Close Window

ND-01 Abstract

Land Cover Conversion in Amazonia, the Role of Environment and Substrate Composition in Modifying Soil Nutrient Cycling and Forest Regeneration

Getulio T. Batista — Universidade de Taubate (SA-PI)
Oliver A. Chadwick — University of California (US-PI)
Dar A. Roberts — University of California (US-PI)


We will integrate remote sensing, GIS and field sampling of soil and vegetation to

develop predictive models relating land cover change to its effect on nutrient cycling

along the environmental gradients proposed as LBA transects. We predict that differences

in substrate composition and environment will drive distinctly different rates of

regeneration and affect the duration of pasture use, thus requiring different approaches

for sustainable management. We will conduct time-series analysis of land-cover change

using Landsat MSS and TM combined with radar to enhance discrimination of pasture and

early stages of forest regeneration. Our initial focus will be in Rondonia and Marabá,

where historical satellite data have been assembled and analyzed, and where a large body

of process level soils and vegetation data already exist. Site selection for soil

biogeochemical sampling will be driven by explicit, rule-driven mapping of terrain

attributes (initially from digitized topographic maps) and land-cover/land-use maps

derived from remote sensing. Soils and vegetation will be sampled for laboratory analysis

of cations, exchange properties, carbon, nitrogen, bulk density, wet chemical extraction

of P from different pools, and possibly 87Sr/86Sr isotopic ratios

for nutrient provenance. We will develop mechanistic models that relate soil nutrient

composition and dynamics to soil. At each sampling locality the effect of land cover

conversion and subsequent management will be documented by analysis of N, P , mineral

composition and cation pools existing in soils on comparable landforms. Analyzed along the

transects, these data will allow development of a mass balance understanding of

weathering, mineral transformation and the nutrient supplying status of Amazon basin

ecosystems as affected by environment and substrate.

Our long-term objective is to develop empirical/predictive biogeochemical models with a

focus on nutrient dynamics linked to environmental gradients and land cover change. We

will predict and test our models using current/future remotely sensed data sets. These

models and observations will be critical to understanding ecosystem dynamics as influenced

by land-use decisions and should contribute to sustainable management practices in the

region. Our research will contribute directly to the larger LBA objectives and will be

informed by results from other team members.

General Strategy

We will be operating over a cascading hierarchy of scales. The broadest scale being the

Amazon basin which will be segmented by the LBA transect structure. The next scale will

focus on the region around the intensively studied tower sites, which we expect will be

underlain by a specific geologic formation. Within the intensive study sites we will

conduct further stratification by two methods. First, we will use digital topography data

to calculate terrain attributes that will allow us to segment the study sites into

landforms and landform elements that are grouped according to similar statistical and

physiographic properties. At the most detailed level of separation (in this case,

determined by data resolution), we will develop variograms to determine the spacing

required for collection of uncorrelated samples. We will then be able to collect random

samples at that level. This explicit, statistically based sampling scheme allow upward

scaling of results based on topography which is extremely important but we will have to

overcome data limitation problems through digitization of paper maps until topographic

data are generated by airborne campaigns or satellite. Second, we will develop land cover

maps and land use history maps using remote sensing data. Our remote sensing strategy is

to build on our preexisting experience mapping land-cover and land-cover change in Manaus

and Maraba to extend these capabilities to Rondonia and other regions along the transect.

We have already assembled all of the necessary optical data for Rondonia and some SIR-C

data. We will pursue additional radar data, such as JERS-1 data sets being assembled by

the Jet Propulsion Laboratory for much of the basin. We have already developed the

programs and obtained the computer facilities that will enable us to generate land-cover

and land-use histories rapidly in any region of the basin where data are available. Once

the topographic hierarchy and land cover/land use history maps are completed, we can

compare the two to determine the landscape pattern of land conversion and to ensure that

the sampling scheme adequately covers all land cover types.

