Michael Steininger

ConvMOS: Climate Model Output Statistics with Deep Learning

Climate models are the tool of choice for scientists researching climate change. Like all models they suffer from errors, specifically systematic and localized representation errors. One way to reduce these errors is model output statistics (MOS) where the model output is fitted to observational data with machine learning. In this work, we present our novel convolutional deep learning climate MOS approach ConvMOS which is specifically designed based on the observation that there are location-specific and systematic errors in the precipitation estimates of climate models. We apply ConvMOS to the simulated precipitation of the regional climate model REMO, showing that a combination of per-location model parameters – suited for reducing location-specific errors – and global model parameters – suited for reducing systematic errors – is indeed beneficial for MOS performance. Furthermore, we find that ConvMOS is able to reduce errors considerably and perform significantly better than three commonly used MOS approaches in most cases.