Modeling Amazon Region Productivity: Including the Effects of Land Use and Land Cover Change

San Francisco, California, December 5, 1998

Christopher Potter (Meeting Chair) introduced the meeting.

Michael Keller (Project Scientist for NASA LBA-Ecology) welcomed workshop participants and reviewed new developments for the LBA project, including the prediction that there should be an international agreement for LBA-Ecology before the end of the year.

M. Keller also mentioned that, because the LBA-Ecology researchers are not yet in the field, there still may be flexibility in their measurement plans if important modeling data needs are identified during this meeting. Keller reminded the group about two constraints with respect to data availability in LBA. First, there is a legal requirement that all the data must be delivered to Brazil within a year of collection. Second, as a NASA requirement, data sets are to be made ‘public’ once they are adequately quality checked. Both these constraints have inherent uncertainties. Therefore the need for close partnerships in LBA was identified. Modelers will need to remain open about what data they are using and how they are using them, making model results available in an expedient manner.

C. Potter summarized the purpose of the workshop meeting. A workshop on ‘Modeling Amazon Region Productivity’ clearly fits in the overall scientific objectives of the LBA project. A major goal for LBA is to increase scientific understanding, through field and modeling studies, of how Amazonia currently functions as a regional entity, and of how this function is affected by land use change. A more specific LBA objective is to quantify, understand and model the physical, chemical and biological processes controlling the energy, water, carbon, trace gas, and nutrient cycles found within Amazonia and to determine how these link to the atmosphere. A related objective is to quantify and model how the energy, water, carbon, trace gas and nutrient cycles respond to deforestation, agricultural practices, and other land use changes, and how these responses are influenced by climate. Modeling regional productivity implies synthesis of observational data over large scales. This requires carbon, water, nutrient, and energy flux information from individual study sites to be integrated across the region.

Hence, the objectives of this modeling workshop were outlined as:

  1. Strengthen communication between LBA-Ecology modeling groups.

  2. Determine connections between ecosystem models and available observational data for regional productivity estimates.

  3. Address specific science issues for modeling regional productivity -- e.g. data gaps, validation techniques, intercomparison of models.

The following groups next contributed presentations of their current modeling efforts.

Principal Investigator: Jonathan Foley

Co-Investigators: Marcos Heil Costa, Jeff Cardille


This group is currently engaged in an effort to develop an integrated dynamical biosphere model -- the Integrated Biosphere Simulator (IBIS). The modeling approach reconciles the disparity of the existing land surface packages and terrestrial biosphere models by representing a more complete hierarchy of ecosystem phenomena, including: surface physics (energy, water, and momentum exchange within the soil-vegetation-atmosphere system), 2.canopy gas exchange (photosynthesis, respiration, and stomatal behavior), 3.plant phenology (seasonal cycles of leaf development, leaf senescence, and plant activity), 4.whole-plant physiology (allocation of carbon and nitrogen, plant growth, tissue turnover, and age-dependent changes), 5.vegetation dynamics (recruitment, competition for resources, mortality, disturbances, and gap formation) , 6.carbon and nitrogen cycling (flow of carbon and nitrogen between the atmosphere, vegetation, litter, and soils). Initial versions of IBIS have been used to investigate global patterns of water balance, carbon cycling, and vegetation cover, as well as the potential impact of increasing CO2 concentrations on the hydrology of the Amazon basin. The team is developing a more sophisticated version of IBIS for simulations of global biogeochemical processes, vegetation dynamics, and terrestrial hydrology as part of our funded NASA EOS Interdisciplinary Science Investigation. Working with other LBA-Ecology investigators, plans are to use the integrated terrestrial biosphere model to evaluate changes in land surface processes, ecosystem dynamics, and terrestrial carbon storage in response to land use activities in Amazonia.

