RainNet v1.0: a convolutional neural network for radar-based precipitation nowcasting

Abstract. In this study, we present RainNet, a deep convolutional neural network for radar-based precipitation nowcasting. Its design was inspired by the U-Net and SegNet families of deep learning models which were originally designed for binary segmentation tasks. RainNet was trained to predict continuous precipitation intensities at a lead time of five minutes, using several years of quality-controlled weather radar composites provided by the German Weather Service (DWD). That data set covers Germany with a spatial domain of 900 by 900 km, and has a resolution of 1 km in space and 5 minutes in time. Independent verification experiments were carried out on eleven summer precipitation events from 2016 to 2017. In order to achieve a lead time of one hour, a recursive approach was implemented by using RainNet predictions at five minutes lead time as model input for longer lead times. In the verification experiments, trivial Eulerian persistence and a conventional model based on optical flow served as benchmarks. The latter is available in the rainymotion library, and had previously been shown to outperform DWD's operational nowcasting model for the same set of verification events. RainNet significantly outperforms the benchmark models at all lead times up to 60 minutes for the routine verification metrics Mean Absolute Error (MAE) and Critical Success Index (CSI, at intensity thresholds of 0.125, 1, and 5 mm/h). Apart from its superiority in terms of MAE and CSI, an undesirable property of RainNet predictions is, however, the level of spatial smoothing. At a lead time of five minutes, an analysis of Power Spectral Density confirmed a significant loss of spectral power at length scales of 16 km and below. Obviously, RainNet had learned an optimal level of smoothing to produce a nowcast at 5 minutes lead time. In that sense, the loss of spectral power at small scales is informative, too, as it reflects the limits of predictability as a function of spatial scale. Beyond the lead time of five minutes, however, the increasing level of smoothing is a mere artifact -- an analogue to numerical diffusion -- that is not a property of RainNet itself, but of its recursive application. In the context of early warning, the smoothing is particularly unfavourable since pronounced features of intense precipitation tend to get lost over longer lead times. Hence, we propose several options to address this issue in prospective research, including an adjustment of the loss function for model training, model training for longer lead times, and the prediction of threshold exceedance in terms of a binary segmentation task. Furthermore, we suggest additional input data that could help to better identify situations with imminent precipitation dynamics. The model code, pretrained weights, and training data are provided in open repositories as an input to such future studies.