Density-based spatial clustering of application with noise approach for regionalisation and its effect on hierarchical clustering
Ramgopal T. Sahu, Mani Kant Verma, Ishtiyaq Ahmad
Abstract
The study assesses DBSCAN-based cluster analysis to obtain a quantile with a minimum or low error (AR bias and R-RMSE). The Mahanadi River Basin was examined for spatiotemporal characteristics of precipitation records prior to testing the DBSCAN technique for regionalising precipitation. An application of the modified Mann-Kendall trend test and BEAST, i.e., Bayesian estimators of abrupt change, seasonality, and trend for studying the spatiotemporal behaviour of the study area. Furthermore, applying DBSCAN and the empirical orthogonal function analysis (EOF) to come up with an effective clustering solution. Gridded rainfall data at a resolution of 0.25° × 0.25° (1901-2017) obtained from IMD Pune is used for computing statistics to be used for precipitation regionalisation. The inaccuracy of ungauged site quantile estimations is contrasted with the error resulting from a hierarchical cluster analysis method that creates site-specific areas specifically for computing quantiles. DBSCAN clustering has identified some non-uniform patterns spread over the lower Mahanadi basin, suggesting lots of outliers (noise points) and small-sized clusters in small regions. This could create difficulties in assessing and designing policies for water resource management for different stakeholders.