+ Models
FORECO-10716; No of Pages 14
Available online at www.sciencedirect.com
Forest Ecology and Management xxx (2007) xxx–xxx
www.elsevier.com/locate/foreco
Soil nutrient–landscape relationships in a lowland
tropical rainforest in Panama
Frauke K. Barthold a,*, Robert F. Stallard b,c, Helmut Elsenbeer a,c
a
University of Potsdam, Institute of Geoecology, Karl-Liebknecht-Strasse 24-25, 14476 Potsdam, Germany
b
US Geological Survey, 3215 Marine Street (Suite E127), Boulder, CO 80303-1066, USA
c
Smithsonian Tropical Research Institute, Balboa, Panama
Received 18 January 2007; received in revised form 30 July 2007; accepted 6 September 2007
Abstract
Soils play a crucial role in biogeochemical cycles as spatially distributed sources and sinks of nutrients. Any spatial patterns depend on soil
forming processes, our understanding of which is still limited, especially in regards to tropical rainforests. The objective of our study was to
investigate the effects of landscape properties, with an emphasis on the geometry of the land surface, on the spatial heterogeneity of soil chemical
properties, and to test the suitability of soil–landscape modeling as an appropriate technique to predict the spatial variability of exchangeable K and
Mg in a humid tropical forest in Panama. We used a design-based, stratified sampling scheme to collect soil samples at 108 sites on Barro Colorado
Island, Panama. Stratifying variables are lithology, vegetation and topography. Topographic variables were generated from high-resolution digital
elevation models with a grid size of 5 m. We took samples from five depths down to 1 m, and analyzed for total and exchangeable K and Mg. We
used simple explorative data analysis techniques to elucidate the importance of lithology for soil total and exchangeable K and Mg. Classification
and Regression Trees (CART) were adopted to investigate importance of topography, lithology and vegetation for the spatial distribution of
exchangeable K and Mg and with the intention to develop models that regionalize the point observations using digital terrain data as explanatory
variables. Our results suggest that topography and vegetation do not control the spatial distribution of the selected soil chemical properties at a
landscape scale and lithology is important to some degree. Exchangeable K is distributed equally across the study area indicating that other than
landscape processes, e.g. biogeochemical processes, are responsible for its spatial distribution. Lithology contributes to the spatial variation of
exchangeable Mg but controlling variables could not be detected. The spatial variation of soil total K and Mg is mainly influenced by lithology.
# 2007 Elsevier B.V. All rights reserved.
Keywords: Soil chemical properties; Lowland tropical rain forest; Soil–landscape relationships
1. Introduction
Soils act as spatially distributed sources and sinks of
nutrients, and the concomitant spatial patterns appear to be very
variable in forest ecosystems in general and in tropical forests
in particular (Silver et al., 1994; Burghouts et al., 1998; Wilcke
et al., 2001) and many variables interact and contribute to the
spatial distribution of soil chemical properties. Tropical forest
soils have been studied for many decades (Milne, 1935;
Sánchez, 1976; Baillie, 1989; Richter and Babbar, 1991), but
their genesis and functioning have not been fully understood
yet. Our understanding of the functioning of tropical forests and
* Corresponding author at: Institute of Landscape Ecology and Resources
Management, University of Giessen, Heinrich-Buff-Ring 26-32, 35392
Giessen, Germany. Tel.: +49 331 977 2047; fax: +49 331 977 2068.
E-mail address: fkbarthold@gmail.com (F.K. Barthold).
of their ecological organization will undoubtedly improve once
we identify the environmental factors that control the spatial
pattern of soil nutrients.
Various studies investigated relationships between soil
chemical properties and landscape features using advances in
soil–landscape modeling (e.g., Moore et al., 1993; Gessler et al.,
1995; Florinsky et al., 2002; Park and Burt, 2002; Baxter and
Oliver, 2005; Henderson et al., 2005), but few dealt with the
tropics (e.g., Holmes et al., 2005). Soil–landscape modeling
helps to predict the spatial patterns of soil properties and to
regionalize point observations by quantifying the relationship
between environmental variables and soil properties of interest. It
relies on the applicability of Jenny’s (1941) state factor equation:
S ¼ f ðcl; o; r; p; t . . .Þ;
where every soil property (S) is a function of climate (cl),
organisms (o), relief (r), parent material ( p), time (t) and
0378-1127/$ – see front matter # 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.foreco.2007.09.089
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089
+ Models
FORECO-10716; No of Pages 14
2
F.K. Barthold et al. / Forest Ecology and Management xxx (2007) xxx–xxx
possible other variables (. . .). This equation may be solved by
varying one factor while keeping the other factors constant.
Soil–landscape modeling focuses on topography as the dominant independent variable that influences the way water moves
through and over the land surface, and that can aid tremendously in exploring spatial patterns of soil properties (McKenzie et al., 2000). Digital terrain analysis offers the means to reexpress the vague term topography in precise, quantitative
variables (Wilson and Gallant, 2000), and various statistical
techniques have been developed to quantify the relationships
between topography-based environmental variables and soil
properties (McBratney et al., 2003). Any spatial patterns
derived by these techniques, however, cannot necessarily be
generalized or extrapolated geographically.
On Barro Colorado Island, Panama, a remarkably coherent
picture of the ecology of a tropical lowland forest emerged
thanks to a comprehensive body of work that includes geology
(Woodring, 1958), topography, soils and hydrology (Dietrich
et al., 1982), long-term rainfall and soil moisture records (Rand
and Rand, 1982) soil nutrient dynamics (Yavitt and Wieder,
1988; Yavitt and Wright, 1996; Yavitt, 2000), and forest
structure and vegetation history (Foster and Brokaw, 1996). The
relationships between landscape features and soil properties
have so far received less attention (Johnsson and Stallard,
1989), although a spatially explicit inventory of soil nutrients
would further contribute to our understanding of the functioning and the organization of this and other tropical forest
ecosystems at a soilscape level.
Hence, our objective was to elucidate the relationships
between environmental variables and selected soil chemical
properties and to quantify soil nutrient stocks in a spatially
explicit manner. We hypothesized that selected soil exchangeable cations (K and Mg) vary with topography, lithology and
vegetation, and we used explorative data analysis and
Classification and Regression Trees (CART) to understand
and quantify the relationships between the landscape features
and exchangeable K and Mg.
Fig. 1. Digital elevation model of Barro Colorado Island and the location of the
sampling sites. Inset: Barro Colorado Island’s location in Panama.
formation of late Oligocene age (Woodring, 1958) (Fig. 2).
Both formations are of sedimentary origin and subdivided into
two facies, volcanic and marine. The two facies of the Bohio
formation mainly consist of conglomerate composed of basaltic
clasts set in a sandy volcaniclastic matrix (Johnsson and
Stallard, 1989), only the volcanic facies is outlined in
Fig. 2. The Caimito marine facies is primarily a foraminiferal
2. Material and Methods
2.1. Study area
The study area is Barro Colorado Island (BCI), an island
which was created when the Panama Canal was flooded,
located in the Gatun Lake in the Republic of Panama (9890 N,
798510 W) (Fig. 1). The island comprises 1500 ha, which is
covered by old growth forest and by low and tall young forest.
The forest is classified as a Tropical Moist Forest in the
Holdridge system (Holdridge and Budowski, 1956). Mean
annual rainfall averages 2600 mm, with a pronounced dryseason between January and April (Dietrich et al., 1982). Mean
annual temperature is 27 8C and the climate is classified as
Tropical Monsoon [Am] in the Köppen Climate Classification
(Croat, 1978). The main soils are Oxisols and Inceptisols
(Baillie et al., in preparation, Soil Survey Staff, 2006). The
island is formed by two geologic formations of Oligocene age:
the Bohio formation dates back to early Oligocene, and Caimito
Fig. 2. Geologic map of Barro Colorado Island with 10 m contour lines.