The vegetation component of IBIS utilizes physical parameters to drive competition among plant functional types. Trees and upper canopy level species have an advantage in terms of light harvesting, while grasses and shrubs or lower level species have an advantage of advantage in terms of surface water harvesting. Validation efforts have already utilized data from flux towers, soils, vegetation, climate over a limited spatial extent and time. In addition to river discharge data for which Brazil, soil moisture networks in Russia and Illinois that are extensive in both space and time. IBIS validation focused lately on the Mississippi River basin with soil moisture estimates, USGS stream gauge data, river routing data calculated from 56 hydrograph stations, and flux data from 15-18 Ameriflux towers within the basin.

Another project goal is to generate the following synthesis of land cover/ and use for Amazonia (at a 5 minute by 5 minute spatial resolution, 1700-1992), by "fusing" several different sources of data, including satellite-based land cover classifications (from AVHRR and Landsat), various historical inventory data, and various in situ vegetation maps.

- natural (or "potential") vegetation (the land cover that existed before major human influences)

- croplands (fraction of gridcell over time)

- pastures (fraction of gridcell over time)

- secondary forest (fraction of gridcell over time)

- forest plantations (fraction of gridcell over time)

- other disturbed lands (fraction of gridcell over time)

Model developments for LBA include; 1) the addition of a land-use and age structure component, 2) use of a permanent cropland (excluding pasture) map at 10 km resolution, 3) linkage to hydrology incorporating runoff and precipitation, 4) modeling of fire frequency in different biomes vs. precipitation.

Model parameter refinements needed from LBA-Ecology include: canopy temperature profiles, soil heat conduction, organic soil layer(s), volumetric soil water content, vertical description of soil texture and rooting profile (for water uptake), potential river routing, topography of the study area including rivers, lakes and reservoirs.

Principal Investigator: Christopher Potter

Co-Investigators: Claudio Reis de Carvalho, Reynaldo Victoria, Joseph Coughlan, Susan Alexander, Steven Klooster, Vanessa Genovese, Matthew Bobo, Jennifer Dungan


The major objective of this research is to test a series of fundamental process-level hypotheses related to specific land use effects on ecosystem productivity, biogeochemistry and trace gas fluxes in the Amazon region. The main tool for these tests is the daily model version of CASA (Carnegie-Ames-Stanford Approach) developed at NASA Ames Research Center (Potter et al., 1998), specifically for Amazon land cover/use types. The modeling tests will be conducted in close collaboration with experimental field studies planned for the LBA intensive research sites. The ultimate research goal is to validate our model applications for different land uses at LBA sites and to scale-up regionally and dynamically the plant-soil biochemical, hydrologic, and production components of the daily NASA-CASA Amazon version, so as to more closely simulate and predict the interannual ecosystem observations from LBA. Research at NASA Ames has been supported previously to refine and extrapolate the monthly and daily moisture balance and regional plant production components of our CASA model version for Brazil, through a joint research grant with collaborators (D. Nepstad et al.) at Brazil's Instituto de Pesquisa Ambiental da Amazon (IPAM) to study moisture relations in forests of the eastern Amazon. Regional model drivers have been assembled and a one-year Amazon-basin version (8-km spatial resolution) of the NASA-CASA model has been completed.

Under LBA-Ecology, model drivers for interannual (1980-1995) CASA simulations are under development and include gridded precipitation, NDVI, surface irradiance, as well as soil and land cover/land use data. Deforestation effects in the Amazon by cutting and burning are programmed to alter in the model's flow equations and storage. Immediately following a deforestation event, land cover settings in the model are changed to represent properties of converted agricultural systems, either as shifting cultivation use, or directly to a pasture cover type, or a temporal sequence of the two. The project will further investigate the effects of changing land use on model predictions, through relatively high resolution (30-m Landsat TM classification) model applications that focus on ecosystem production, nutrient cycling, and biogenic trace gases (BTG) exchange along important eco-climatic transects and at intensive LBA study site locations. The team is developing a land use classification scheme for the Amazon basin using five spectral signatures from Landsat Pathfinder TM and a ‘Spatial Segmentation Application’ for smoothing. Land cover/use types include Primary forest, Secondary forest (young), Cerrrado, Pasture, Cultivation, Burned, Wetland, and Open water. We have completed classification and geo-rectification of seven Landsat TM scenes covering most of Rondonia. Similar land cover maps are being generated by our team for the Tapajos area of Para. Landsat TM land cover classifications will be used to improve CASA ecosystem model predictions of carbon, water, and trace gas fluxes for the Amazon region. High resolution (30-m) information on land cover class distributions will be used to refine spatial model predictions at LBA intensive study areas. Land cover distributions from TM images will be compared to 1-km AVHRR land cover maps for the purposes of validation and scaling analysis of spatially nested model predictions.

