Statistical post-processing of wind speed forecasts using convolutional neural networks

TL;DR

This paper introduces convolutional neural networks to improve probabilistic wind speed forecasts by utilizing spatial patterns from NWP models, outperforming traditional methods in accuracy and skill scores.

Contribution

It demonstrates the effectiveness of CNNs in spatial post-processing of wind forecasts and compares different density estimation methods, highlighting the superiority of quantized softmax.

Findings

CNNs improve Brier skill scores for wind speeds

CNNs outperform fully connected networks and quantile forests

Quantized softmax yields best probabilistic forecasts

Abstract

Current statistical post-processing methods for probabilistic weather forecasting are not capable of using full spatial patterns from the numerical weather prediction (NWP) model. In this paper we incorporate spatial wind speed information by using convolutional neural networks (CNNs) and obtain probabilistic wind speed forecasts in the Netherlands for 48 hours ahead, based on KNMI's deterministic Harmonie-Arome NWP model. The probabilistic forecasts from the CNNs are shown to have higher Brier skill scores for medium to higher wind speeds, as well as a better continuous ranked probability score (CRPS) and logarithmic score, than the forecasts from fully connected neural networks and quantile regression forests. As a secondary result, we have compared the CNNs using 3 different density estimation methods (quantized softmax (QS), kernel mixture networks, and fitting a truncated normal distribution), and found the probabilistic forecasts based on the QS method to be best.