We will focus detailed sampling on comparisons among primary forest and various land

cover classes. As other researchers have done, we expect to make intensive use of pasture

chronosequences, which will give us a time history of biogeochemical change. Finally, we

will compare biogeochemical behavior in and between different land cover types across the

LBA transects. We anticipate that as LBA our soil biogeochemistry data will be useful to

plant ecologists and other biogeochemists on the LBA Science Team and expect to adjust

sampling priorities to meet the overall needs of the LBA Team.

Remote Sensing Strategy

The remotely sensed methodologies we will employ build heavily off of seven years of

prior experience working in several areas within the Amazon basin. We will employ spectral

mixture models using reference end-members as described by Adams et al., (1993, 1995) and

Roberts et al., 1997a. Through past research in the area we have already developed a

spectral library that is applicable to much of the basin. Where new leaf level and canopy

level spectra become available they will be incorporated into our existing libraries.

We will use a binary decision tree approach to classify spectral mixture models from

Landsat TM and MSS into at least seven classes, including primary forest, second growth,

pasture, water, construction (roads/urban), recently burned, and cloud/smoke obscured. We

will employ multi-temporal techniques described by Roberts et al, (1997a) to subdivide

pasture and second growth forest into age classes to establish rates of pasture

maintenance and regeneration. Potential for regeneration and sustainable land use will

vary as some function of intensity of pasture use and age. Images will be screened for

clouds using techniques described by Roberts et al. (1997a). Classification errors can be

reduced through use of time series as well. Difficulties will occur in separating green

pastures and some types of agriculture from second growth forest. This problem is expected

to be particularly significant in Rondonia, which has extensive pastures, croplands and

some plantations. To facilitate mapping we will explore use of L band radar, using SIR-C

data where available (such as Rondonia) and JERS-1 in other regions.

If new remotely sensed optical and microwave data sets become available during the

course of LBA we will take advantage of them opportunistically. For example, we have

extensive experience in the analysis of AVIRIS data for mapping vegetation. We have

already obtained and analyzed some AVIRIS data collected in the state of Rondonia during

the SCAR-B program. If enhanced TM or LEWIS HSI data become available we will use them as

part of our strategy for testing predictions for regeneration success and explore their

capabilities for improvements in spectral libraries and vegetation classification.

Landscape Modeling, Site Selection, and Scaling of Results

A suite of Earth processes combine with land use to act as driving variables that

determine the present state of soil biogeochemistry in Amazon ecosystems. The importance

of individual controls is scale dependent. At the physiographic region level, gradients in

climate coupled with underlying geology drive the properties and processes of soils and

ecosystems. Rainfall may be seasonally limiting in parts of Amazonia; at the same time

nutrients may be conserved relative to the wetter regions. The nature of soil parent

material and the depth to fresh rock (containing nutrients) are significant determinants

of biogeochemical functioning. More locally where climate and geology are relatively

uniform, topographic variation controls soil biogeochemical functioning because of

redistribution of water and soluble nutrients along flow paths. Topography modifies Human

land use in predictable ways (i.e. certain landforms will be utilized more than others).

In this proposal we will quantify variations in soil biogeochemistry as affected by

natural and anthropogenic processes acting at these different scales.

The spatial distribution of soil properties (soil-landscape) is part of the overall

ecosystem and is a primary control on nutrient dynamics. A soil-landscape is a complex

geomorphologic, biogeochemical and hydrological system which controls the composition and

productivity of ecosystems and their response to disturbance. The characteristics of the

soil-landscape provide essential information for modeling the pools, pathways and fluxes

of water and nutrients. The key components for determining soil biogeochemical properties

are quantitative spatial models that include, but are not limited to soil properties such

as A-horizon depth, overall soil depth, clay content, cation exchange capacity, organic

matter content, available nutrients, nutrient supplying capacity and bulk density.

Topographic variation within the soil-landscape system modifies ecosystem functioning by

changing fluxes of water, energy and mass. Long-term differences in fluxes can be inferred

from differences in measured soil properties. The spatial distribution of soil properties

drive the location of monitoring sites for short-term ecosystem-specific fluxes within the

soil-landscape system such as water balance and nutrient availability.