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089
+ Models
FORECO-10716; No of Pages 14
F.K. Barthold et al. / Forest Ecology and Management xxx (2007) xxx–xxx
limestone with abundant pelecypods and the Caimito volcanic
facies consists of volcaniclastic sandstone (Johnsson and
Stallard, 1989). The only non-sedimentary lithology on the
island is an andesite flow capping the island (Fig. 2). The
andesite reaches a maximum thickness of 85 m (Woodring,
1958). It is not younger than the Caimito formation and
represents the episode of volcanic and intrusive activity
between Oligocene and early Miocene. It is a non-vesicular,
phenocrystic andesite with veins and vugs (Johnsson and
Stallard, 1989). The main minerals of the andesite are
plagioclase, clinopyroxene, orthopyroxene and magnetite
whereas the veins and vugs contain quartz, calcite and zeolite
(Johnsson and Stallard, 1989).
The topography on BCI is considerably contrasting and
clearly related to the underlying geology (Fig. 2). The
topography on the western part of the fault system is cuestalike with the top corner of the dipslope in the NE of the hilltop,
and a general gentle dip to the SW (Fig. 2) (Baillie et al., in
preparation). The dipslope, which appears to be structurally
controlled by the upper extant surface of the andesite flow, has
several subunits, with the large gently dipping upper plateau
fringed to the south and west by gently graded treads, which are
separated by slightly steeper and moderately bouldery risers
(Baillie et al., in preparation). There is a further low rocky riser
down to the lowland on the marine sediments in the south and
west (Baillie et al., in preparation). The streams are arranged in
a radial pattern around the hilltop and are little incised on the
andesite and the marine sedimentary dipslope terrains and
deeply incised on the Bohio volcanic formation (Woodring,
1958) (Fig. 2). This formation, covering the north, northwest
and an area stretching through the centre of the island, also
forms the scarp element of the cuesta with steeper terrains
(Baillie et al., in preparation) (Fig. 2). The Caimito volcanic
formation to the east of the fault system also appears to have a
slight cuesta-like form, although this is lower and less
pronounced. Its slope dips towards the south and the scarp,
which extends down to the lake, has a discontinuous midslope
ledge.
2.2. Sampling design
We used a design-based, stratified sampling plan (Brus and
DeGruijter, 1997) with lithology, vegetation and topography as
stratifying variables, along the lines of McKenzie et al. (2000).
Using spatial data on lithology, vegetation and topography
(stored in GIS systems), we created four lithology classes, three
vegetation classes and three topography classes. Vegetation is
expressed as forest age and topography is represented as
Topographic Wetness Index (TWI). The Topographic Wetness
Index was calculated as followed:
A
TWI ¼ log
tan a
where A is the specific contributing area and a is slope.
The density function of the TWI was used to calculate
quantiles on an equal area basis, which provided the class
3
boundaries. The combination of forest age and Topographic
Wetness Index yielded in nine classes on each lithology or 36
classes for the whole study area. In order to avoid sampling
in a wrong class due to dislocation in the field we applied
exclusion rules. Areas smaller than 0.01 ha were excluded.
This size was chosen because our GPS (Trimble Asset
Surveyor v. 5.00, Model TSC1) has an accuracy of 1–10
horizontal meters. Also, a 50 m-wide buffer zone was
established along all lithological boundaries, and patches
within this buffer zone were excluded. This is because the
lines drawn on the geologic map are not based upon visible
contacts in the field and also the topography suggests that for
example the andesite unit extends much further across its
map boundaries on its western side (BCI Soil Survey 2006,
personal observations). After application of the exclusion
rules three replicates of each class were randomly drawn
which resulted in 108 patches. From each patch one sampling
site was also randomly selected. The sites provide the exact
coordinates for sampling. The locations of the sampling sites
are illustrated in Fig. 1.
2.3. Data
We used lithology, vegetation and terrain attributes
(Table 1) as predictor variables. The lithology is the ultimate
source of cations, the vegetation influences soil chemistry by
uptake and release of nutrients, and the topography determines
processes that transport nutrients along the surface. A geologic
map by Woodring (1958) modified by Johnsson and Stallard
(1989) already existed in digital form. Digital data on
vegetation included forest age, which was inferred from a
color range reflected from an aerial photograph (Svenning
et al., 2004). Topographic information was gained from a
digital elevation model (DEM). We generated two sets of the
same topographic variables from two different DEMs. One
DEM had been generated from a 1:25,000 topographic map
with 10 m contour lines by the Defense Mapping Agency on
the basis of an aerial photograph. Another DEM was developed
from a 1:10,000 topographic map with 6 m contour lines by
Miller (1927). Both maps were transformed to DEMs in
ArcInfo using the Topogrid algorithm, which is based upon the
ANUDEM program by Hutchinson (1989). The selected
topographic variables are listed in Table 1 with a description of
their generation (software and algorithm) and detailed
definitions and significances are provided by Wilson and
Gallant (2000).
2.4. Field methods
During the rainy season of 2005, we located sites with a GPS
device (Trimble Asset Surveyor v. 5.00, Model TSC1) and
sampled five different depths of mineral soil down to 1 m: 0–
5 cm, 5–25 cm, 25–50 cm, 50–75 cm and 75–100 cm, unless
bedrock was encountered. We used an Eijkelkamp auger with
7 cm in diameter and drilled at three places within the site to
obtain a support of 115 cm2. Depth to weathered rock was
determined in the field. The samples were oven dried at 60 8C.
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089
+ Models
FORECO-10716; No of Pages 14
4
F.K. Barthold et al. / Forest Ecology and Management xxx (2007) xxx–xxx
Table 1
Selected topographic variables and description of their generation
Software
Primary terrain attributes
Elevation
Slope
Flow direction
Specific catchment area (SCA)
Curvature
Plan curvature
Profile curvature
Total upslope length
Longest upslope length
Secondary terrain attributes
Topographic Wetness Index (TWI)
Transport Capacity Index (TCI)
Arc Map
TauDEM
TauDEM
TauDEM
Arc Map, Spatial Analyst
Arc Map, Spatial Analyst
Arc Map, Spatial Analyst
TauDEM
TauDEM
TauDEM
Arc Map, Raster Calculator
Stream Power Index (SPI)
Arc Map, Raster Calculator
Distance to stream
TauDEM
Strahler Network Order
TauDEM
Basic algorithm
Source of algorithm
D1
D1
D1
FD8
FD8
FD8
Variables which
were generated
using D8 (flow direction)
Variables which
were generated
using D8 (flow direction)
Tarboton (1997)
Tarboton (1997)
Tarboton (1997)
Moore et al. (1991), Zevenbergen and Thorne (1987)
Moore et al. (1991), Zevenbergen and Thorne (1987)
Moore et al. (1991), Zevenbergen and Thorne (1987)
O’Callaghan and Mark (1984)
D1
Using variables which
were generated with D1
Using variables which
were generated with D1
Defined by D8 flow
directions to the streams
Variables which were
genereated using D8
(Flow Direction)
Tarboton (1997)
Tarboton (1997)
Bulk density was measured at selected soil pits from the BCI
Soil Survey 2006 (Baillie et al., in preparation) in 10 cm
increments down to 50 cm and at selected sites measurements
were taken in 10 cm intervals down to 1 m. Additional
measurements down to 50 cm were also taken at selected sites
of the study of Grimm et al. (in preparation). Measurements
were conducted after the compliant cavity procedure (4A5)
(Soil Survey Staff, 1996). A cylindrical hole was excavated at
the zone surface to the desired depth and the excavated soil
bagged, oven dried at 105 8C in the lab and then weighed in
order to obtain a value for the total mass. The dry soil was then
wet-sieved through a 2 mm mesh and the obtained fine-earth
fraction was again dried at 105 8C and then weighed. The
resulting hole was lined with thin plastic, water was added to a
datum level and the volume of the water was measured. Soil
corrections were made for weight and volume of the rock
fragments (assuming a rock density of r = 2.65 g/cm3) and bulk
density was calculated as followed:
BD ¼
Wf
Ve
where BD is the bulk density (g cm3), Wf is the oven-dry
weight of <2 mm material (g), Ve is the excavation volume of
<2 mm material (g cm3).
