Abstract

Dekads in the prediction of yields under associated millet and cowpea crops were studied in three localities located in the Sahelian zone. Multiple linear regressions was applied to create predictive models based on variables extracted from the decadal Normalized Difference Vegetation Index (NDVI) image series from the SPOT VEGETATION satellite, the GeoWRSI tool, and the recorded ground rainfall. The earliest forecast of millet occurs as early as the third dekad of July with an estimation error of 53 Kg ha-1. Combined variables derived from NDVI, Water Requirements Satisfaction Index (WRSI) and rainfall explain better the variation of millet yield in the first and second dekads of September with estimation errors ranging from 50 to 75 Kg ha-1 according to the localities. The earliest forecast of cowpea can occur in the third dekad of August with an estimation error of 68 Kg ha-1. Cowpea forecasting also occurs later in the season in the second dekad of September by combining NDVI with rainfall or WRSI with rainfall with estimation errors ranging from 52 to 70 Kg ha-1. However, when sorghum is close to the associated millet and cowpea crops in the cultivated area and when the soil moisture are satisfactory for crops, NDVI and WRSI have little or no explanation for variation in millet and cowpea yields. The forecast under associated cereal and legume crops encountered in Sahelian production systems can be envisaged based on the environment topography and on high spatial resolution NDVI images.

Key words: Cowpea, millet, normalized difference vegetation index (NDVI), prediction, water requirements satisfaction index (WRSI), yield.