Downscaling Satellite Precipitation Estimates With Multiple Linear Regression, Artificial Neural Networks, and Spline Interpolation Techniques

Abstract

Satellite precipitation estimates (SPEs) have been widely used in various applications. However, when applied to small basins and regions, the spatial resolution of SPEs is too coarse. In this study, we present three downscaling algorithms based upon the relationships between SPEs and cloud optical and microphysical properties in northeast Austria. Different downscaling techniques, namely, multiple linear regression, artificial neural networks, and spline interpolation, were adopted for the downscaling of Integrated Multi-satellitE Retrievals for GPM (IMERG) precipitation data. In this respect, linear and nonlinear relationship among IMERG data and different cloud variables, such as cloud effective radius, cloud optical thickness, and cloud water path, was evaluated. Downscaled SPEs, as well as the original IMERG product, were subsequently validated using 54 rain gauges at a daily timescale. According to the results, all downscaled products were more accurate than the original IMERG data. Furthermore, all downscaling techniques captured the spatial patterns of precipitation reasonably well with more detailed information when compared with the original IMERG precipitation. However, the spline interpolation slightly outperformed the other techniques, followed by multiple linear regression and artificial neural network, respectively. Moreover, the proposed methods, which consistently showed increased correlation (e.g., from 0.30 to 0.56 for spline interpolation) and reduced mean absolute error and root-mean-square error (e.g., from 10.14 to 6.55 mm and 13.5 to 8.76 mm, respectively) for average of all events, can more accurately produce downscaled precipitation data.