The parameters available for basin scale simulation with NASA-CASA are monthly NDVI, climate, land cover type from Stone (1994), and a soils map developed from the RADAM dataset (available for ftp at Soil C, N and pH at 8-km grid resolution are also available for ftp. Weather station data from a 10-year historical records (1980-1990) is being interpolated to generate gridded precipitation data sets. Early results indicate that it is difficult to reproduce high precipitation events, which may be important to nutrient cycling processes. Geo-statistical analyses are underway to investigate these patterns. Historical surface radiation data are obtained from the SeaWifs project at monthly 2.5 degree resolution.

Model parameter refinements needed from LBA-Ecology include: foliar N and lignin measurements, vertical root distribution, maximum rooting depth, land cover ground truth, daily surface radiation.


Principal Investigator: Jeffrey Richey

Co-Investigators: Reynaldo Victoria, Victoria Ballester, Niro Higuchi, John Hedges, Allan Devol, Miles Logsdon, Emilio Mayorga, Robin Weeks, Jim Tucker


The objective of this research group is to track the sequence of hydrological and linked biogeochemical processes that govern the movement of water, nutrients and organic matter from fixation in the system through to basin export by the river systems. This team developing a variable-scale drainage basin element model for biogeochemical processes which include water and carbon balances and river routing (in conjunction with hydrology work elsewhere in LBA). A goal of the research is to use these model inputs to accurately predict the timing between river height and precipitation and the movement of carbon (DOC, POC and black carbon) through soil column, riparian zone and into rivers. This requires development of multi-scale digital elevation models and spatial data layers (topography, soils, land use) sufficient to evaluate 4-km, 1-km, and ultimately 30-m resolution. The enhanced hydrological models will allow the team to implement a more detailed consideration of the mechanistic processes within a modeling environment ("HYDRO-CASA") for the evolution and partitioning of organic matter, nutrients, and trace gases. A key element is the use of internally consistent data sets for biophysical drivers and validation data sets describing the spatial and temporal variability of the land surface and hydrologic response. These are based on AVHRR time series data, using response filters to minimize fine scale variance from clouds and smoke in the estimation of remotely sensed parameters, including precipitation, temperature, radiation, NDVI and FPAR. For example, significant variations are seen in surface response (MFSV, FPAR, NDVI) in the year following El Nino, rather than during the year itself. Comparisons of model calculations and river gauge data for the upper Amazon suggest that the behavior of precipitation quantity and distributions are not being captured by the existing (very sparse) gauging network. In the future, regional extrapolation in the overall context of biogeochemical cycling will be based on time series multi-spectral image measurements along the LBA transects to validate and determine uncertainties in remote sensing inputs (e.g., FPAR, net solar radiation, temperature, non-photosynthetic vegetation) from image data.


Principal Investigator: Mathew Williams

Co-Investigators: Edward Rastetter, Yosio Shimabokuro.