Although it is difficult to measure below ground soil-landscape characteristics, we can

develop statistical relationships between soil properties and terrain attributes such as

slope, aspect, hill-slope position, and slope shape. Our approach, developed by Paul

Gessler and others (Gessler et al. 1995, McSweeney et al., 1994, Moore et al., 1993),

integrates a geographic information system (GIS), digital terrain modeling, and

statistical modeling to quantify relationships between soil properties and terrain

attributes, which are calculated from continuous digital elevation models using GIS map

algebra tools. Terrain attributes are inexpensive to map over large areas and

quantitatively describe the distribution of landscape processes (e.g. water flow

convergence/divergence, sediment and solute transport, solar radiation). The distributions

of terrain attributes can be used to design sampling schemes that capture the full range

of landscape variation and to predict soil biogeochemical characteristics. By then

sampling across the range of terrain attributes, we build statistical models relating

landscape structure to soil biogeochemical properties. These models allow explicit

aggregation of point measurement data or simulation model output (based on point source

input data) to landscape scales.

Implementation procedures in Amazonia are as follows:

  1. Work in conjunction with the science team to identify priority physiographic provinces

    to cover a range of climate and substrate. We expect to start in Rondonia as part of the

    mesoscale catchment study and the first tower cluster along western transect

  2. Develop digital elevation models from various sources: digitized 1:100,000 scale maps

    being competed by Tom Dunne's research team (we expect to digitize maps as well), new

    topographic mapping technology such as TOPSAR, Global Topographic Mapping Mission and

    laser altimetry, and semi-quantitative measures of topographic attributes using remote

    sensing data

  3. Calculate (Moore et al., 1993) and map (Gessler et al., 1995) terrain attributes that

    are appropriate to particular landscapes (we expect to be hampered the lack of data at

    appropriate scales)

  4. Use variogram analysis to determine minimum distances between sample sites required for

    random sampling of soil and biogeochemical properties

  5. Overlay land use and land cover classes on the terrain attributes to determine

    preferential locations of land cover change

  6. Modify topographically driven soil biogeochemistry sampling plan to ensure that all land

    cover types are represented.

Soil Biogeochemical Sampling and Analysis

At each site we will characterize the weathering and nutrient status of soils by

sampling 3 detailed profiles further augmented by randomly distributed core samples to

document soil variability. Core sample site selection will be driven by the statistical

analysis of digital elevation models described above. Soil profiles will be described

according to USDA specifications (Soil Survey Staff, 1992). Channel samples of each

horizon will be taken for analysis of total elements, mineral composition, bulk density,

pH, exchange capacity, exchangeable cations, cations in soil solution, phosphorous,

carbon, and nitrogen.

Weathering and elemental losses from soils will be studied. Soil weathering leads to

significant leaching losses of the soil-derived nutrient elements (K, Ca, Mg, P), Si, and

even Al. It is critical to document the weathering state of the different substrates

because it is indicative of the nutrient reserve for the ecosystem. The amount of

elemental loss relative to the parent material provides a measure of the weathering state

of a soil profile and its present nutrient supplying capacity. Recent development of

physio-chemical strain gauge based on mass balance allows the use of accumulated strain to

calculate chemical gains and losses (Chadwick et al., 1990). An immobile element, i, (e.g.

Zr) is used to compute the change in volume during pedogenesis as follows:

i,w = (pCi,p/wCi,w) - 1 (1)

Where is dry bulk density (g cm-3), C is element mass (wt. %), w is

weathered horizon, and p is parent material. Negative values represent collapse of the

profile. Once the bulk strain is computed, chemical mass gains and losses of a mobile

element, j, per unit volume of parent material, j,w (g cm-3) are calculated

with eq. 2:

j,w = (wCj,w(i,w + 1) - pCj,p)/100. (2)

Elemental analyses will be made by inductively coupled plasma spectrometry on fused or

acid (H2SO4, HNO3, & HF) dissolved soil material for

all major rock forming elements.

