Ecosystem Model-Data Intercomparison
About EMDI

The Ecosystem Model-Data Intercomparison (EMDI) activity will provide, for the first time ever, a formal opportunity for a wide range of global carbon cycle models to be compared with measured net primary production (NPP). The goals of the EMDI are to compare model estimates of terrestrial carbon fluxes (NPP and net ecosystem production (NEP), where available) to estimates from ground-based measurements, and improve our understanding of environmental controls of carbon allocation. The primary questions to be addressed by this activity are to test simulated controls and model formulation on the water, carbon, and nutrient budgets with the observed NPP data providing the constraint for autotrophic fluxes and the integrity of scaled biophysical driving variables.

Background — Progress in modeling the global carbon cycle has previously been inhibited by the lack of adequate observational data for model parameterization and validation, such as NPP from field measurements (Scurlock et al. 1998). Total NPP and its individual components are key ecosystem variables. The International Geosphere-Biosphere Programme (IGBP) project on Global Analysis, Interpretation and Modeling (GAIM) held global NPP model intercomparison meetings in 1994 and 1995 at the Potsdam Institute for Climate Impact Research (PIK), Germany (Hibbard and Sahagian, 1998). Here it was found that further progress was inhibited by the lack of appropriate field observations of NPP. The Global Primary Production Data Initiative (GPPDI) was set up as a Focus 1 activity of the IGBP Data and Information System to address this problem (Prince et al. 1995). More recently, the Vegetation/Ecosystem Modeling and Analysis Project (VEMAP) (Schimel et al., 1997) cited similar difficulties in comparing model predictions with currently available NPP estimates and an analysis of terrestrial carbon sinks cited the need for additional data to refine model constraints (Fan et al. 1998).

GPPDI — Under the auspices of GPPDI, the Worldwide NPP Working Groups 1, 2 and 3 (WG1-WG3) were held at the National Center for Ecological Analysis and Synthesis (NCEAS) between December 1997 and October 1998, to take extensive but incomplete datasets and make them usable for analyses and models by estimating total NPP in a consistent manner, for points and for large grid cells (Prince et al. 1997 proposal to NCEAS). In WG1 three biomes were considered (tropical dry and wet forest, boreal forest and grasslands). For each biome, the principal components of NPP and available field observations were identified. In all biomes, most existing field data measure only partial components of the total NPP, and so techniques were developed to estimate the missing components. WG2 addressed the issue of converting field measurements of NPP, estimated using the techniques elaborated in WG1, into estimates of NPP for large grid cells, up to 0.5° x 0.5°. Forest, crop and range inventory data were studied and biophysical data sets were considered. WG3 incorporated further expertise on forest inventory and below-ground data and produced drafts of a set of 12 open-literature papers associated with WG1-WG3. GPPDI resulting in compiling NPP data for over 130 intensive sites, for 2000-2500 extensive sites, and 2000-3000 cells (Table 1). This work was supported by IGBP-DIS, NASA Terrestrial Ecosystem Program, and the NSF NCEAS.

Planning — In October 1998, a GAIM-organized planning meeting was held to discuss opportunities to dovetail NPP data sets from the GPPDI with a global model/data intercomparison. The overall tentative schedule is given in Table 2. It was proposed that the model-data comparison would need two major working groups to facilitate the comparisons. The first would be a data-intensive workshop where primary (NPP) and ancillary (climate, soils, etc.) data sets would be compiled, input and output formats agreed upon, data formatted, and data distributed to participating modeling groups. NCEAS will host this workshop, tentatively scheduled for April 28- May 1, 1999. The second workshop would entail visualization and analyses of model results. Following the 1994 and 1995 model workshops, PIK has offered to host this workshop in August or September 1999. Many of the 17 models that were included in the 1995 PIK workshop are candidates for the EMDI exercise.

