A comparison of remote sensed semi-arid grassland vegetation anomalies detected using MODIS and Sentinel-3, with anomalies in ground-based eddy covariance flux measurements.

Abstract: Remote sensed vegetation biophysical indicators are derived from various sensors and methods and contribute to the quantification of vegetation growth. They are used as inputs to early warning systems for crop yield estimation, thus improving crop management and food security. However, the performance of any such system is affected by the performance of the chosen remote sensed biophysical indicator. Performance based selection, requires evaluation of the performance of various remote sensed indicators, in how well each indicator represents actual vegetation growth. Scientific Problem Such performance evaluation has previously been done in northern-latitude focused studies, with little or none having been conducted in southern hemisphere semi-arid grasslands specifically, such as those used for sheep grazing in the Nama-Karoo biome in South Africa. Study Aim The aim of this study is to test the strength of the associations between remote sensed indicators of fAPAR, GDMP and MODIS GPP, with ground-based eddy covariance-derived GPP, at a semi-arid Nama-Karoo grassland site in South Africa. Methods Correlation between standard scores (z-scores) of time-series biophysical indicator data was conducted, to test the relative strengths of each indicator and eddy covariance reference data. Additionally, linear regression yielded models (R2 = 0.748, 0.538 and 0.129, RMSE = 0.599, 1.34 and 1.88 gC/m2/day) that could be used as a fast and simple predictors of Nama-Karoo type semi-arid grassland biomass availability for grazing management, in similar ecozone’s such as parts of mid-latitude Australia, South America and Mexico. Study Findings The findings of the study indicate the strongest correlation exists between standard scores within the MODIS GPP data (r = 0.849, p < 2.2e-16), followed by Sentinel-3 GDMP (r = 0.667, p = 4.3e-12) and lastly Sentinel-3 fAPAR (r = 0.239, p = 0.045). These performance differences are likely due to temporal response differences relating to changes in temperature, vapour pressure deficit, soil moisture, light use efficiency and other variables. The importance of this is that standard scores calculated using Sentinel-3 GDMP observational data may prove more useful, for semi-arid grassland vegetation anomaly early warning systems, than fAPAR, and only slightly less well than MODIS GPP.