Background

Amazonia contains some of the most productive ecosystems in the world, experiences notable patterns of climate variability on seasonal to interrannual time-scales, and is undergoing significant changes in landuse. It is thus a region with important fluxes and changes in fluxes of carbon, nutrients, and water. Regional-scale ecosystem modeling, database development and remote sensing analysis will be important tools for evaluating potential consequences of land-use change on the biogeochemistry in Amazonia. Our group has significant experience in modeling of natural and human-managed ecosystems, remote sensing, GIS analysis, and large database assembly and dissemination. In this Background Section, we review our relevant experience.

II.1. Analysis of natural ecological systems

Our earlier work1 in Amazonia showed the dependence of NPP on water availability. The recycling of local evaporation and precipitation is well-accepted,1 and as large areas of the basin are subject to active deforestation there is a concern about how such land surface disruptions will affect the water cycle in the tropics2. Interannual variability in precipitation leads to large variations in downstream hydrographs3 4(Fig. 3), NPP, and NEP (Fig. 2). The capability to model and understand changes in the Amazon regional water system is important to studies of terrestrial productivity, and also the significant effects of floodplain inundation on trace gas emissions.

II.2. Studies of managed and disturbed ecosystems

We are currently developing a model for landcover and landuse change. The model consists of: 1) a land evaluation module that assesses the suitability and availability of land for crops and pasture based on biophysical constraints such as climate, soil and topography; and 2) a landuse module, GEOMOD,5 for simulating spatial patterns of landuse/landcover as well as the rates of change from landscape to regional scales based upon biophysical factors and socio-economic factors (e.g., population density, land tenure system, timber price). Algorithms within GEOMOD represent the principles of adjacency, dispersion, regional heterogeneity, relative growth, energy efficiency, and resource quality. The land evaluation module has been applied globally to examine the impact of different transient climate change scenarios in the 21st century on the availability and suitability of land for crop cultivation.6 GEOMOD has been used to simulate spatial distribution of landuse/landcover change in the Tropics over time at a regional scale (South and Southeast Asia, Africa and middle America) and at a landscape scale in Costa Rica (Fig. 4), northern Thailand, peninsular Malaysia and northeastern China.10

Conversion of land to pasture and cropland has had a significant impact on the biogeochemistry of many areas of Amazonia, and these ecosystems must be included in any regional analysis. The DNDC model7 can be used to simulate carbon and nutrient biogeochemistry in agroecosystems. DNDC has simulated 30 years of pasture biogeochemistry following conversion of forest to pasture. Estimated SOM and N2O fluxes are in general accord with a chronosequence study in Costa Rica (Fig. 5a,b) and simulation of N2O flux from fertilized maize grown on recently cleared land is also in agreement with measurement (Fig. 5c). DNDC has successfully simulated N2O fluxes and soil organic matter (SOM) dynamics in temperate and subtropical regions.8 DNDC has been used to simulate N2O flux from agricultural lands in the US,9 and is currently being used to evaluate N2O emissions from agricultural lands in China, including wet subtropical areas. Tropical agriculture and pasture simulations with DNDC have been focused on Costa Rica (Fig. 5), and will be adapted to the LBA region.

II.3. Management of data and GIS capabilities

Our NASA EOS-IDS and Hughes Applied Information Systems are collaborative partners in a prototyping project that will provide full GIS capability over the internet and is fully interoperable with NASA's EOS Data and Information System Core System (ECS). One aspect of this effort is to provide a means for LBA participants to conduct geospatial data searches and queries via the internet. We call this initial prototype the UNH EOS Explorer. The Explorer is a world-wide-web-based GIS (http://www.unh-ecs.sr.unh.edu). Specific components of the system are a Java client and server, a Spatial Data Engine (SDE/ESRI10) client and server, and an Oracle database.

In preparation for the LBA research, our effort has focused on data covering Brazil's Legal Amazon region. Currently, the Explorer contains a total of 37 data layers (Table 1). Twelve layers are satellite-based and originate from several archive centers: the EROS11 Data Center (EDC), EOSAT, Landsat Ground Station Operator Working Group (LGSOWG) and Spot Image Corp. The satellite footprint data were made available through the efforts of the Humid Tropical Forest Inventory Project (HTFIP) at UNH which is part of NASA's Landsat Pathfinder Project. Four additional data layers were also made available by the HTFIP group and our EOS-IDS research effort: Soils, IBGE vegetation, RADAM vegetation, and counties. The soils, IBGE vegetation, and RADAM vegetation layers were originally acquired from the Brazilian Ministry of Mines and Energy and the Brazilian Ministry of Agriculture. There are also 10 Hydro-Climatology layers provided by the UNH Global Hydrology Research Group. These data were obtained from the Global Hydrological Archive and Analysis System (GHAAS) database which was also developed, in part, by our EOS-IDS efforts. An additional set of data layers includes several types of land cover and topology coverage data from the Digital Chart of the World (DCW) archive from ESRI.12

II.4. Remote sensing

Issues related to validating spatial patterns of model predictions have been explored using remote sensing. We examined model results of the Vegetation/Ecosystem Modeling and Analysis Project of net primary productivity (NPP),13 estimated in the conterminous USA at a spatial resolution of 0.5o by 0.5o grids3. One goal was to check the realism of the spatial variability of model estimates using long-term monthly mean NDVI for each 0.5x0.5o grid cell, using the fact that NPP and NDVI are linearly related. Correlations for the entire domain were relatively high (R2=0.6-0.7). However, comparison of the mean deviates of both NDVI and simulated NPP (each grid cell value subtracted from the mean of all grid cell values in an ecosystem type) were uncorrelated within biomes. Thus the models appeared to be representing across-biome patterns of NPP, but no conclusion could be made about within-biome variability. This type of analysis is potentially very powerful for evaluating model-predicted NPP (Fig. 6).

A remotely sensed vegetation index can also yield insight into patterns of response of ecosystems to climate, providing a means to evaluate model response and model-based hypotheses. We14 used global AVHRR data and gridded air temperature from the Microwave Sounding Unit to estimate the magnitude of immediate and lagged response to temperature (Fig. 7). Patterns of zero-, one-, and two-year lagged responses of NDVI to temperature variability were ecosystem dependent and consistent with the hypothesis that biogeochemical mechanisms play an important role in mediating global relationships between CO2 and temperature.

We have successfully retrieved canopy biophysical variables, including the fraction of PAR absorbed by the canopy (fAPAR) and albedo, using a radiative transfer model and AVHRR data for a transect in the Central African Republic.15 Though observations of a single pixel have a single sun-sensor geometry, we gathered neighboring pixels in temporally composited scenes, having similar functional ecosystem type, in order to simulate a multiple sampling of geometry within ~0.5x0.5o "cells". With the advent of MISR data from EOS AM-1, we expect both improved accuracy and the ability to perform inversions using much smaller spatial windows.