National Aeronautics and Space Administration

Gross primary productivity (GPP) quantifies the photosynthetic uptake of carbon by ecosystems and is an important component of the terrestrial carbon cycle. Empirical light use efficiency (LUE) models and process-based Farquhar, von Caemmerer, and Berry (FvCB) photosynthetic models are widely used for GPP estimation. In this paper, the MODIS GPP algorithm using the LUE approach and the Boreal Ecosystem Productivity Simulator (BEPS) based on the FvCB model in which a sunlit and shaded leaf separation scheme is evaluated against GPP values derived from eddy-covariance (EC) measurements in a variety of ecosystems. Although the total GPP values simulated using these two models agree within 89% when they are averaged for the conterminous U.S., there are systematic differences between them in terms of their spatial and temporal distribution patterns. The spatial distribution of MODIS GPP therefore differs substantially from that produced by BEPS. These differences may be due to an inherent problem of the LUE modeling approach. When a constant maximum LUE value is used for a biome type, this simplification cannot properly handle the contribution of shaded leaves to the total canopy-level GPP. When GPP is modeled by BEPS as the sum of sunlit and shaded leaf GPP, the problem is minimized, i.e., at the low end, the relative contribution of shaded leaves to GPP is small and at the high end, the relative contribution of shaded leaves is large. Compared with monthly and annual GPP derived from eddy covariance data at 40 tower sites in North America, BEPS performed better than the MODIS GPP algorithm. The difference between MODIS and BEPS GPP widens as with the fraction of shaded leaves increases. The simpler LUE modeling approach should therefore be further improved to reduce this bias issue for effective estimation of regional and temporal GPP distributions.