Abstract
Spatial distribution of drought plays key role specifically in hydrological research. Drought is a complex stochastic natural hazard caused by prolonging shortage of rainfall and available water resources. The multi-scalar drought indices (based on probability distribution) are commonly used for making effective drought mitigation policies. In addition, the multi-scalar drought indices have great extent of the inaccurate determination of drought classes due to its probabilistic nature. However, the interpretation and applicability of various time scales are cumbersome for multi-scalar drought in various meteorological stations at a particular region. In this regards, accurate estimation and continuous monitoring of future drought at regional level requires a more appropriate and important time scale with respect to regions under study. In this study, we aimed to investigate the appropriate time scale for multi-scalar drought indices by using geo-reference points of meteorological stations. We used Boruta algorithm with two random forest adapters of machine learning algorithms for regionalized optimization of drought monitoring time scale. We explored the appropriate time scale for the Standardized Precipitation Temperature Index (SPTI) of 52 meteorological stations that are located across Pakistan. Results show that the significant importance of SPTI-1 (1-month time scale) for further spatial and temporal studies. That is, being high ranked and prominence, SPTI-1 has the ability to capture the characteristics of all other time scales that are in some circumstances applicable for drought characterization and classification.