Stoniness data was taken from two different sources and the
information was combined for uncertainty propagation
purposes. Field estimates from the BCI Soil Survey 2006
(Baillie et al., in preparation) served as input source as well as
O’Callaghan and Mark (1984)
Tarboton (1997)
O’Callaghan and Mark (1984)
O’Callaghan and Mark (1984)
lab measurements of rocks larger than 2 mm from the bulk
density determination.
2.5. Laboratory methods
2.5.1. Exchangeable cations (K and Mg)
Exchangeable potassium and magnesium were extracted
using 1M CaCl2 (Helmke and Sparks, 1996) and analyzed with
an Atomic Absorption Spectrometer. Relative precision of lab
methods is 17.6% for K and 19.8% for Mg.
2.5.2. Total elements (K and Mg)
Total concentrations of elements were analyzed with a HF/
HClO4-digestion method modified after Heinrichs and
Hermann (1990). Samples were ground with an agate mortar.
Then, between 0.15 g and 0.2 g of soil were weighed and the
samples were moistened with H2OMilliQ. Two milliliters of
HClO4 (70%) and 6 ml of HF (40%) were added. The
solutions were heated at 85 8C over night. Teflon containers
were then opened and the solutions were fumed off at 185 8C.
One milliliter concentrated HCl and 8 ml H2OMilliQ were
added to the residue and the solution was heated again to
85 8C over night. The cooled digestion solution was filtered
and filled up to 50 ml with H2OMilliQ. Samples were measured
using an ICP-OES (Vista-MXP, Varian). For each element,
two wavelengths were used and the one providing the better
results was used (determined by intensity, symmetry and
degree of interference). For K, this was 766.491 nm and for
Mg, 279.553 nm. Relative precision of lab methods is 2.2%
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089
+ Models
FORECO-10716; No of Pages 14
F.K. Barthold et al. / Forest Ecology and Management xxx (2007) xxx–xxx
5
Fig. 3. Grouped boxplots of exchangeable K concentration on each of the four lithologies, the notches represent 95% confidence intervals around the median. Sample
sizes range from 9 to 27.
for K and 9.4% for Mg (value is based upon three replicates
on eight sites (24 samples).
measurements from the bulk density survey:
2.5.3. Statistical analysis and modeling
We used the R software (R Development Core Team, 2005)
for exploratory data analysis and resistant interval estimation.
We used the 95% confidence interval (CI) as a measure of
uncertainty in all our analyses. Nutrient stocks on a mass basis
were calculated using concentration, bulk density, thickness of
the soil layer and stoniness after Ellert and Bettany (1995) and
Huntington et al. (1989):
where S is the stoniness (skeletal fraction >2 mm); SFE is the
stoniness from field estimations; SBD is the stoniness from bulk
density measurements (skeletal fraction >2 mm).
Variables that are a function of other variables contain two
different sources of uncertainty: (1) the uncertainty of
measurement and (2) the uncertainty due to spatial variation
of the individual variables which are in this case stoniness and
bulk density. In order to obtain the value of uncertainty due to
spatial variation of the soil chemical element mass (in Mg/ha)
we performed standard uncertainty propagation (Taylor, 1997)
assuming that stoniness and bulk density are independent from
each other, due to the fact that the measurements of bulk density
did not account for fragments >10 cm which make up a large
component in stoniness quantifications on BCI.
For model building purposes we applied the method of
Classification and Regression Trees performed with the rpartpackage in R (Therneau and Atkinson, 2005). CART is a nonparametric method used to explore and predict data. CART
handles non-linear and non-additive relationships; it uses
M element ¼ Celement BD T 10; 000 m2 ha1
0:001 Mg kg1 ð1 SÞ
where Melement is the mass of element per unit area (Mg ha1);
Celement is the concentration of element (kg Mg1); BD is the
bulk density of the soil (Mg m3); T is the thickness of the soil
layer (m); S is the stoniness of the soil (unitless).
The value for stoniness for each soil type was calculated as
the sum of the field estimates from soil survey data and of the
S ¼ SFE þ SBD
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089
+ Models
FORECO-10716; No of Pages 14
6
125.1 (27.1)
78.2 (18.0)
82.1 (21.0)
121.2 (27.3)
89.9 (14.9)
70.4 (14.9)
recursive partitioning of the dataset into smaller and
successively homogenous subsets. Details are provided in
Breiman et al. (1984), Myles et al. (2004), and De’ath and
Fabricius (2000).
3. Results
3.1. Potassium
a
(21.8)
(6.8)
(5.7)
(3.7)
(3.7)
(3.1)
(3.7)
27.4
19.6
11.7
11.7
7.8
7.8
7.8
0–5
5–10
10–20
20–30
30–50
50–80
80–110
(1.7)
(0.8)
(20.6)
(5.7)
33.2
16.8
9.0
2.4
105.6 (3.1)
176.0 (7.7)
0–10
10–20
20–50
50–100
0–5
5–25
25–50
50–75
75–100
(15.2)
(13.8)
(8.8)
(8.1)
(7.3)
106.6
69.1
50.3
34.5
31.6
Cano Negro,
Nicaragua
n = 246
Kukra,
Nicaragua
n = 246
Sample sizes for BCI soil depths: 107 (0–5 cm), 108 (5–25 cm), 98 (25–50 cm), 83 (50–75 cm), 64 (75–100 cm).
0–20
20–40
40–60
74.3 (15.8)
54.7 (13.1)
54.7 (21.0)
78.2 (9.8)
62.6 (13.1)
70.4 (16.4)
Creek,
Taiwan
n=9
Leeward
slope,
Taiwan
n = 12
Windward slope,
Taiwan n = 8
Ridge,
Taiwan
n = 11
Mackensen et al.
(2000)
Belem,
Para,
Brazil
n=4
Blair (2005)
Depth
(cm)
This
study
BCIa
Depth
(cm)
Table 2
Arithmetic means ( one-half CI) of exchangeable K (mg/kg)
Markewitz et al.
(2001)
Paragominas,
Para, Brazil
n=3
Depth
(cm)
Depth
(cm)
Chen et al. (1997)
F.K. Barthold et al. / Forest Ecology and Management xxx (2007) xxx–xxx
3.1.1. Exchangeable K concentrations
Exchangeable K is almost equally distributed across the
whole island. Box plots conditioned by lithology illustrate that
values on Andesite and Caimito volcanic are slightly lower than
on Bohio volcanic and Caimito marine, but these differences
are not significant except for the deepest depth (75–100 cm)
where Bohio volcanic (43.9 10.3) (median 95% CI) is
significantly higher than andesite (21.7 11.19) (Fig. 3).
On all lithologies, the concentration of exchangeable K
decreases significantly with increasing depth between the
topsoil (0–5 cm) and the mineral soil (>5 cm).
Table 2 compares mean exchangeable K concentrations on
BCI with those from other tropical lowland forests across
Central and South America and Taiwan. The table shows that
the nutrient status on BCI is on the upper end of nutrient
concentrations in regard to exchangeable K, similar to the sites
in Nicaragua (Blair, 2005), slightly higher than Taiwan (Chen
et al., 1997) and much higher than the sites in Paragominas and
Belem, Para, Brazil (Mackensen et al., 2000; Markewitz et al.,
2001).
