Performance of four state-of-the-art GPP products (VPM, MOD17, BESS and PML) for grasslands in drought years

Abstract Accurate estimation of gross primary production (GPP) is of significance for understanding the changes of carbon uptake and its responses to extreme climate events like droughts. Emerging new GPP products with higher spatial and temporal resolutions (500-1000 m, 8-day) from the Moderate Resolution Imaging Spectroradiometer (MODIS) Photosynthesis (MOD17), the Vegetation Photosynthesis Model (VPM), the Breathing Earth System Simulator (BESS), and the Penman-Monteith-Leuning (PML) models, provided unprecedented opportunities to understand the spatial and temporal variations of GPP. However, their performances under drought conditions remain obscure. Here we evaluated the performance of these four state-of-the-art GPP products in grasslands, using FLUXNET data as reference. The results showed that all the four models have reasonable accuracies under non-drought years. In drought years, the VPM performed best, followed by the MOD17, PML and BESS, with the RMSEs of 1.67, 1.69, 1.72 and 1.77 gC m−2 day−1, respectively. The VPM, BESS and PML overestimated annual GPP by 2%, 13% and 21%, respectively, while MOD17 underestimated annual GPP by 10% in drought years. This varied model performances under drought years could be partially attributed to the differences in quantifying the water stress effects. The water constraint factor in the VPM, which is derived from the Land Surface Water Index (LSWI) and directly indicates the overall water content of leaf, plant stand and soil background, could better capture the vegetation response to water content variation than that in MOD17, PML and BESS, all of which used an atmospheric moisture related indicator (the Vapor Pressure Deficit for MOD17 and PML, and the relative humidity for BESS). This study suggests that water stress factors, which reflect the physiological and ecological characteristics of vegetation itself (e.g., LSWI) rather than atmospheric moisture (e.g., VPD) or other meteorological surrogates, should be further considered in GPP models when applied in drought conditions