy loam to sand but generally was

in the range of loamy sand. Bos et al. (1984) reported

average double acid extractable Mg and K values of 500 mg/kg

and 57 mg/kg and NH4OAC extractable Mg and K values of 177

mg/kg and 41 mg/kg for samples taken from the medium grid.

Average levels of NH4OAC extractable Mg and K for the whole

field were 2.1 meq/100 g (254 mg/kg) and 3.1 meq/100 g (58

mg/kg), respectively. Magnesium values ranged from 0.25 to

6.02 meq/100 g and K values ranged from 0.017 to 3.295

meq/100 g. Although the four bases Na, K, Ca, and Mg were



o I Oi ~- o ) (I I CNM (
0) C) 0 o ClO - 0 r

m 3 CM- C o I it)

m O O r 0 0 O


mT )N C) C- c ( c -

0O m
oi Y a- 0
E c

En C C- CO IC t CM, C'O CD C

Sn 0 o 0 0 5


L C 0) m 0 \ I C0
C . C . J ;

co 0. N CO- (N
o I DO L C) C) O ) M o

0 Sr
oM 0I 0 0t I O O

-0 4 ) IC Nb t )
4, 0' r- n.
O Dl rn un ip r o o C+ M o

o o
o o 0

a) m z L C c 0 04ct V C aL -0 m mOcr
SI 1 0 co CM C-- 0a L C) L
T- C 3 4u P S-.

M I C )

o 1 \ ( x
> oan = cco uw
t- 005 C E hr

05 0 2 .- C O 0 4C 05C) 0 ) 4

1-1, >) ia a) C) C) o. (.3a

included in the CEC determination, only K and Mg were

individually scrutinized. Potassium values appeared to

correlate well with Na values and Mg values with Ca values.

The Univariate procedure indicated that none of the

populations were strictly normal. HO2-pH and KCl-pH

populations were closest to normal. A population with a

small CV may be best described by either a normal or

lognormal distribution (Rao et al., 1979). Elevation and

sand populations may fit into this category.

In addition to the analysis of data from surface soils,

some data were analyzed on subsoil layers at selected

sites. Averages were computed for nine locations sampled to

115 cm or greater depths in the medium grid (Table 2). The

averages represent the sum of the layer values divided by

the number of layers in each profile. These nine locations

were chosen out of a possible 15 because all the layers in

each of the nine profiles were very close to the 25 cm ideal

length as described in the materials and methods section.

these profile averages varied slightly from the grid-wide

averages for surface soils, although it was not known

whether the differences were significant or not. Since it

was assumed that the sampling points were not random and

independent, statistical analysis was not pursued. Trends

were observed, however.

Surface soils, for example, had greater OC contents,

more extractable Mg and K, higher CEC, and higher

extractable acidity than subsurface soils. In surface soils


o o o o 0 o C 0 "I
0 0 0 0 0 0 0 0

SC) I) N

(0 -0 C
rC ( C-
C I m C) MO 0- MO cM C)
I I C i C N m MoN 0 LO N

.,- C O
a) x ,

00 0

S > CO

0rl Lr- I N C N C N O l

-4 0 c> n n () C) C C) c) t, C

.,- -C- C1 MO MO NO C) C) C) C) -: CL 4-1
En 1 wM C) MO M I
00 I x


[o 0c I -0 N 0co N 0no

4ro-'1 N, I LCo Ln L c ) L M )' CD o M
4 -- !C O I . . . . i

(I c 0 F7 0

0 0 X 0 C )
0 0I

O ( 0 0 0 0 D MO 0 0 0 0 0C)
0I ) -O ) ) C C C
C 0

C . . . . 4
[-. 0. C) O M O ON O ^- O N C
0 OM

3o I ..
C O O N -

and subsoils, KC1-pH was equivalent but H80-pH was higher in

the surface soils. The surface soils had a slightly finer

texture than those averaged over depth. Dolomite and

fertilizer were applied during pasture establishment and

probably account for some of the differences in H20-pH, KC1-

pH, Mg and K contents between surface soils and subsoils.

The average OC content of the surface soils was greater

than that of the subsoils, indicating incipient A horizon

formation. Otherwise, there was no trend (increasing or

decreasing) with depth. The sample found to contain the

most OC occurred, in fact, at a depth of 175 cm.

Large differences between the averages were not

expected because of the nature of material placement. The

dragline operators did not segregate the OB materials

according to likeness of properties, so there was just as

much chance of having clayey and sandy materials next to

each other as on top of each other.

Differences in spoil characteristics (slowly increasing

or decreasing values) were noted locally, specifically with-

in the medium-sized grid. All values except elevation were

averaged by row, over the six east-west rows of sampling

points (Table 3). Of these averages, only extractable K

values did not show a regular increasing or decreasing trend

across the rows. There were increases in KC1-pH, H20-pH,

CEC, OC, and extractable Mg from row A southward with the

highest values in either the E or F rows. Extractable

acidity and sand contents decreased in the same regular

manner as other values increased.