3.1.2. Total K concentrations
The total K concentration varies with lithology (Fig. 4), with
a concentration almost twice as high on the volcanic facies of
the Bohio formation than on the other lithologies. However, the
dot plots illustrate that these differences are not significant
between the mineral soil (>5 cm) of the Bohio volcanic facies
and andesite, between the depth from 5 to 25 cm on Bohio
volcanic and Caimito marine and between the deepest depth
from 75 to 100 cm on Bohio volcanic and all other lithologies,
which is due to the wide confidence intervals of K values on
Bohio volcanic (Fig. 4). There is a slight downward trend with
decreasing concentrations across the island but this trend is not
significant (Fig. 4).
3.1.3. Exchangeable and total K stocks
The mean content of exchangeable K on BCI in the upper
50 cm is 0.25 0.03 Mg/ha (mean one-half CI). On all
lithologies except on the Caimito volcanic formation, the stocks
increase to a depth of 25 cm and then decrease again down to
50 cm (Table 3). The mean content of total K in the upper 50 cm
is 5.2 1.0 Mg/ha (mean one-half CI). In the uppermost
0.5 m, the total K stock increased with depth in the following
intervals: 0–5 cm; 5–25 cm and 25–50 cm (Table 3).
3.1.4. Regression Tree Models
The Regression Tree Models which we developed for every
depth with concentration of exchangeable K as response
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089
+ Models
FORECO-10716; No of Pages 14
F.K. Barthold et al. / Forest Ecology and Management xxx (2007) xxx–xxx
7
Fig. 4. Dot plots of total K distribution on the four lithologies on Barro Colorado Island. The empty circles represent the value of each sample, the filled circles
represent the median and the crosses represent the lower and upper 95% confidence limits. Sample sizes range from 9 to 12.
variable explained 0, 4, 13, 1 and 8% (for 0–5 cm, 5–25 cm, 25–
50 cm, 50–75 cm and 75–100 cm, respectively) of the spatial
variability with the soil class being the most important predictor
(Table 4). Topography and forest age explained only a
negligible fraction and none of the spatial variation,
respectively (Table 4).
America, Hawaii and Taiwan. The table shows that exchangeable Mg concentrations on BCI are similar to K at the upper end
of the concentration range across the tropics: similar to the sites
in Panama and Colombia (Golley, 1986) but exceeding the sites
in Hawaii (Scowcroft et al., 2004), and by far those in Brazil
(Mackensen et al., 2000; Markewitz et al., 2001) and Taiwan
(Chen et al., 1997).
3.2. Magnesium
3.2.1. Exchangeable Mg concentrations
The concentrations of exchangeable Mg are up to 10 times
higher than those of exchangeable K (Figs. 3 and 5),
exchangeable Mg, however, is not equally distributed
throughout the study area (Fig. 5). Box plots of exchangeable
Mg, again conditioned by lithology (Fig. 5), illustrate that in
contrast to exchangeable K concentrations there is a large range
in measured exchangeable Mg values within each lithology.
The lowest concentrations are on andesite and Caimito
volcanic, which are alike, and the highest values are on Bohio
volcanic and Caimito marine, which are also alike. However,
concentrations on Bohio volcanic are significantly higher than
on andesite and Caimito volcanic, while concentrations on
Caimito marine are only significantly higher than on Caimito
volcanic excluding the uppermost 5 cm (Fig. 5).
There is a downward trend with decreasing concentrations
but differences are only significant on Caimito volcanic.
Table 5 lists mean exchangeable Mg concentrations of BCI
in comparison with mean exchangeable Mg concentrations
from other tropical lowland forests across Central and South
3.2.2. Total Mg concentrations
Values of total Mg concentrations have large ranges on all
lithologies except on Caimito volcanic but highest values are
found on Bohio volcanic and Caimito marine (Fig. 6). However,
the dot plots illustrate that these differences are not significant
from the lower values on andesite and Caimito volcanic. Total
Mg concentration does not change with depth.
3.2.3. Exchangeable and total Mg stocks
The mean content of exchangeable Mg is 3.0 0.3 Mg/ha
(mean one-half CI). In contrast to the exchangeable K stocks,
exchangeable Mg increases continuously from 0 to 50 cm
(Table 6). The mean content of total Mg is 25.8 5.1 Mg/ha
(mean one-half CI). The total Mg content in the uppermost
0.5 m increases with depth in the following intervals: 0–5 cm;
5–25 cm and 25–50 cm (Table 6).
3.2.4. Regression Tree Models
The Regression Tree Models for exchangeable Mg concentrations yielded slightly better results than for exchangeable K
and explained 0, 23, 33, 47 and 33% (for 0–5 cm, 5–25 cm,
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089
+ Models
FORECO-10716; No of Pages 14
8
5.19 (1.02)
2.39 (0.43)
Predictor variables
Cross-validated error
Exchangeable
K (mg/kg) in
Exchangeable
K (mg/kg) in
Exchangeable
K (mg/kg) in
Exchangeable
K (mg/kg) in
Exchangeable
K (mg/kg) in
–
1.02
Soil type
0.96
Soil type
0.87
Soil type,
plan curvature
Soil type
0.99
0–5 cm
5–25 cm
25–50 cm
50–75 cm
0.92
75–100 cm
Bohio v.
1.62 (0.93)
2.87 (1.60)
3.31 (1.76)
7.8 (2.56)
1.11 (0.64)
2.00 (0.95)
2.53 (1.32)
5.63 (1.75)
4. Discussion
0.079 (0.016)
0.085 (0.021)
0.082 (0.018)
0.246 (0.031)
0.062 (0.019)
0.048 (0.019)
0.034 (0.018)
0.144 (0.032)
25–50 cm, 50–75 cm and 75–100 cm, respectively) of the
spatial variability (Table 7). Again, soil class was the most
important variable. More importantly, topography and forest
age explained only little to none of the spatial variability
(Table 7).
Andesite
3.81 (0.57)
1.01 (0.30)
1.99 (0.63)
2.19 (0.74)
0.51 (0.15)
0.87 (0.30)
1.01 (0.26)
Response variable
BCI
0.72 (0.17)
1.50 (0.32)
1.59 (0.43)
Table 4
CART results
Caimito v.
Total K
Caimito m.
Caimito v.
BCI
F.K. Barthold et al. / Forest Ecology and Management xxx (2007) xxx–xxx
The soils of BCI are, regarding exchangeable K and Mg,
as fertile as other soils in Central America and Taiwan
and more fertile than soils in the Amazon basin. This may
be accredited to the younger geological history of the
Isthmus of Panama, which formed only 3.1–2.8 million years
ago (Coates and Obando, 1996) and therefore supports a
younger landscape with shallower and presumably less
weathered soils than in the continental tropics of the Amazon
Basin.
4.1. Differences in exchangeable cation concentrations
within the tropics
0.257 (0.062)
0.197 (0.038)
0–50
0.344 (0.105)
0.074 (0.032)
0.098 (0.039)
0.085 (0.036)
0.100 (0.056)
0.145 (0.079)
0.099 (0.038)
0.056 (0.016)
0.084 (0.027)
0.057 (0.022)
0–5
5–25
25–50
Caimito m.
Bohio v.
Andesite
Exchangeable K
Depth (cm)
Table 3
Exchangeable and total K stocks (Mg/ha) (mean one-half CI)
4.2. Exchangeable cations
Exchangeable K is bound to the outer surfaces of clay
minerals and humic substances (Helmke and Sparks, 1996).
The decreasing exchangeable K concentrations with depth
show the same trend as the organic carbon concentrations on
BCI (Grimm et al., in preparation). This suggests that
exchangeable K is preferentially bound to humic substances.
