A Bayesian Fuzzy Clustering Approach for Design of Precipitation Gauge Network Using Merged Remote Sensing and Ground‐Based Precipitation Products. Issue 2 (10th February 2022)

• Record Type: Journal Article
• Authors: Sreeparvathy, Vijay; Srinivas, V. V.
• Is Part Of: Water resources research. Volume 58:Issue 2(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2022-02-10
• Subjects: rain gauge network -- infinite Bayesian clustering -- fuzzy entropy -- random forest -- merging satellite precipitation products -- India; Hydrology -- Periodicals; 333.91
• DOI: 10.1029/2021WR030612
• Languages: English
• ISSNs: 0043-1397

Abstract

Abstract: A two‐level clustering approach is proposed for optimal design/expansion of a ground‐based precipitation monitoring network (GPN). It harnesses the advantages of Infinite Bayesian fuzzy clustering in the first level to partition the study area into homogeneous precipitation zones by considering structural/statistical characteristics and temporal variability of the observed precipitation. In the second level, an ensemble of hierarchical and partitional clustering techniques is considered in the time domain to effectively partition each zone into groups by considering weighted inter‐site dissimilarities of precipitation. The dissimilarities account for correlation, temporal dynamics, and fuzzy mutual information of precipitation at existing stations and possible new gauge locations. Key station's location in each group is identified by a proposed ranking procedure that accounts for population density, land‐use/landcover, and fuzzy marginal entropy of precipitation. For use with the approach, information on precipitation was derived for fine resolution ungauged grids covering the study area using random forest‐based regression relationships developed for gauged grids between merged multiple satellite‐based precipitation products (CHIRPS, IMERG) and ground‐based precipitation measurements. The potential of the proposed approach over other clustering‐based procedures is illustrated through a case study on a GPN comprising 1, 128 gauges in Karnataka state (191, 791 km 2 ) of India. Potential locations for installing new gauges and areas where there is scope for relocating existing stations are identified. The proposed methodology appears promising and could be extended to design networks monitoring various other hydrometeorological variables. Key Points: Proposes a two‐level clustering approach for optimal design/expansion of ground‐based precipitation monitoring network (GPN) Approach harnesses advantages of infinite Bayesian fuzzy clustering and an ensemble of hierarchical and partitional clustering methods Merged ground and multiple satellite‐based precipitation derived using random forest method is used to locate key stations to redesign GPN