Millet yield can consistently be assessed using AVHRR NDVI data. Two years of data showed that one linear regression line between grain yield and integrated NDVI could be statistically justified. Adding environmental information and using multiple regression techniques improved the yield model further. The agricultural domain was stratified into intensive and extensive cultivated land. By adding the environmental variable of livestock density to the millet yield integrated NDVI model, the level of explained yield variance was improved to 88 per cent for the intensively cultivated area. For the rest of the agricultural domain, the variable percentage cultivated land was included to the yield integrated NDVI model explaining altogether 76 per cent of the yield variance. GIS spatial interpolation tools were used to generated surfaces of per cent cultivated land and livestock density from point observations. The use of Photosynthetically Active Radiation (PAR) data along with NDVI to assess millet grain yield or total crop biomass, was found to be of limited use since no single regression line valid for both years could be established and the level of explained variance was reduced compared with using the NDVI alone.