Standardized Precipitation Index (SPI) is one of the most used drought monitoring tools. It is easy to compute as it only needs cumulative precipitation amount for an input. However, it is more of a meteorological drought index rather than an agricultural one. In this study, we reconstruct SPI for monitoring a specific crop by calibrating the cut-off values that separates drought classes. For this purpose, two objectives to optimize are obtained: area under receiver operating characteristics (ROC) curve and misclassification rate of a multivariate decision model. By maximizing the area under ROC curve, we are able to calibrate thresholds for the realized states of the drought. By multivariate decision problems, crop and location specific information is used to regulate the size of the classes so that they can reveal agricultural wise meaningful information. Rain-fed wheat monitoring at Polatli station of Turkey is studied for an implementation.