Evaluation of Selected Sentinel‐2 Remotely Sensed Vegetation Indices and MODIS GPP in Representing Productivity in Semi‐Arid South African Ecosystems

Abstract

The ability to validate satellite observations with ground‐based data sets is vital for the spatiotemporal assessment of productivity trends in semi‐arid ecosystems. Modeling ecosystem scale parameters such as gross primary production (GPP) with the combination of satellite and ground‐based data however requires a comprehensive understanding of the associated drivers of how the carbon balance of these ecosystems is impacted under climate change. We used GPP estimates from the partitioning of net ecosystem measurements (net ecosystem exchange) from three Eddy Covariance (EC) flux tower sites and applied linear regressions to evaluate the ability of Sentinel‐2 vegetation indices (VIs) retrieved from Google Earth Engine to estimate GPP in semi‐arid ecosystems. The Sentinel‐2 normalized difference vegetation index (NDVI), enhanced vegetation index (EVI) and the land surface water index (LSWI) were each assessed separately, and also in combination with selected meteorological variables (incoming radiation, soil water content, air temperature, vapor pressure deficit) using a bi‐directional stepwise linear regression to test whether this can improve GPP estimates. The performance of the MOD17AH2 8‐day GPP was also tested across the sites. NDVI, EVI and LSWI were able to track the phase and amplitude patterns of EC estimated gross primary production (GPPEC) across all sites, albeit with phase delays observed especially at the Benfontein Savanna site (Ben_Sav). In all cases, the VI estimates improved with the addition of meteorological variables except for LSWI at Middleburg Karoo (Mid_Kar). The least improvement in R2 was observed in all EVI‐based estimates—indicating the suitability of EVI as a single VI to estimate GPP. Our results suggest that while productivity assessments using a single VI may be more favorable, the inclusion of meteorological variables can be applied to improve single VIs estimates to accurately detect and characterize changes in GPP. In addition, we found that standard MODIS products better represent the phase than amplitude of productivity in semi‐arid ecosystems, explaining between 68% and 83% of GPP variability.