I MI 0 0 0 0 N 0
m -H -H +1 -H -H

S 0 0 0 0 0 0
I *

O hOI 4- 4- 4 -H 4 -H
(o I N x C- t- --

X 0 0 >- C' n
a O

h O O t-- CO K' O^
E C) N

*H -H +1 +1 +1 -
-0 c Il 1
E *HE e ^- N N N
0 I 0 -H -H -H +- -H -H
- I *H I "- 0 cO O 0> O
'D O I> N -K- N O

4E FI u I Cr Lr 0 a C
LOLIm W ,D I -t

< H
4P c'I 0 1 M In <- N a N

S(00.1 O "- N a O0 N

- H O+1 +1 +-1
S o I ) I on o a o 0

SI 4 4-I N 4-I 4-I -i
S' 1001 CO M 1 I N n-


N 0 M N C O
oCI 4 4 I -H -H 4 ,,
O -r 0 0" G CO N 0"

MO, CO M *
Io 0 0 0 C n

E.-I In -- CMa) M
U) I 4H +H -H -H 1-H -H

(0 CO II t< U O I CO

043 ^ tM -' -t- ^- K

One of the agronomically important aspects of creating

this mine land type spoil-capped ST, is the manner in which

the spoil is spread over the ST. The spreading process

seemed to have caused the general trend in values from row A

to F, in one of two ways.

First, there may simply have been differences in types

of materials. It is possible that during mining, the last

spoil dumped on top of the piles was sandier, slightly more

acid, and relatively low in dolomite. During reclamation,

as the bulldozers spread the spoil from the islands over the

ST, this material was spread. This explanation is not

highly plausible since, as noted above, the dragline

operators do not segregate materials during mining.

A more likely explanation was that significant amounts

of leaching occurred in the spoil between the time of mining

and reclamation. Mining ended in 1976, and final grading of

the site was not completed until 1980. This interval may

have been long enough for the upper 1 or 2 m of spoil to be

leached of some native, dolomite-derived Ca and Mg

carbonates. The difference in Mg concentrations between

rows A and F is 1.32 meq/100g, which roughly equates to 256

Kg per cubic meter of soil (assuming a bulk density of about

1.61 g/cm3). If half of the yearly rainfall (140 cm)

infiltrates and equilibrates with the dolomite (67 mg/l in

solution), it will take about 5 years to leach the 256 Kg

from the top meter of soil.

It was the upper, more leached spoil layers that were

spread out over the ST. The spreading uncovered the less

weathered spoil beneath. The spoil of rows A, B, and C

consisted, therefore, of the more leached materials, and

rows D, E, and F were dominated by the relatively

unweathered materials.


Values of certain soil parameters apparently depend

heavily on location relative to spoil islands. It is

important to design sampling schemes that correctly assess

that relationship for this and similar fields. Part of the

sampling design would be to study the spatial variability of

individual parameters to determine how close together the

sampling points must be to ensure that some points will fall

on the spoil islands and some will fall between the spoil


It is possible to determine an optimum sampling scheme

that can represent the field soil parameters to any desired

accuracy by using geostatistics. Geostatistical analyses

show the interdependence of sample values and their

locations in the field through the semi-variogram.

The semi-variograms (using all 204 values) of CEC, OC,

Mg, and K were described by spherical models (Table 4).

Semi-variograms of H20-pH, KC1-pH, and extractable acidity

were described by exponential models. The elevation semi-

variogram was best described by a parabolic model, and the

sand semi-variogram by a Gaussian model. The nugget

OC CO) 0


LO 0 0 "O
0 0 0 Co

C 4-











0) -

0 0; 0 0


In 'C CO R c O 0
K' T) N 0 T

Co 0

0 0

S 4- 4--a

o r-
-- -4 -4-

> I Ic
0 0 0
m0 0 0
c. a a
a)o x x


ca a a
-4 4 C
c)) a





0 '0
o -4



03 0



C *




.0 r


0' rC


I \ I n 2 0
- N C- )

variance, range, sill, anisotropy, and drift were the

important parts of structure determination. They provided

information on the sample population plus the adequacy of

the sampling scheme.

A semi-variogram with a non-zero intercept indicates

presence of a nugget variance. The nugget variance in this

study ranged from 0 for elevation to 17 for sand content

(Table 4). A nugget of zero indicated that all variablity

was associated with position, which was accounted for by the

sampling scheme. The higher nugget variances for the other

parameters indicated that the samples were not taken close

enough together to eliminate all positional variability.

Even though some of the nugget variances were

relatively close to the sill values (i.e., pure nugget

effect), all of the semi-variograms did exhibit structure.