The goal of this project is to test, develop and strengthen a set of hypotheses embedded in two process-based models, one related to canopy gas exchange and the other to ecosystem biogeochemistry, in order to improve our understanding of the dynamic nature of C storage and nutrient fluxes in Amazonia. These tasks require close co-operation with field teams. For example, in an earlier collaboration with A. Nobre (INPA) and J. Grace and Y. Malhi (Edinburgh) factors were considered that control daily and seasonal carbon (C) and LE fluxes in an Amazonian rain forest, by comparing a detailed model of the soil-plant-atmosphere continuum (SPA; Williams et al. 1996) against a long-term data set collected using eddy covariance at an undisturbed rain forest site north of Manaus, Brazil (ZF2 tower). Initial application of the SPA model was parameterized with simple measurements of canopy structure, and driven with local meteorological data. It made effective predictions of C and LE exchange during the wet season, but dry season predictions were overestimates in both cases (Williams et al. 1998). Sensitivity analyses indicated that the best explanation for this behavior was a seasonal change in below-ground hydraulic resistance. An optimization routine was then used to estimate the increase in below-ground hydraulic resistance during the dry season that would be required to explain the reduced dry season fluxes. The local soil, a clay latosol, is typical of much of Amazônia, having very low available water and low hydraulic conductivity. It appears that an increase in hydraulic resistance in the dry season introduces a significant seasonal cycle to carbon and water fluxes from this tropical forest. In LBA, this hypothesis will be tested further, working with groups to analyze eddy flux, sapflow, soil moisture, leaf gas exchange and root distribution data to further understanding of seasonality in ecosystem processes in both primary forest and disturbed or managed habitats. An aggregated version of the model (Williams et al. 1997) will be used within a GIS to investigate the regional implications of seasonality on the functioning of the Amazon system, and its sensitivity to global and land use change. A second modeling component will examine the coupling between nutrient dynamics and longer term C storage and exchange over decadal timescales, using an ecosystem biogeochemical model (MBL-GEM; Rastetter et al .1991). A primary goal will be to test whether one can predict the long-term responses of biomass and soil organic matter to land-use change, using chronosequence data for validation.

These models should prove effective tools for regional and temporal scaling of C and LE fluxes. The primary data requirements for SPA are regional and temporal information on meteorology (temperature, humidity, irradiance, precipitation), leaf area index (LAI), foliar N, critical leaf water potentials, and plant and soil hydraulic characteristics (soil moisture retention curves, rooting depths and distributions, soil water potentials). The primary requirements for GEM are data on C, N and P stocks in soils and biomass, and flux rates between the major pools. SPA will be used to provide a productivity component directly to GEM.

Model parameter refinements needed from LBA-Ecology include: seasonal changes in LAI, foliar N measurements, vertical root distribution and sap flow.



Principal Investigator: Berrien Moore III (represented by George Hurtt and Paul Moorcroft)

Co-Investigators: Carlos Nobre, Steve Pacala, Jerry Melillo, Charles Vorosmarty, Bruce Peterson, Changsheng Li, George Hurtt, Paul Moorcoft, Steve Frokling, Bobby Braswell, Ernest Linder, Xiangming Xiao


This research will produce be a set of coupled, hierarchically structured models accessible through a common model framework. This framework will provide the means for investigating the team’s principal research objectives. The LBA region will be considered within the context of two broad environmental conditions: 1) natural ecosystems where perturbations in biogeochemical states are driven primarily by natural variability of climate and fire, and 2) disturbance gradients that are induced by human land-use activities and/or human-induced climate change. Models of terrestrial biogeochemical cycles, vegetation dynamics, hydrology and land-use change will be evaluated for applicability to the LBA study region. The ecosystem and hydrology models will be driven by the physical climate; whereas, the land-use model will be driven by biophysical, ecological, and economic constraints. The linked models will be incorporated into a Geographic Information systems (GIS) context, accessing numerous data sets from LBA or data layers housed at the host institutions. Evaluation of model performance will be by comparison and field measurements from LBA as well from published data sources. Satellite remote sensing analysis will be used as a means to evaluate the spatial and temporal patterns of model performance at the regional scale. Finally, formal statistical methods will be applied to characterize model uncertainty, in addressing the question of human impacts on the Amazonian landscape.

A new model has been developed for predicting the structure and function of ecosystems that formally includes sub-grid scale heterogeneity caused by disturbances such as treefall gaps, blowdowns, fires, and land use. Issues for model testing include the potential use of IGBP fire data, IGBP land cover classification data, and the Worldwide Organic Soil Carbon and Nitrogen Database. A special emphasis will be placed on obtaining data on forest age structure, land use and land use history, and disturbance regimes, particularly around LBA tower sites.