On BCI, exchangeable K resources are limited (S.J. Wright,
personal communication). The higher values of exchangeable
K in the topsoil than in the subsoil suggest that plants obtain the
bulk of their K from the topsoil and that in soils with limited
cation concentrations the recycling processes mainly take place
in the topsoil.
These results are in line with the exchangeable Mg results.
In contrast to K, Mg is not a limited nutrient on BCI.
The exchangeable Mg concentrations do not show such
a significant trend with increasing depth except on the
Caimito volcanic facies. This emphasizes the fact that
recycling processes in soils with limited nutrients take place
in the topsoil, from which the plants mainly obtain their
nutrients.
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089
+ Models
FORECO-10716; No of Pages 14
160.4 (53.9) 255.2 (58.9) 179.9 (51.0) 269.8 (78.5)
82.6 (39.2) 158.0 (48.8) 119.1 (29.3) 162.8 (65.4)
119.1 (45.7) 179.9 (38.6) 102.1 (32.4) 143.4 (54.2)
Windward
slope,
Taiwan
n=8
n.a., not available.
a
Sample sizes for BCI soil depths: 105 (0–5 cm), 107 (5–25 cm), 97 (25–50 cm), 83 (50–75 cm), 66 (75–100 cm).
302
347
0–15
(19.3)
(7.7)
(7.7)
(3.9)
(3.9)
(3.1)
(3.1)
48.6
21.9
14.6
12.2
9.7
9.7
9.7
0–5
5–10
10–20
20–30
30–50
50–80
80–110
0–10 147.3 (62.2)
10–20
75.6 (90.1)
20–50
61.0 (89.4)
50–100 39.1 (50.6)
940
(72.3) 583
(90.2)
(109.9)
(116.0)
(119.3)
769.1
709.5
695.4
637.8
609.7
0–5
5–25
25–50
50–75
75–100
Panama NW
Colombia
n=8
n=6
BCIa
9
4.3. Total cations
0–20
20–40
40–60
Ridge,
Taiwan
n = 11
Bottom, Hakalau
Forest,
Big Island,
HI n = n.a.
Slope, Hakalau
Forest,
Big Island,
HI n = n.a.
Belem, Para,
Brazil n = 4
Paragominas,
Para,
Brazil n = 3
Markewitz
et al. (2001)
Depth
(cm)
Golley (1986)
This
study
Depth
(cm)
Table 5
Arithmetic means ( one-half CI) of exchangeable Mg (mg/kg)
Depth
(cm)
Mackensen
et al. (2000)
Depth Scowcroft et al. (2004)
(cm)
Depth (cm) Chen et al. (1997)
Leeward
slope,
Taiwan
n = 12
Creek,
Taiwan
n=9
F.K. Barthold et al. / Forest Ecology and Management xxx (2007) xxx–xxx
The total concentration of cations comprises four different
forms in which the cations exist: solution, exchangeable, fixed
(or non-exchangeable) and mineral (or structural). The fixed
form is that portion of K, that is held between adjacent
tetrahedral layers of dioctahedral and trioctahedral micas,
vermiculites and integrade clay minerals such as chloritized
vermiculite whereas the mineral form of K is bonded within the
crystal structure of soil mineral particles (Helmke and Sparks,
1996). Helmke and Sparks (1996) mention that 98% of the total
concentration belongs to the mineral form. The spatial
distribution of total cation concentrations might therefore be
closely linked to the underlying lithology and its mineral
composition. Table 8 provides a summary of the rock types and
their mineral and chemical composition. The abundance of total
K in the soils developed on Bohio volcanic results from the
mineral composition of this formation, which includes
hornblende. Andesite, Caimito volcanic and marine all are
made up of minerals that contain only little or no K, which is
reflected in the total K concentration of the soils (Fig. 6).
The spatial distribution of total Mg concentrations can also
be explained by the mineral composition of the rock types of the
different formations. Fig. 6 illustrates that total Mg concentrations are highest on Bohio volcanic and Caimito marine and
lowest on Andesite and Caimito volcanic. The magnesium
sources for the pool in Bohio volcanic soils are the abundant
minerals hornblende and clinopyroxene. On the Caimito
marine facies the high Mg concentrations may stem from a
magnesium-containing calcite. Although there are no studies
on the chemical composition of this mineral on BCI, the age of
this formation refers to a time when the Mg2+/Ca2+-ratio in
seawater were high, which is accredited to a change in volume
of the mid-atlantic ridge zone to smaller values, and which led
to precipitation of calcite where Mg2+ replaces Ca2+ in the
crystal structure (Stanley, 1999). This mechanism is today
expressed in high total Mg concentrations in the soils on the
Caimito marine facies.
The slightly higher Mg values on andesite than on Caimito
volcanic may be attributed to the andesite containing
clinopyroxenes, orthopyroxenes and zeolites (Table 8) whereas
the Caimito volcanic facies does not contain any minerals that
are made of Mg.
There is abundant evidence that suggests that total nutrient
concentrations are controlled by lithology, however, we cannot
put any numbers on the chemical composition of the rock types.
4.4. Relationships between soil chemical properties and
landscape features
Besides assessing nutrient pools on BCI, the second aim of
this study was to investigate the spatial distribution of K and
Mg, and to determine the importance of soil–landscape
relationships for a lowland tropical rain forest soil chemistry.
Johnsson and Stallard (1989) describe the contrasting landscapes on BCI where flat terrains with low landscape gradients
are developed on two contrasting lithologies, on the Andesite
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089
+ Models
FORECO-10716; No of Pages 14
10
F.K. Barthold et al. / Forest Ecology and Management xxx (2007) xxx–xxx
Fig. 5. Grouped boxplots of exchangeable Mg concentration on each of the four lithologies, the notches represent 95% confidence intervals around the median.
Sample sizes range from 10 to 27.
flow and the volcanic sediments of the Caimito volcanic facies,
and steeper terrains are developed on Bohio volcanic and
Caimito marine facies. They conclude from the composition of
stream sediments derived from different terrains that steeper
terrains are weathering-limited and support soils that are
shallower and richer in cations than soils that are located on the
flatter, transport-limited terrains. Our results are not consistent
with these conclusions and we need to discuss the two cations
of interest separately.
Our exchangeable Mg results (Fig. 5) are in line with the
findings of Johnsson and Stallard (1989): Mg concentrations
differ according to lithology and values are higher on the deeply
dissected terrains of Bohio volcanic and Caimito marine, if only
just significantly different from Bohio volcanic. It is therefore
surprising that the topography does not emerge as a controlling
factor in the regression trees. There are several reasons that
might explain the weakness of the relationships between the
topography and spatial variation of exchangeable Mg:
First, two different sets of the same terrain attributes were
generated from two different DEMs that we used to quantify the
relationship between topography and soil chemical properties.
Each DEM was developed from a different source of
topographic information, and their qualities and potentials in
reflecting the true landform vary. One DEM was generated from
a 1:25,000 m topographic map by the Defense Mapping
Agency that was made from an aerial photograph. Many
subtleties of the earth surface are not shown due to the lush
vegetation on BCI, which often obscures topographic
irregularities (Johnsson and Stallard, 1989). The second
DEM was developed from a 1:10,000 m topographic map by
Miller (1927). The map is developed on a finer scale but it is
hand drawn and can only be as good as the human perception of
the environment is. Therefore, not every point in the study area
may truly represent the form of the earth surface.
