Estimating Per-Pixel GPP of the Contiguous USA Directly from MODIS EVI Data

We estimated gross primary production (GPP) of the contiguous USA using enhanced vegetation index (EVI) data from NASA's moderate resolution imaging spectroradiometer (MODIS). Based on recently published values of correlation coefficients between EVI and GPP of North American vegetations, we derived GPP maps of the contiguous USA for 2001-2004, which included one La Nina year and three moderately El Nino years. The product was a truly per-pixel GPP estimate (named E-GPP), in contrast to the pseudo-continuous MOD17, the standard MODIS GPP product. We compared E-GPP with fine-scale experimental GPP data and MOD17 estimates from three Bigfoot experimental sites, and also with MOD17 estimates from the whole contiguous USA for the above-mentioned four years. For each of the '7 by 7' km Bigfoot experimental sites, E-GPP was able to track the primary production activity during the green-up period while MOD17 failed to do so. The E-GPP estimates during peak production season were similar to those from Bigfoot and MOD17 for most vegetation types except for the deciduous types, where it was lower. Annual E-GPP of the Bigfoot sites compared well with Bigfoot experimental GPP (r = 0.71) and MOD17 (r = 0.78). But for the contiguous USA for 2001-2004, annual E-GPP showed disagreement with MOD17 in both magnitude and seasonal trends for deciduous forests and grass lands. In this study we explored the reasons for this mismatch between E-GPP and MOD17 and also analyzed the uncertainties in E-GPP across multiple spatial scales. Our results show that the E-GPP, based on a simple regression model, can work as a robust alternative to MOD17 for large-area annual GPP estimation. The relative advantages of E-GPP are that it is truly per-pixel, solely dependent on remotely sensed data that is routinely available from NASA, easy to compute and has the potential of being used as an operational product.