Experimental Design - The protocol for model-data intercomparison will be finalized at the NCEAS Working Group in April. The design will need to be transparent yet fair, with no modeling group having the opportunity to "fine-tune" their model to the data in advance. The proposed experimental design will consists of a multi-tiered approach to make maximum use of the available NPP and NEE measurements. The NPP data sets emerging from the GPPDI (see Table 1) are derived from both point (e.g. m2) and spatially explicit (e.g. county, polygon) sampling designs, thus enabling a valid comparison between point and area-based models and data. Analyses and visualizations would be carried out within each tier to investigate the model controls on NPP and their underlying formulations.

1. Site model-data comparisons - The first tier of the comparison would be to compare the globally distributed intensive point data (see Table 1) to models (e.g. biogeochemical) that predict carbon fluxes at a single point, or 1-D models. The driver fields for the points would be acquired and distributed to modeling groups so that the global models (i.e., not fine-tuned for each point) could be run for the selected points. It was proposed to do this comparison in two-phases: first to run the models 100% blind, then, as the models submit their preliminary results, to release 1/2 of the NPP data so groups can identify model strengths and weaknesses and potentially modify their models prior to the workshop and final comparison with all NPP values.

2. Grid-cell model-data comparisons - The second tier of the comparison would be to compare the globally distributed areal data (see Table 1) to models. Areal NPP data are derived from county and polygonal data for boreal and temperate forests, grasslands, and crops. Approximately ten global sites were identified as candidates for spatial modeling comparison, including Canadian boreal forest, St. Petersburg, Russian boreal forest, northeastern U.S. temperate forests, Superior National Forest, China, mid-western U.S. crops, H.J. Andrews temperate forests, the Great Plains grasslands, Queensland grasslands, and temperate forests in Wisconsin. Each areal data set will be disaggregated into component vegetation-types and their spatial extent provided. The inclusion of spatially-extent data will require models to simulate carbon fluxes and make valid comparisons with the measured NPP data. The driver fields for the areal sites would be acquired and distributed to modeling groups.

3. Global model-data comparisons - The third level of comparison would take advantage of the large collections of extensive NPP estimates (approximately 2500-3000 sites) to provide a broader comparison between patterns of measured NPP and model outputs. The extensive collections represent a mix of years, methods, and available documentation - often not known or readily available. However, the carbon allocation rules developed by the NCEAS Working Groups can be applied to these data to provide reasonably consistent set of NPP estimates when only partial information is available. A NASA-funded TECO project at ORNL (Techniques for validation of regional and global scale models against field data" PI: R. J. Olson) is developing robust statistical tests to compare models and data at different scales using relationships with environmental variables rather than geographic pairings. If these statistical approaches are accepted by the EMDI steering committee, then the extensive NPP data can be compared to global model outputs.

4. FLUX Data — An important source of ecosystem production measurements (with monthly time resolution) to include in model-data intercomparison is being compiled by the community of scientists associated with the global network of CO2 flux towers. The flux community is currently developing robust and consistent methods to aggregate hourly observations to monthly and annual sums of NEE. However, it is realized that most models estimate NPP while the flux towers measure NEP. It is anticipated that 10-12 AmeriFlux sites and 10-15 EUROFLUX sites will be able to provide monthly, multiple-year estimates of net ecosystem exchange (NEE) to include in the EMDI exercise. Those biogeochemical models that maintain a complete carbon budget within their models would be asked to run their models for flux tower sites and compare their monthly NEP estimates to the monthly flux tower estimates.

Anticipated Results — A number of publications in major international journals are expected to result from the overall EMDI process. The publications will be immediately relevant to preparing the IGBP statement on C cycle in late 1999 and to the IPCC process. The NPP and associated data will constitute the state-of-the-art in validating and developing global carbon cycle models with field measurements. The data assembled for EMDI are also likely to have other ecological applications (biodiversity studies, nutrient cycling, etc.), as well as being likely to improve the calibration of remote sensing (e.g. the EOS Validation NPP product). Beneficiaries will include global change modelers, ecologists, and remote sensing scientists in the United States and worldwide.