Secondly, processes that are responsible for spatial variation
of soil chemical properties might take place on a smaller scale
than the one captured with our DEM. Overland flow, which is
thought to play an important role in surface processes that
erodes and accumulates material, does not occur uniformly
across the island (Godsey et al., 2004). In theory, various factors
such as rain intensity, depth to an impermeable soil layer and
topography influence the generation of overland flow. Topographic variables such as slope, size of catchment area and
topographic wetness index, are thought to determine overland
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089
+ Models
FORECO-10716; No of Pages 14
11
25.76 (5.05)
12.47 (6.30)
31.72 (9.71)
36.14 (10.12)
17.45 (6.60)
3.00 (0.32)
1.31 (0.32)
2.84 (0.50)
3.45 (0.81)
0–50
3.26 (0.66)
4.23 (1.14)
10.18 (3.14)
11.34 (3.78)
2.28 (1.40)
4.61 (3.91)
5.58 (4.73)
4.60 (1.95)
12.25 (5.85)
14.87 (7.50)
6.63 (3.41)
13.23 (6.12)
16.28 (7.31)
3.19 (1.61)
8.04 (5.68)
6.22 (2.96)
0.56 (0.09)
1.16 (0.20)
1.27 (0.23)
0.43 (0.14)
0.44 (0.21)
0.44 (0.20)
0.55 (0.23)
1.09 (0.32)
1.20 (0.31)
0.63 (0.27)
1.20 (0.39)
1.43 (0.45)
0.55 (0.18)
1.39 (0.53)
1.50 (0.58)
Caimito v.
Caimito m.
Bohio v.
Depth (cm)
Table 6
Exchangeable and total Mg stocks (Mg/ha) (mean one-half CI)
0–5
5–25
25–50
Andesite
Andesite
BCI
Total Mg
Exchangeable Mg
Bohio v.
Caimito m.
Caimito v.
BCI
F.K. Barthold et al. / Forest Ecology and Management xxx (2007) xxx–xxx
Table 7
CART results
Response variable
Predictor variables
Cross-validated error
Exchangeable Mg
(mg/kg) in 0–5 cm
Exchangeable Mg
(mg/kg) in 5–25 cm
Exchangeable Mg
(mg/kg) in 25–50 cm
Exchangeable Mg
(mg/kg) in 50–75 cm
Exchangeable Mg
(mg/kg) in 75–100 cm
–
1.02
Soil type
0.77
Soil type
0.67
Soil type
0.53
Soil type
0.67
flow generation and are used to identify spatial patterns of
overland flow. Godsey et al. (2004) though found that overland
flow generates already at shallow slopes and small catchment
areas, primarily along microtopographic features such as
concentrated flow lines and wash areas instead of following
topographic gradients outlined by the DEM. Pipeflow
mechanisms might be responsible for these spatial variations.
If such microtopographic features as concentrated flow lines
and wash areas determine the generation of overland flow rather
than topographic features such as slope, catchment area and
wetness index, then it is also difficult to capture the mechanisms
that are responsible for the spatial patterns of exchangeable
nutrients. This implies that the resolution of our DEM is too
coarse to capture those processes that are controlled by
microtopography.
A careful examination of the box plots in Fig. 3 reveals a
slight trend with higher values of exchangeable K on the steeper
terrains of Bohio volcanic and Caimito marine, but the
differences are not significant from the Andesite and Caimito
volcanic facies. Therefore, it is not surprising that the
Regression Tree Models using lithology and terrain attributes
explain only little to none of the variation of exchangeable K.
Various factors that may explain the weakness of the
relationships between landscape features and exchangeable
Mg have been mentioned above and may be true for
exchangeable K as well.
A more likely possibility for the explanation of the weak
relationships between landscape features and exchangeable K
is that biogeochemical processes have a large impact on the
distribution of exchangeable K in the soil. Potassium is a very
labile element and its biogeochemical cycle has been
investigated and described in many studies (Vitousek and
Sanford, 1986; Likens et al., 1994; Tobón et al., 2004).
Potassium is being leached from green leaves during rainfall
(Wilcke et al., 2001; Tobón et al., 2004). However, a larger
amount of K is stored in the litter fall and returned to soil via
litter leachate (Wilcke et al., 2001; Tobón et al., 2004). Loos
et al. (unpublished results) confirm this process for BCI where
they found that the amount of K that actually reaches the soil
via litter leachate is much larger than what is being leached via
through fall. Most of the K that is recycled in the plant-watersoil system and leaves the living plant is accumulated in the
litter and returns to the soil. Yavitt et al. (2004) reported that
residence time of K is short in the leaf litter and suggest rapid
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089
+ Models
FORECO-10716; No of Pages 14
12
F.K. Barthold et al. / Forest Ecology and Management xxx (2007) xxx–xxx
Fig. 6. Dot plots of total Mg distribution on the four lithologies on Barro Colorado Island. The empty circles represent the value of each sample, the filled circles
represent the median and the crosses represent the lower and upper 95% confidence limits. Sample sizes range from 9 to 12.
Table 8
Summary of rock types and their mineral and chemical composition of the different formations (compiled from Johnsson and Stallard, 1989)
Formation
Rock type
Main minerals
Chemical compositiona
Andesite
Non-vesicular, phenocrystic
andesite with veins and vugs
Plagioclase (feltspar group)
Clinopyroxene
Orthopyroxene
Magnetite
Veins and vugs
Quartz
Calcite
Zeolite
nNa[AlSi3O8] + nCa[Al2Si3O8]
(Ca, Mg, Fe, Al)2(Si, Al)2O6
(Mg, Fe)SiO3
Fe3O4
Bohio volcanic
Caimito marine
Caimito volcanic
a
b
Conglomerate with
basaltic clasts in a sandy
volcaniclastic matrix
Clasts
Plagioclase
Hornblende (amphibole group)
Clinopyroxene
Matrix:
Volcanic lithic fragments
Plagioclase
Amphibole
Magnetite
SiO2
CaCO3
M2/nO*Al2O3*xSiO2*yH2Ob
nNa[AlSi3O8] + nCa[Al2Si3O8]
(Ca, Na, K)2–3(Mg, Fe, Al)5[(OH, F)2(Si,Al)2Si6O22]
(Ca, Mg, Fe, Al)2(Si, Al)2O6
nNa[AlSi3O8] + nCa[Al2Si3O8]
Fe3O4
Foraminiferal limestone
with abundant pelecypods
Vitric volcaniclastic debris
Calcite
Plagioclase
Quartz
CaCO3
nNa[AlSi3O8] + nCa[Al2Si3O8]
SiO2
Volcaniclastic sandstone
(basaltic agglomerate and
tuffaceous graywackes)
Volcanic lithic fragments
Plagioclase
nNa[AlSi3O8] + nCa[Al2Si3O8]
From Berry et al. (1983).
M, alkali or alkaline atom; n, charge on that atom; x, a number from 2 to 10; y, a number from 2 to 7.
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089
+ Models
FORECO-10716; No of Pages 14
F.K. Barthold et al. / Forest Ecology and Management xxx (2007) xxx–xxx
mineralization. Although plants may take up K directly from
leaf litter, the more important mechanism seems to be uptake
from the mineral soil after leaching from the litter layer. We
conclude that K is recycled fast and efficient in the tropical rain
forest ecosystem on BCI and that the importance of topographic
processes for its spatial variation is confounded by these
biogeochemical processes.
Finally, vegetation features might play a more important role
than topography and lithology. We did include forest age as a
variable that characterizes differences in species composition
due to different stages of forest succession, and in forest
structure (Foster and Brokaw, 1996). But these differences are
now rather subtle (Foster and Brokaw, 1996) and therefore
forest age as a variable characterizing forest structure is too
general. For example, Chen et al. (1997) attributed the spatial
variation of exchangeable K to variables such as canopy height,
average basal area and density. Goodland and Pollard (1973)
reported strong correlations between vegetation structure
(height, basal area and density) and soil N, P and K contents
in the cerrado vegetation of Brazil. This suggests that
vegetation features may confound the importance of the
topography.
