Exploring the ability of regional extrapolation for precipitation nowcasting with deep learning

Abstract

Precipitation nowcasting refers to the prediction of precipitation intensity in a local region and in a short timeframe up to 6 hours. The evaluation of spatial and temporal information still challenges state-of-the-art numerical weather prediction models. The increasing possibilities to store and evaluate data combined with the advancements in the developments of artificial intelligence algorithms make it natural to use these methods to improve precipitation nowcasting. In this work, a Trajectory Gated Recurrent Unit (TrajGRU) is applied to radar data of the German Weather Service. The impact of finetuning a network pretrained at a different location and for several precipitation intensity thresholds with respect to the training time is evaluated. In cases with little availability of training data at the target location, for example when heavy rainfall is rare, the finetuned model can benefit from the original model performance at the pretraining location. Furthermore, the skill scores for the different thresholds are shown for a prediction time up to 100 minutes. The results highlight promising regional extrapolation capabilities for such neural networks for precipitation nowcasting.