The Potential of Sentinel-2 for Crop Production Estimation in a Smallholder Agroforestry Landscape, Burkina Faso

Journal article, 2020

Crop production statistics at the field scale are scarce in African countries, limiting potential research on yield gaps as well as monitoring related to food security. This paper examines the potential of using Sentinel-2 time series data to derive spatially explicit estimates of crop production in an agroforestry parkland in central Burkina Faso. This type of landscape is characterized by agricultural fields where cereals (millet and sorghum) and legumes (cowpea) are intercropped under a relatively dense tree canopy. We measured total above ground biomass (AGB) and grain yield in 22 field plots at the end of two growing seasons (2017 and 2018) that differed in rainfall timing and amount. Linear regression models were developed using the in situ crop production estimates and temporal metrics derived from Sentinel-2 time series. We studied several important aspects of satellite-based crop production estimation, including (i) choice of vegetation indices, (ii) effectiveness of different time periods for image acquisition and temporal metrics, (iii) consistency of the method between years, and (iv) influence of intercropping and trees on accuracy of the estimates. Our results show that Sentinel-2 data were able to explain between 41 and 80% of the variation in the in situ crop production measurements, with relative root mean square error for AGB estimates ranging between 31 and 63% in 2017 and 2018, respectively, depending on temporal metric used as estimator. Neither intercropping of cereals and legumes nor tree canopy cover appeared to influence the relationship between the satellite-derived estimators and crop production. However, inter-annual rainfall variations in 2017 and 2018 resulted in different ratios of AGB to grain yield, and additionally, the most effective temporal metric for estimating crop production differed between years. Overall, this study demonstrates that Sentinel-2 data can be an important resource for upscaling field measurements of crop production in this agroforestry system in Burkina Faso. The results may be applicable in other areas with similar agricultural systems and increase the availability of crop production statistics.