Studies relating topography and nutrient availability often
show contradictory results. Although a few studies could
establish a link between topography and the spatial pattern of
soil exchangeable cations, e.g. Chen et al. (1997) and Silver
et al. (1994), others could not, e.g. Johnson et al. (2000),
Scowcroft et al. (2004) and Wilcke et al. (2001). Our study
belongs to the latter group. The contradictory results may be
attributed to different processes in response to specific
environmental features discussed in the individual studies.
Scale is likely to be another issue. Chen et al. (1997) and Silver
et al. (1994) conducted their studies at the hillslope and the
catchment scale, with an extent of 10 ha and 12 ha (two
catchments), and sample sizes of n = 40 and n = 87, respectively. In contrast, Johnson et al. (2000) study was conducted at
the catchment scale with an extent of 214 ha and a sample size
of n = 72, and our study was conducted at the landscape scale
with an extent of 1500 ha and 108 samples. The ratio of sample
size to extent is much larger in the former two studies than in
the latter.
Our results clearly show that topography does not control the
spatial variation of exchangeable K and Mg in the tropical
forest soilscape of Barro Colorado Island at the observed scale.
Though, it is not impossible, that microtopography or the
topography at a scale smaller than the one we captured with our
DEM might be important for the spatial distribution of the
observed soil chemical properties.
5. Conclusions
We investigated the spatial distribution of total and
exchangeable magnesium and potassium and its relationship
to terrain attributes at the landscape scale. This relationship
turned out to be weak for various possibilities, such as
microtopography (in the case of exchangeable Mg), whose
features go undetected at the resolution of our DEM, or
13
biogeochemical processes unrelated to topography (in the case
of exchangeable K). Fast nutrient recycling may well mask any
effects of topography, which would explain the lack of terrain
attributes’ predictive power of spatial patterns. The spatial
distribution of total cation concentrations does not depend on
the topography, but on lithology only. The apparent lack of a
strong relationship between terrain attributes and the selected
soil properties suggests that rather simple sampling designs are
adequate to estimate the spatial mean of these and related
nutrients on Barro Colorado Island and similar soilscapes. Our
results prompt us to view critically the predictive power of
terrain attributes for soil properties at the observed scale.
Acknowledgments
Frauke Barthold acknowledges support by the German
Academic Exchange Service (DAAD). The authors thank Ian
Baillie for helpful suggestions and the STRI staff on BCI for the
logistical, technical and administrative support.
References
Baillie, I.C., 1989. Soil characteristics and classification in relation to the
mineral nutrition of tropical wooded ecosystems. In: Proctor, J. (Ed.),
Mineral Nutrients in Tropical Forest and Savanna Ecosystems. Blackwell
Scientific Publications, Oxford, pp. 15–26.
Baxter, S.J., Oliver, M.A., 2005. The spatial prediction of soil mineral N and
potentially available N using elevation. Geoderma 128, 325–339.
Berry, L.G., Mason, B., Dietrich, R.V., 1983. Mineralogy. W.H. Freeman, San
Francisco.
Blair, B.C., 2005. Fire effects on the spatial patterns of soil resources in a
Nicaraguan wet tropical forest. J. Trop. Ecol. 21, 435–444.
Breiman, L., Friedman, J.H., Olshen, R.A., Stone, C.J., 1984. Classification and
Regression Trees. Wadsworth, Belmont, CA.
Brus, D.J., DeGruijter, J.J., 1997. Random sampling or geostatistical modelling? Choosing between design-based and model-based sampling strategies
for soil. Geoderma 80, 1–59.
Burghouts, T.B.A., Van Straalen, N.M., Bruijnzeel, L.A., 1998. Spatial heterogeneity of element and litter turnover in a Bornean rain forest. J. Trop. Ecol.
14, 477–505.
Chen, Z.-S., Hsieh, C.-F., Jiang, F.-Y., Hsieh, T.-H., Sun, I.-F., 1997. Relations
of soil properties to topography and vegetation in a subtropical rain forest in
southern Taiwan. Plant Ecol. 132, 229–241.
Coates, A.G., Obando, J.A., 1996. The geologic evolution of the Central
American Isthmus. In: Jackson, J.B.C., Budd, A.F., Coates, A.G. (Eds.),
Evolution and Environment in Tropical America. University of Chicago
Press, Chicago, pp. 21–56.
Croat, T.B., 1978. Flora of Barro Colorado Island. Stanford University Press,
Stanford, California.
De’ath, G., Fabricius, K.E., 2000. Classification and Regression Trees: a
powerful yet simple technique for ecological data analysis. Ecology 81,
3178–3192.
Dietrich, W.E., Windsor, D.M., Dunne, T., 1982. Geology, climate and hydrology of Barro Colorado Island. In: Leigh, E.G.J., Rand, A.S., Windsor, D.M.
(Eds.), The Ecology of a Tropical Forest: Seasonal Rhythms and Long-term
Changes. Smithsonian Institution Press, Washington, D.C., pp. 21–46.
Ellert, B.H., Bettany, J.R., 1995. Calculation of organic matter and nutrients
stored in soils under contrasting management regimes. Can. J. Soil Sci.
529–538.
Florinsky, I.V., Eilers, R.G., Manning, G.R., Fuller, L.G., 2002. Prediction of
soil properties by digital terrain modelling. Environ. Model. Software 17,
295–311.
Foster, R.B., Brokaw, N.V.L., 1996. Structure and history of the vegetation of
Barro Colorado Island. In: Leigh, E.G., Rand, S., Windsor, D.M. (Eds.),
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089
+ Models
FORECO-10716; No of Pages 14
14
F.K. Barthold et al. / Forest Ecology and Management xxx (2007) xxx–xxx
The Ecology of a Tropical Forest: Seasonal Rhythms and Long-term
Changes. Smithsonian Institution, Washington, DC, pp. 67–82.
Gessler, P.E., Moore, I.D., McKenzie, N.J., Ryan, P.J., 1995. Soil-landscape
modelling and the spatial prediction of soil attributes. Int. J. Geogr. Inf. Syst.
95, 421–432.
Godsey, S., Elsenbeer, H., Stallard, R., 2004. Overland flow generation in two
lithological distinct rainforest catchments. J. Hydrol. 295, 276–290.
Golley, F.B., 1986. Chemical plant–soil relationships in tropical forests. J. Trop.
Ecol. 2, 219–229.
Goodland, R., Pollard, R., 1973. The Brazilian cerrado vegetation: a fertility
gradient. J. Ecol. 61, 219–224.
Heinrichs, H., Hermann, A.G., 1990. Praktikum der Analytischen Geochemie.
Springer Verlag, Berlin-Heidelberg-New York.
Helmke, P.A., Sparks, D.L., 1996. Lithium, sodium, potassium, rubidium, and
cesium. In: Methods of Soil Analysis. Part 3. Chemical Methods, Soil
Science Society of America and American Agronomy Society, Madison,
WI, pp. 551-574.
Henderson, B.L., Bui, E.N., Moran, C.J., Simon, D.A.P., 2005. Australiawide predictions of soil properties using decision trees. Geoderma 124,
383–398.
Holdridge, L.R., Budowski, G., 1956. Report of an ecological survey of the
Republic of Panama. Carribean Forester 17, 92–110.
Holmes, K.W., Kyriakidis, P.C., Chadwick, O.A., Soares, J.V., Roberts, D.A.,
2005. Multi-scale variability in tropical soil nutrients following land-cover
change. Biogeochemistry 74, 173–203.
Huntington, T.G., Johnson, C.E., Johnson, A.H., Siccama, T.G., Ryan, D.F.,
1989. Carbon, organic matter, and bulk density relationships in a forested
spodosol. Soil Sci. 148, 380–386.
