Combining distribution‐based neural networks to predict weather forecast probabilities

Mariana C.A. Clare,Cyril J. Morcrette,Omar Jamil

doi:10.1002/qj.4180

Mariana C.A. Clare, Cyril J. Morcrette + Show 1 more

Open Access

https://doi.org/10.1002/qj.4180

Copy DOI

Abstract

AbstractThe success of deep learning techniques over the last decades has opened up a new avenue of research for weather forecasting. Here, we take the novel approach of using a neural network to predict full probability density functions at each point in space and time rather than a single output value, thus producing a probabilistic weather forecast. This enables the calculation of both uncertainty and skill metrics for the neural network predictions, and overcomes the common difficulty of inferring uncertainty from these predictions. This approach is data‐driven and the neural network is trained on the WeatherBench dataset (processed ERA5 data) to forecast geopotential and temperature 3 and 5 days ahead. Data exploration leads to the identification of the most important input variables. In order to increase computational efficiency, several neural networks are trained on small subsets of these variables. The outputs are then combined through a stacked neural network, the first time such a technique has been applied to weather data. Our approach is found to be more accurate than some coarse numerical weather prediction models and as accurate as more complex alternative neural networks, with the added benefit of providing key probabilistic information necessary for making informed weather forecasts.

Highlights

For over 100 years, advanced mathematical techniques have been used for weather prediction
There are several methods to deal with this issue and in this work we focus on two: (i) using a meta-learner to combine the outputs from several neural networks trained individually; (ii) using data exploration techniques to identify the important variables in the dataset and discard the unimportant ones
Because we are applying the full stacked neural network approach, we must split the dataset in a different way to in Section 2 as there are two stages of neural networks

Summary

Introduction

For over 100 years, advanced mathematical techniques have been used for weather prediction. Numerical Weather Prediction (NWP) is an advanced discipline which uses some of the world’s largest supercomputers to solve complex non-linear differential equations. The forecast skill of these models has been improving by approximately one day every ten years, i.e. the 5-day forecast today is as accurate as the 4-day forecast was ten years ago (see Bauer et al, 2015). This improvement has been achieved through the scientific and technological development of both NWP models and computers (Bauer et al, 2015). Research has been mainly focused on the supervised learning techniques of neural networks

Methods

Results

Conclusion