Remote estimation of GPP in temperate grassland: implications of the uncertainty in GPP estimation in semi-arid ecosystems using MODIS data

Abstract

This study analyzed grassland gross primary production (GPP) estimated by the Temperature and Greenness (TG) model and the Moderate Resolution Imaging Spectroradiometer (MODIS) algorithm along the mean precipitation gradient and as a function of interannual variability in site-level precipitation. The calibrated TG model and MODIS algorithm appeared to provide accurate GPP estimations at three study sites with varying precipitation. However, the evaluation for each site/year demonstrated the variations of the accuracy of GPP estimates among different sites and years. GPP were overestimated at the driest site among three study sites, and during the dry years of the semiarid site. Both models provided more accurate GPP estimates for the wet site and during the wet and normal years of the semiarid sites. Calibrating both models for each site/year showed that the parameters of both models varied among sites and years, especially for the TG model. The relationship between flux-tower GPP observations and (scaled EVI *scaled LST) for the TG model and the relationship between GPP observations and (fPAR*PAR*Tmin scalar*VPD scalar) for the MODIS algorithm were different during green-up and dry-down period of grassland during the dry years at semiarid sites. This result implied that different relationships at different growing stages might be one of the major reasons for the overestimation of GPP by the TG model and the MODIS algorithm for semiarid grassland where water is a limiting resource. Thus, both TG model and MODIS algorithm should be used with caution in the arid and semiarid grassland regions