Performance of dry and wet spells combined with remote sensing indicators for crop yield prediction in Senegal

Abstract

Studying the relationship between potential high-impact precipitation and crop yields can help us understand the impact of the intensification of the hydrological cycle on agricultural production. The objective of this study is to analyse the contribution of intra seasonal rainfall indicators, namely dry and wet spells, for predicting millet yields at regional scale in Senegal using multiple linear regression. Using dry and wet spells with traditional indicators i.e. proxies of crop biomass and cumulated rainfall, hereafter called remote sensing indicators (NDVI, SPI3, WSI and RG), we analysed the ability of dry and wet spells alone or combined with these remote sensing indicators to provide intraseasonal forecasts covering the period 1991–2010. We analysed all 12 regions producing millet and found that results vary strongly between regions and also during the season, as a function of the dekad of prediction. At the spatial scale, the strongest performing combinations include the dry spell indicators DSC20 and DSxl in the peanut basin. While in the south of the country, the combination of wet period indicators WS1 or WSC5 with the RG is fairly reliable. Focussing on Thies, our best region in the groundnut basin, we showed that dry and wet spells indicators can explain up to 80% of yield variations, alone or in combination with remote sensing indicators. Regarding the timing of prediction, millet yield can be forecast as early as July with an accuracy of 40% of the mean yield but the best forecast is obtained in early September, at the peak of crop development (accuracy of 100 kg/ha i.e. 20% of the mean yield). Although, the estimated yields show biases over some years identified as extremely deficient or in oversupply in terms of agricultural yields.

Keywords

• Dry/Wet spells
• Remote sensing
• Crop yields
• Multiple linear regression model