Hutchinson, M.F., 1989. A new method for gridding elevation and stream line
data with automatic removal of pits. J. Hydrol. 106, 211–232.
Jenny, H., 1941. Factors of Soil Formation – A System of Quantitative
Pedology. McGraw-Hill, New York.
Johnson, C.E., Ruiz-Méndez, J.J., Lawrence, G.B., 2000. Forest soil chemistry
and terrain attributes in a Catskills watershed. Soil Sci. Soc. Am. J. 64,
1804–1814.
Johnsson, M.J., Stallard, R.F., 1989. Physiographic controls on the composition
of sediments derived from volcanic and sedimentary terrains on Barro
Colorado Island, Panama. J. Sediment. Petrol. 59, 768–781.
Likens, G.E., Driscoll, C.T., Buso, D.C., Siccama, T.G., Johnson, C.E., Lovett,
G.M., Ryan, D.F., Fahey, T., Reiners, W.A., 1994. The biogeochemistry of
potassium at Hubbard Brook. Biogeochemistry 25, 61–125.
Mackensen, J., Tillery-Stevens, M., Klinge, R., Fölster, H., 2000. Site-parameters, species composition, phytomass structure and element stores of a
terra-firme forest in East-Amazonia, Brazil. Plant Ecol. 151, 101–119.
Markewitz, D., Davidson, E.A., Figueiredo, R.D.O., Victoria, R.L., Krusche,
A.V., 2001. Control of cation concetrations in stream waters by surface soil
processes in an Amazonian watershed. Nature 410, 802–805.
McBratney, A.B., Mendonca Santos, M.L., Minasny, B., 2003. On digital soil
mapping. Geoderma 117, 3–52.
McKenzie, N.J., Gessler, P.E., Ryan, P.J., O’Connell, D.A., 2000. The role of
terrain analysis in soil mapping. In: Wilson, J.P., Gallant, J.C. (Eds.),
Terrain Analysis – Principles and Applications. John Wiley & Sons,
Inc., pp. 245–265.
Miller, F., 1927. Special Map of Barro Colorado Island. In. Engineer Production
Plant, Panama Canal Department, U.S. Army, Corozal, Canal Zone.
Milne, G., 1935. Some suggested units of classification and mapping particularly for East African soils. Soils Res. 4.
Moore, I.D., Gessler, P.E., Nielsen, G.A., Peterson, G.A., 1993. Soil attribute
prediction using terrain analysis. Soil Sci. Soc. Am. J. 57, 443–452.
Moore, I.D., Grayson, R.B., Landson, A.R., 1991. Digital terrain modelling: a
review of hydrological, geomorphological, and biological applications.
Hydrol. Processes 5, 3–30.
Myles, A.J., Feudale, R.N., Liu, Y., Woody, N.A., Brown, S.D., 2004. An
introduction to decision tree modeling. J. Chemom. 18, 275–285.
O’Callaghan, J.F., Mark, D.M., 1984. The extraction of drainage networks
from digital elevation data. Comp. Vision Graphics Image Processing 28,
323–344.
Park, S.J., Burt, T.P., 2002. Identification and characterization of pedogeomorphological processes on a hillslope. Soil Sci. Soc. Am. J. 66, 1897–1910.
R Development Core Team, 2005. R: A Language and Environment for
Statistical Computing. R Foundation for Statistical Computing.
Rand, A.S., Rand, W.M., 1982. Variation in rainfall on Barro Colorado Island.
In: Leigh, E.G., Rand, S., Windsor, D.M. (Eds.), The Ecology of a Tropical
Forest: Seasonal Rhythms and Long-term Changes. Smithsonian Institution
Press, Washington, D.C., pp. 47–59.
Richter, D.D., Babbar, L.I., 1991. Soil diversity in the tropics. Adv. Ecol. Res.
21, 316–389.
Sánchez, P.A., 1976. Properties and Management of Soils in the Tropics. Wiley,
New York.
Scowcroft, P.G., Haraguchi, J.E., Hue, N.V., 2004. Reforestation and topography affect montane soil properties, nitrogen pools, and nitrogen transformations in Hawaii. Soil Sci. Soc. Am. J. 68, 959–968.
Silver, W.L., Scatena, F.N., Johnson, A.H., Siccama, T.G., Sanchez, M.J., 1994.
Nutrient availability in a montane wet tropical forest: spatial patterns and
methodological considerations. Plant Soil 164, 145–164.
Soil Survey Staff, 1996. Soil Survey Laboratory Methods Manual. In. National
Soil Survey Center, Natural Resources Conservation Service, U.S. Department of Agriculture, Lincoln, NE.
Soil Survey Staff, 2006. Keys to Soil Taxonomy. U.S. Department of Agriculture. Natural Resource Conservation Service, Washington, D.C..
Stanley, S.M., 1999. Earth System History. W.H. Freeman and Company, New
York, N.Y., USA.
Svenning, J.-C., Kinner, D.A., Stallard, R.F., Engeblbrecht, B.M.J., Wright,
S.J., 2004. Ecological determinism in plant community structure across a
tropical forest landscape. Ecology 85, 2526–2538.
Tarboton, D.G., 1997. A new method for the determination of flow directions
and contributing areas in grid digital elevation models. Water Resour. Res.
33, 309–319.
Taylor, J.R., 1997. An Introduction to Error Analysis – The Study of Uncertainties in Physical Measurements. University Science Books, Sausalito,
Ca., USA.
Therneau, T.M., Atkinson, B., 2005. R port by Brian Ripley. rpart: Recursive
Partitioning. R package version 3.1–23.
Tobón, C., Sevink, J., Verstraten, J.M., 2004. Litterflow chemistry and nutrient
uptake from forest floor in northwest Amazonian forest ecosystems. Biogeochemistry 69, 315–339.
Vitousek, P.M., Sanford, R.L., 1986. Nutrient cycling in moist tropical forest.
Ann. Rev. Ecol. Syst. 17, 137–167.
Wilcke, W., Yasin, S., Valarezo, C., Zech, W., 2001. Change in water quality
during the passage through a tropical montane rain forest in Ecuador.
Biogeochemistry 55, 45–72.
Wilson, J.P., Gallant, J.C., 2000. Digital terrain analysis. In: Wilson, J.P.,
Gallant, J.C. (Eds.), Terrain Analysis - Principles and Applications. John
Wiley & Sons, Inc., pp. 1–26.
Woodring, W.P., 1958. Geology of Barro Colorado Island, Canal Zone. Smithsonian Miscellaneous Collections 135, 1–39.
Yavitt, J.B., 2000. Nutrient dynamics of soil derived from different parent
material on Barro Colorado Island, Panama. Biotropica 32, 198–207.
Yavitt, J.B., Wieder, R.K., 1988. Nitrogen, phosphorus, and sulfur properties of
some forest soils on Barro Colorado Island, Panama. Biotropica 20, 2–10.
Yavitt, J.B., Wright, S.J., 1996. Temporal patterns of soil nutrients in a
Panamanian moist forest revealed using ion-exchange resins and experimental irrigation. Plant Soil 183, 117–129.
Yavitt, J.B., Wright, S.J., Wieder, R.K., 2004. Seasonal drought and dry-season
irrigation influence leaf-litter nutrients and soil enzymes in a moist, lowland
forest in Panama. Aust. Ecol. 29, 177–188.
Zevenbergen, L.W., Thorne, C.R., 1987. Quantitative analysis of land surface
topography. Earth Surf. Processes Landforms 12, 47–56.
Please cite this article in press as: Barthold, F.K., et al., Soil nutrient–landscape relationships in a lowland tropical rainforest in Panama, Forest
Ecol. Manage. (2007), doi:10.1016/j.foreco.2007.09.089