Drought is a natural disaster with creeping cumulative effect. To reduce drought-related losses, water resources management strategies in arid and semi-arid regions are of great importance. The objective of this study was to predict agricultural water shortage based on development of a Drought Early Warning System (DEWS) consisting of hydrological indicators for monitoring, forecasting, and determining the level of water shortage for allocated agricultural water. More specifically, the DEWS involves forecasting of drought, water shortage, and drought alert index. The results in the Karkheh river basin in southwest Iran showed that the 3-month standardized hydrological drought index (SHDI-3) can adequately characterize drought severity in the study region. SHDI-3 was then divided into five classes so that the next month drought class was predicted based on the SHDI-3 of preceding and current months using support vector machine. Using K-medoids algorithm and the correlation between consequent months, agricultural water shortages were predicted. The overall accuracy, underestimation error, overestimated error, and the risk index were 0.81, 0.08, 0.11, and 0.7%, respectively. As a result, the monthly flow and demand variables were good predictors for prediction of upcoming water shortage. The findings of this study improve understanding of drought impacts on meeting agricultural water demands and to mitigate the present and upcoming water shortage consequences.