HR: 0800h
AN: B51C-0233    [Abstracts]
TI: Adaptive Rule-Based Piece-Wise Regression Models for Estimating Regional Net Ecosystem Exchange in Grassland and Shrubland Ecoregions Using Regional and Flux Tower Data
AU: * Fosnight, E A
EM: fosnight@usgs.gov
AF: USGS/EROS/SAIC, USGS EROS, Sioux Falls, SD 57042 United States
AU: Wylie, B K
EM: wyie@usgs.gov
AF: USGS/EROS/SAIC, USGS EROS, Sioux Falls, SD 57042 United States
AU: Zhang, L
EM: lizhang@usgs.gov
AF: USGS/EROS/SAIC, USGS EROS, Sioux Falls, SD 57042 United States
AB: The scientific understanding of the global carbon cycle requires quantitative documentation, monitoring, and projection of carbon stocks and fluxes at various scales across the landscape. The challenge is to develop predictive models using carbon flux towers at site-specific locations, and to extrapolate these models to landscapes and regions. We use remote sensing and national climate and soil databases within data-driven models to estimate carbon fluxes. To accommodate the study of coupled human-environmental relationships and their influences on carbon dynamics, a coherent suite of models is being developed for agricultural, wooded and wetland ecosystems within predominantly grassland and shrubland ecoregions. In previous work, we have mapped carbon fluxes in terms of Net Ecosystem Exchange (NEE), Gross Primary Production (GPP), and Respiration (Re) in the Northern Great Plains, the Sagebrush Steppes and the Kazakh Steppes at 1-km resolution and 10-day time steps. We now extend this work beyond fairly uniform ecological conditions to accommodate more complex spatial mixtures of ecological types within ecoregions. The models need to adapt to both the complexity of the environmental variables and the land cover patterns. Our rule-based models adapt to local climatic, soil and phenology through the definition of piece-wise regression models. A suite of such models is needed to capture the phenologic and climatic variability across the wide range of shrubland and grassland ecoregions that exist. The result is a multi-year time series of 1-km maps of carbon flux that are suitable for trend and anomaly analysis. We seek sensitive models that permit the effective study of localized carbon dynamics while avoiding over-fitting the available carbon flux tower measurement data. Two critical components of the project are (1) sensitivity and cross-validation studies to evaluate the internal consistencies of the models and (2) intercomparison studies to help isolate methodological artifacts from variability resulting from climate and management of the land.
DE: 0414 Biogeochemical cycles, processes, and modeling (0412, 0793, 1615, 4805, 4912)
DE: 0429 Climate dynamics (1620)
DE: 0439 Ecosystems, structure and dynamics (4815)
DE: 0480 Remote sensing
SC: Biogeosciences [B]
MN: Fall Meeting 2005