Principal Investigator: Emilio Moran (represented by J. C. Randolph)

Co-Investigators: -- Flavio Ponzoni, Dalton Valeriano, J. C. Randolph, Eduardo Brondizio, Paul Mausel


This group has collected data that they hope to contribute to LBA modeling efforts. The study sites in Agarape-Acu (European) and Tome-Acu (Japanese) in the state of Para characterized and quantified the nature and rate of production in secondary growth by developing local carbon budgets. Methods used ranged from interviews for land-use history, nested plots for biomass determination, leaf-level studies, litterfall and decomposition data, vegetation chemistry and meteorological data at Tome-Acu. Site characteristics include crop type, temperature (24-34 degree range), rainy and dry season CO2 flux data from tree profiles. The vegetation data (includes basal area) is collected from 20 sites, grouping young and old plots. Aboveground live biomass shows differences in the two areas and management practices. There is also root data for total biomass calculation. Carbon stocks in large tree and small tree foliage, wood, roots, and soil were measured. Most of the carbon is found in the top 20 cm and 20-40 cm but there is also data to 60 cm. Litterfall was collected using trays and bags. Carbon concentration remains constant with age but increases with decreasing precipitation. Soil concentrations of phosphorus are an order of magnitude higher than litter P. Photosynthesis rates are higher in early succession species. Chronological data sets, with approximately 60 secondary forest plots, are well defined and can be used to evaluate land use questions including whether sites are becoming depleted by reuse and the cycling time of abandonment and reuse.

Subsequent workshop discussions were organized around several key questions for regional modeling.

  1. What processes are represented now in LBA-Ecology models of regional productivity and what important processes are missing?

As a summary statement, most of the models presented at this workshop have the common capacity to represent the processes of net carbon assimilation by plants at the tree or stand level. Soil water dynamics and uptake by vegetation are commonly simulated, along with latent heat (LE) fluxes from the plant canopy and changes in surface (ca. 0-2 m) soil moisture content. Several models can also represent decomposition of plant litter and root material in forest soils for estimates of soil carbon fluxes and net ecosystem exchange (NEE). Nevertheless, LBA models vary substantially in their respective levels of mechanistic control and aggregation of NEE processes, such as photosynthesis, respiration, and microbial activity (see Table 1).

It appears that the following parameters and processes are not well represented in LBA-Ecology models for ecosystem productivity. Hence, this list provides LBA field teams with a number of potential measurement targets that would be of value to ecosystem modelers.

Soil hydraulic properties - moisture, retention, conductivity

Soil heat conductance

Rooting depth and distribution

Microbial biomass dynamics

Soil respiration

Nutrient uptake processes - soil to vegetation (i.e., nitrogen and phosphorus)

Leaf phenology, longevity, and chemistry

Sap flow rates

Canopy profiles (e.g., temperature, conductance)

Growth and biomass allocation to plant tissues

Influence of flooding on productivity

Disturbance processes - rates, types, extent

Agricultural land use histories (i.e., pasture, crop)

Biodiversity and plant forms (e.g., secondary forest, vines)

A survey of model inputs and outputs was conducted to summarize regional data sets that are planned during the course of LBA-Ecology. Table 1 consolidates information presented by investigators at this workshop, and includes data provided previously by LBA-Ecology modeling teams to the LBA-Ecology project office.

  1. How do existing model predictions compare to existing observational data? How do model predictions compare with one another?

  2. Limited information now exists to answer this question. The participants therefore agreed that model comparisons of canopy fluxes (carbon, water, energy exchange) were both desirable and potentially achievable for presentation at the third LBA-Ecology Science Team meeting in April 1999. Primary objectives of this first LBA model comparison activity will be to better understand general approaches, different model algorithms, underlying hypotheses, data gaps, and the diversity of responses in predicting tower fluxes for an Amazon forest stand.