Nutrient and mineralogical distribution within soils. In addition to the total mass of

elements, selective wet chemical extractions are used to document the form that an element

is in: exchangeable, organically bound, bound in secondary minerals, bound in primary

minerals. Exchange bases are determined using NH4OAC extraction, exchange Al by

KCl extraction and exchange capacity by the sum of bases and Al (Soil Survey Staff, 1992).

Soil organic C and N will be determined using a Carlo Erba CNS analyzer; Standard

procedures will be used for bulk density and pH (Soil Survey Staff, 1992).

The quantity, chemical forms, and availability of soil phosphorus are particularly

important. The dominant forms of phosphorus vary systematically with soil mineralogical

and chemical properties (Walker and Syers, 1976; Cole and Heil, 1981; Schimel et al.,

1985; Smeck, 1985, Cross and Schlesinger, 1995). That variation often regulates carbon and

nitrogen dynamics in the long term. We propose to evaluate the forms of soil P using

Tiessen and Mohr's (1993) modification of the Hedley et al. (1982) sequential extraction

procedure. This procedure yields P fractions that are operationally defined, but

experience has shown that they correlate quite well with Walker and Syers (1976)

conceptually defined pools of P (Crews et al., 1995).

Mineral composition which controls cation exchange capacity and P sorption properties

will be determined using X-Ray Diffraction, and for selected samples, a series of

sequential extractions (Chadwick et al., 1990; Chadwick et al., 1994). For each of these

samples, gravimetric weight loss is measured after each treatment. Selective dissolution

procedures are as follows:

  1. removal of organo-metal complex with peroxide (Wada, 1989)

  2. removal of non-crystalline allophane, imogolite, and ferrihydrite using acid ammonium

    oxalate in the dark (McKeague and Day, 1966)

  3. removal of crystalline goethite and hematite by Na-dithionite and Na-citrate using the

    Blakemore and others (1987) modification of Holmgren (1967) procedure

  4. removal of the poorly crystalline aluminosilicates and gibbsite using (Jackson and

    others, 1986)

  5. removal of kaolin by heating the residue to 500 °C followed by a 0.5 M NaOH extraction.


Quantification of nutrient provenance is important to understand biogeochemical

dynamics. Strontium isotopes can be used to determine the provenance of calcium and

quantify the relative amounts contributed by silicate weathering and atmospheric dust and

rain (Capo et al., in press; Stewart et al., in press; Graustein, 1989). In order to

interpret 87Sr/86Sr values, clear differences must exist between the

87Sr/86Sr values of soil parent material and atmospheric input. At

this time we do not know if we can apply the technique uniformly across Amazonia. As part

of this project we will collect preliminary samples for analysis and if warranted we will

propose a parallel study designed to identify the nature of nutrient recharge to the

Amazon ecosystems.

Plant and Litter Nutrients. The nutrient status of plants and the distribution

of species in a landscape provide a natural measure of the nutrient supplying capability

of the soil where they are growing. The youngest fully-expanded stem of mature, sun leaves

will be analyzed for leaf mass per area, N, P, base cations, Fe and Al. We will measure N

mineralization using intact core incubations and N and P availability using resin bags. We

anticipate that leaves taken from trees growing on from nutrient rich soils will have

enhanced P concentrations compared with leaves from nutrient poor soils. This approach is

not without problems - element storage within vegetation and dilution of element

concentrations by growth as well as soil nutrient availability affect foliar

concentrations (Chapin et al., 1990).

The selection of measurements is based on data required to develop mechanistic models

of soil biogeochemical processes. We have over 20 years of experience interpreting the

processes that lead to measured properties. Although developing mechanistic models is our

major focus, a subset of these data can also be used as input to ecosystem simulation

models such as CENTURY (Parton et al., 1988). We can run CENTURY simulations to compare

with our mechanistic understanding and contribute the edaphic input parameters to other

members of the LBA Science Team.

Close Window