On the Bayesian network based data mining framework for the choice of appropriate time scale for regional analysis of drought Hazard

Abstract

Data mining has a significant role in hyrdrologic research. Among several methods of data mining, Bayesian network theory has great importance and wide applications as well. The drought indices are very useful tools for drought monitoring and forecasting. However, the multi-scaling nature of standardized type drought indices creates several problems in data analysis and reanalysis at regional level. This paper presents a novel framework of data mining for hydrological research—the Bayesian Integrated Regional Drought Time Scale (BIRDts). The mechanism of BIRDts gives effective and sufficient time scales by considering dependency/interdependency probabilities from Bayesian network algorithm. The resultant time scales are proposed for further investigation and research related to the hydrological process. Application of the proposed method consists of 46 meteorological stations of Pakistan. In this research, we have employed Standardized Precipitation Temperature Index (SPTI) drought index for 1-, 3-, 6-, 9-, 12-, 24-, and ()month time scales. Outcomes associated with this research show that the proposed method has rationale to aggregate time scales at regional level by configuring marginal posterior probability as weights in the selection process of effective drought time scales.