ABSTRACT
Regionalization is highly beneficial when there is lack of gauging stations or historical data and is also beneficial in estimating the data at ungauged sites. Droughts, which are a frequently occurring natural phenomenon in many parts of the world, have been studied using a regionalized approach. Standardized precipitation indices (SPIs) for different timescales are computed using rainfall data. Fuzzy c-means clustering algorithm, which is a soft partitional clustering algorithm, has been used for the identification of homogenous drought regions based on drought indices and geographical attributes. Five cluster validity indices, namely, partition coefficient, partition entropy, extended Xie–Beni index, Fukuyama–Sugeno index, and Kwon index, were used to find the optimum value of the number of clusters and fuzzifiers. The regions were checked for homogeneity using L-moments-based homogeneity tests. The whole study area was divided into three regions: west, east, and south. These regions resemble the actual rainfall zones of the state. Further, the wavelet regression approach was used for the prediction of drought indices at the region level. Root mean square error (RMSE), coefficient of correlation (R), and Nash–Sutcliffe efficiency index (NS) were used to evaluate the accuracy of the model for regional predictions. These regional predictions can be extended for the estimation of the drought indices at ungauged stations within the regions.