    As a first test of LBA-Ecology model comparison, Mathew Williams has agreement with his ABRACOS collaborators (e.g., Antonio Nobre, John Grace, and Yadvinder Malhi) at the Manaus ZF2 tower site on a plan for sharing among the workshop participants hourly meteorological driver data plus daily estimates of NEE and LE flux data. These data are now distributed via email. Several site characteristics for vegetation overstory and soils have been defined also for standard use by the LBA modeling groups. Other diagnostic model estimates may be generated and compared, such as GPP, and soil moisture and temperature profiles.

  3. What variables can we predict with high confidence? Low confidence? No confidence? And at what scales?

  4. The overall conclusion of the participants about confidence levels was that, owing to the scarcity of data both spatially and temporally, there is little confidence in most model variables, except those drivers like air surface temperature, which may show relatively low seasonal and diurnal variation over the region.

  5. What experimental or observational studies are needed to improve representation of existing/missing processes? What interactions with LBA-Hydromet teams should be planned?

Items identified through participant discussion included several remotely sensed and ground-based studies:

  • Light use efficiency of plants under cloudy conditions, in relation to direct and diffuse radiation

  • Surface radiation fluxes for the regional at minimum of 0.5 degree resolution

  • Atmospheric corrections for satellite vegetation index time series data

  • Biophysical parameters to drive models of canopy light use, including

  • * LAI

  • * specific leaf area

  • * foliar concentrations of nitrogen, chlorophyll, lignin, cellulose, and water

  • * leaf thickness

  • * leaf hemispherical reflectance/transmittance spectra for species of interest

  • * soil reflectance

  • Controls of tree phenology over several years of variable moisture and light

  • Soil respiration fluxes on a diurnal basis

  • Below ground production and root turnover rates

  • Plant nutrient uptake and soil availability for N and P

  • Soil P mineralization and occlusion rates

  • Asymbiotic and symbiotic N fixation

  • Wet and dry N deposition rates

  • Tower studies at a relatively aseasonal site in the western Amazon and at a transitionally dry site (LBA is in the initial stages of developing a such site in Mato Grosso)

  • Studies of disturbance processes, their frequency and their impacts

  • Studies of succession following disturbance events

Interaction with the newly formed LBA-Hydromet teams was discussed at length with Roni Avissar, who suggested that LBA-Ecology and LBA-Hydromet try to overlap for a day of joint meetings in April. Roni offered to coordinate the development of several regional meteorological data sets that would meet the needs of LBA-Ecology modelers. Much work remains however to define requirements and priorities.


  1. What data are needed for future model checking and validation? Where will these data come from?

Validation can be defined as the evaluation of the predictability of a model (together with the model parameter estimates) based on objective comparison with a data set not used for model building and estimation. A model may be valid for one purpose, or at one scale, and not valid at another. The objective of validation is to examine whether the model is an adequate description of the validation data set in terms of its behavior and of the application proposed. There are at least two types of validation: external and internal. External validation is the application of the developed model to a new data set (validation data set) from another independent study. Internal validation refers to the use of data-splitting and resampling techniques (cross-validation and bootstrapping) for evaluation purposes.

For LBA-Ecology simulations of regional carbon cycling and process controls, items identified through participant discussion included several potential validation data sets:

  • River discharge and routing at the basin scale

  • Multi-year tower flux LE and CO2, plus soil moisture profiles

  • Seasonal NDVI (LAI and fPAR) for phenology

  • Seasonal NDVI for fire monitoring

  • Above ground biomass and increments, from inventory methods and new remote sensing technologies such as SAR and Lidar

  • Soil C stocks and dynamics from field surveys






Susan Alexander

CSU Monterey Bay


Roni Avissar

Rutgers Univ


Matt Bobo



Jeff Cardille

U. Wisconsin


Joseph Coughlan



Jon Foley

U. Wisconsin


Vanessa Genovese



George Hurtt



Michael Keller

USDA/Forest Service


Steve Klooster

CSU Monterey Bay


Robin Martin



Emilio Mayorga

Univ. of Wash

Paul Moorcroft



Chris Potter



J.C. Randolph

Indiana University


Jeffrey Richey

Univ. of Wash


Alicia Torregrosa

CSU Monterey Bay


Mathew Williams