An early warning system for agricultural drought in an arid region using limited data

In this paper, an early warning system for agricultural drought in an arid region (Jodhpur district, Rajasthan, India) is developed by estimating grain yield of pearl millet, the major food crop in the region. Two multiple linear regression models, namely the intermediate warning (IW) model that estimates the yield about a month before the crop is harvested, and the final warning (FW) model that estimates the yield at about harvest time, are developed for the purpose. Explanatory variables used in the models are delay in crop sowing, monthly rainfall, and number of rainy days in a month. These variables are derived from the daily rainfall data which are easily available in arid regions. Using the 1963–1987 data for model development, up to 74% and 81% yield variations could be explained by the IW and FW models, respectively. When these models were tested on the 1988–1991 data, mean absolute per cent error in the estimated yields was found to be 18.055% and 11.052% for the IW and the FW model, respectively. Models are evaluated and some measures for their improvement discussed.