Temporal Super-Resolution, Ground Adjustment, and Advection Correction of Radar Rainfall Using 3-D-Convolutional Neural Networks

Abstract

Weather radars are highly sophisticated tools for quantitative precipitation estimation and provide observations with unmatched spatial representativeness. However, their indirect measurement of precipitation high above ground leads to strong systematic errors compared to direct rain gauge measurements. Additionally, the temporal undersampling from 5-minute instantaneous radar measurements requires advection correction. We present ResRadNet, a 3D-convolutional residual neural network approach, to reduce these errors and, at the same time, increase the temporal resolution of the radar rainfall fields by a 5-minute short-range prediction of 1-minute timesteps. The network is trained to process spatiotemporal sequences of radar rainfall estimates from a composite product derived from 17 C-band weather radars in Germany. In contrast to previous approaches, we present a method that emphasizes the generation of spatiotemporally consistent and advectioncorrected country-wide rainfall maps. Our approach significantly increased the Pearson correlation coefficient of the radar product (from 0.63 to 0.74) and decreased the root mean squared error by 22 percent when compared to 247 rain gauges at a 5-minute resolution. An additional large-scale comparison to 8 years of data from 1138 independent manual daily gauges confirmed that the improvement is robust and transferable to new locations. Overall, our study shows the benefits of using 3D convolutional neural networks for weather radar rainfall estimation to provide 1-minute, ground-adjusted, that is, bias-corrected with respect to on-ground sensors, and advection-corrected radar rainfall estimates.