A 95% confidence interval around the population variance

(not part of the geostatistics package) was calculated for

each sill value and in every case excluded the nugget

value. There was a chance that these nuggets were

artificially high since the smallest lag interval tested was

7 m. A distance of 7 m was used as the smallest lag because

any smaller lag distances did not have enough pairs to

adequately support the semi-variance at that distance.

Twenty samples taken 1 m apart were not enough to influence

the semi-variance at the first lag. The nugget variance

also indicated the occurrence of some randomness among

sample values independent of position. This randomness may

be due to one or more of (i) true, position-independent

variability in the field, (ii) sampling error in drawing

subsamples for lab analyses, and (iii) error in analytical


Range values indicate the distance beyond which pairs

of samples are independent. Ranges varied from 90 m for OC

for 400 m for extractable K (Table 4).

Just as Gajem et al. (1981) found different ranges and

sills of the semi-variograms depending on the spacing

between samples, differences were noted in this study

between the ranges and sills as calculated for the first 126

samples (the medium grid) versus the total number of

samples. The range and sill of the semi-variograms

calculated for all points increased from those ranges and

sills calculated for the first 126 points.

It is impossible at this time to identify whether

measurement bias (Jury, 1984) or the spatial structure of

the soils has a greater effect on the increased range that

occurred in going from 126 to 204 values. The structure may

also result from the unequal numbers of samples taken from

the three different grid spacings.

The most significant differences involved the shapes of

the semi-variograms of CEC and elevation. Bos et al. (1984)

reported a pure nugget of 4.5 for CEC (Fig. 13) and a

Gaussian curve for elevation (Fig. 14) with no range or

sill. The semi-variogram of CEC, with all 204 values




S...... ... .... 4 .....,..4 ..........1. .. ..... ........ ............ ...... .,.1.. ....... ....... .
S 2 4 6 1 12 14 1




12 ---- ---------------- -----------------

...... ....... ...... ,.......... .......... I ............... ............ ......... .. .. ........ .... .. : .
0 49 10 : 150 21 2514 30 : 50


Fig. 13. Cation exchange capacity direction-independent
semi-variogram calculated from 126 values
(medium grid) (modified from Bos et al., 19B4) and
from all 204 values.


0 0

o ... ... .. ....... ....... ....I.. ....I.. .I. .. .......... I .......... ..........
0 2 4 6 8 1 12 1 14 16 1- 20


C1C"i 1,

I - -


Fig. 14. Elevation direction-independent semi-variogram
calculated from 126 values (medium grid)
(modified from Bos et al., 1984) and from all
155 values.

4. i

-*. 0 -


included, was described by a spherical model with a range of

150 m and a sill of 9.6 (Fig. 13). The semi-variogram of

elevation still had no range and sill but was best fit by a

parabolic curve (Fig. 14).

When a semi-variogram exhibits a pure nugget, it

usually indicates that the samples were not spaced closely

enough to show any kind of structure. It appeared that the

sampling scheme used in this first study was inadequate to

show either the smaller-scale or the larger-scale structure

for CEC. The addition of the values from the second phase

of sampling showed a structure on a field-sized scale. The

60 widely spaced samples completely masked the pure nugget

and any effects of the 18 samples spaced 1 m apart. The

nugget of the new semi-variograms was larger than the pure

nugget (sill) of the first and the new sill, 9.6, was twice

as large as the old sill.

The different results show the need of choosing a

sampling scheme such that the number of closely spaced

samples is equivalent to samples spaced farther apart (e.g.,

65 samples in the small, 65 in the medium, and 65 in the

large grids). This kind of sampling scheme will identify

the structure much better than a system with different

numbers of samples at varying distances. In the present

study (13 samples in the small, 126 in the medium, and 60 in

the large grid), the best alternative was to choose the

limits of lag distance classes that the computer would use

to calculate each point of the semi-variogram. This

resulted in combining several odd distance classes together

so that the numbers of pairs supporting each semi-variance

were equal. These odd distance classes invariably masked

some semi-variogram structure that would have otherwise been


The change in the elevation semi-variogram from

Gaussian to parabolic does not seem significant at first

glance. Bos et al. (1954) correctly reported only the first

half of the elevation semi-variogram, which did not show any

noticeable shape. The Gaussian shape is most noticeable if

the same semi-variogram is shown to 200 m (Fig. 14).

Although as much confidence could not be placed on the

second half of the semi-variogram as on the first half, it

was still possible to see the tendency for the curve to

level out at about 180m. Comparing this to the parabolic

shape of the semi-variogram of all values (Fig. 14) showed a

much more significant difference. The new semi-variogram

showed no tendency to level out.

The difference between these two semi-varibgrams was

clearly caused by the presence of the slope in the field.

All 125 samples used in the calculation of the first semi-

variogram were on one of the level parts of the field. The

majority of the added samples were located on the sloping

land. The mean value of elevation decreased from the north-

central part of the field (medium-grid) across the slope to

the southern border. If this trend in the means could have

University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs