Enhancing Drought Risk Assessment Through Improved Statistical Pooling Techniques

Malaysia is known for its abundant amount of rainfall that might leading to significant challenges. However, the occurrence of droughts periods also introduces the same problem that give a major effect on many economic sectors. Drought frequently occurs in sequence when extended periods of low flow is interrupted by a small excess period, result in multiple number of small drought events rather treated as a single drought events. Thus, the statistical properties of drought characteristics show inconsistency in characterizing droughts, making the extreme value modelling unreliable with underestimation of drought characteristics due to multiple small drought events. This study addresses the limitation of traditional drought indices, which often fail to accurately capture the drought duration and intensity. Understanding on the behaviour of streamflow and its impact will improve a strategic planning and risk preparedness in dealing with drought. The detailed evaluation of the threshold-level method combined with pooling procedures, inter-event time (IT) method and moving average (MA) are used to remove the dependency between drought events. Both pooling methods are used to identify and characterize drought events efficiently. Hence, sensitivity analysis is utilized to determine an optimal parameter of IT and MA, critical duration \(\:{t}_{c}\)and averaging interval \(\:n\), respectively in order to pooled drought events. It if found that \(\:{t}_{c}=5\) and \(\:n=10\) are the optimal pooling parameters. This study also analyzed the suitability for a two series, annual maximum series (AMS) and partial duration series (PDS) to be applied for modelling extreme drought events. The three parameters of generalized extreme value (GEV) and generalized pareto distributions (GPD) are used to model the series. Four parameter estimations, maximum likelihood estimation (MLE), generalized maximum likelihood estimation (GMLE), L-moments (LMOM) and Bayes are compared to find a robust estimation. The L-moment ratio diagram and the goodness-of-fit (GOF) between empirical data and theoretical distributions are evaluated for each station. It is found that most of the PDS are well fitted by the GPD models, rather than the GEV.