Abstract
The accuracy of gridded precipitation data depends on the availability of a uniformly spaced rain gauge network and an appropriate spatial interpolation method that considers the rainfall variability and other factors that influence the precipitation patterns in the region of interest. In the current study, conceptually superior variants of a widely used spatial interpolation algorithm, Shepard's method, are proposed, formulated and evaluated to overcome one of the major limitations in neighbourhood selection, that is, arbitrary selection of rain gauges. The variants provide mechanisms to objectively select the rain gauges (control points) based on correlation (variant 1), distribution similarity (variant 2) and a combination of both (variant 3). The improved variants were used in the development of gridded rainfall data at a resolution of 5 km over the Kabini River basin in south India, and in the state of Kentucky, United States. Results from multiple experiments using the original Shepard's method and its variants indicate improvements in the accuracy of precipitation estimates. Also, these variants have preserved the site‐specific statistics and distributional characteristics of the rainfall data. A variant 1 that uses a correlation‐based neighbourhood selection criterion performed better for daily and monthly data compared to others and is suitable for generation of gridded rainfall data. The variant 1 when used with information from clustering of sites for selection of the neighbours has led to improvement in gridded precipitation data estimates. The proposed variant 1 can also be used for point data estimation useful for filling missing data at any site.