Comparison of MODIS-Algorithms for Estimating Gross Primary Production from Satellite Data in semi-arid Africa

Abstract: The climatic patterns of the world are changing and with them the spatial distribution of global terrestrial carbon; the food and fiber of the world and in itself an important factor in the changing climate. Knowledge of how the terrestrial carbon stock is changing, its distribution and quantity, is important in understanding how the patterns of the world are changing and large scale models using remotely sensed data have emerged for this purpose. This study compares four vegetation related MODIS (Moderate Resolution Imaging Spectroradiometer) products, derived from MODIS satellite data using algorithms which calculates the two vegetation indices, Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI), and the two biophysical factors, Leaf Area Index (LAI) and absorbed Fraction of Photosynthetically Active Radiation (FPAR). The comparison is in their ability to estimate intra-annual variations of Gross Primary Production (GPP); this is done using the time-series data of quality screened eddy covariance (EC) Flux Tower stations from the Carbo Africa network as truth data. The results show a modest agreement between the different vegetation metrics and EC Flux Tower derived GPP, with an overall average coefficient of determination (R2) of 0.63 for LAI, R2 of 0.51 for NDVI, R2 of 0.52 FPAR and a R2 of 0.49 for EVI, using all stations and years of data. When each station received the same weight, i.e. using the correlation of all observation for each station and then calculating the average, the overall correlation improved, still showing LAI as the best predictor of Flux Tower GPP with a R2 of 0.62, but with an improved EVI with a R2 of 0.61, while NDVI and FPAR had an R2 of 0.57 and 0.59 respectively. This result and the observed large variation in between stations, e.g. NDVI between an R2 of 0.62 and 0.83 for the station Demokeya compared to an R2 of 0.32 and 0.49 of NDVI for the station Tchizalamou, may indicate a site specific proficiency of the vegetation metrics. When the observations within the growing period were tested separately a strong decrease in correlation was observed, with an average R2 between 0.41 – 0.56 for all station and years and an average R2 between 0.36 – 0.45 for all sites using all observations for each station regardless of year, lending strength to the assumption that the non-vegetation period observations affect the correlation greatly. The study concludes that up scaling of an intra-annual standardized major axis regression model based solely on the relationship between any of these metrics and Flux Tower estimated GPP is inadvisable due to the modest overall intra-annual agreement between the metrics and GPP. It is also concluded that since the vegetation metrics display site specific proficiency, models of GPP would benefit from site specific ancillary data that describes vegetation-limiting factors, e.g. water availability.