Exploring Bayesian deep learning for weather forecasting with the Lorenz 84 system

Yang Liu; Jisk Attema; Wilco Hazeleger

doi:10.5281/zenodo.4146850

Exploring Bayesian deep learning for weather forecasting with the Lorenz 84 system

Yang Liu, Jisk Attema, Wilco Hazeleger

Research output: Contribution to conference › Poster › Academic

Abstract

Recent developments in deep learning have led to many new neural networks potentially applicable to weather forecasting. However, these techniques are always based on deterministic deep neural networks (DNN) and therefore prone to over-confident forecasts. This brings Bayesian deep learning (BDL) into our scope. In this study, we use Bayesian Long-Short Term Memory neural networks (BayesLSTM) to forecast output from the Lorenz 84 system with seasonal forcing, so as to examine if BDL is useful for weather forecast.

Original language	English
DOIs	https://doi.org/10.5281/zenodo.4146850
Publication status	Published - 30 Sept 2020
Event	ECMWF-ESA Workshop on Machine Learning for Earth System Observation and Prediction - Online Duration: 5 Oct 2020 → 8 Oct 2020

Workshop

Workshop	ECMWF-ESA Workshop on Machine Learning for Earth System Observation and Prediction
Period	5/10/20 → 8/10/20

Access to Document

10.5281/zenodo.4146850Licence: CC BY

https://edepot.wur.nl/572637Licence: CC BY

Cite this

@conference{ff10174cd8f74c5cb248ff80d31754a3,

title = "Exploring Bayesian deep learning for weather forecasting with the Lorenz 84 system",

abstract = "Recent developments in deep learning have led to many new neural networks potentially applicable to weather forecasting. However, these techniques are always based on deterministic deep neural networks (DNN) and therefore prone to over-confident forecasts. This brings Bayesian deep learning (BDL) into our scope. In this study, we use Bayesian Long-Short Term Memory neural networks (BayesLSTM) to forecast output from the Lorenz 84 system with seasonal forcing, so as to examine if BDL is useful for weather forecast.",

author = "Yang Liu and Jisk Attema and Wilco Hazeleger",

year = "2020",

month = sep,

day = "30",

doi = "10.5281/zenodo.4146850",

language = "English",

note = "ECMWF-ESA Workshop on Machine Learning for Earth System Observation and Prediction ; Conference date: 05-10-2020 Through 08-10-2020",

}

TY - CONF

T1 - Exploring Bayesian deep learning for weather forecasting with the Lorenz 84 system

AU - Liu, Yang

AU - Attema, Jisk

AU - Hazeleger, Wilco

PY - 2020/9/30

Y1 - 2020/9/30

N2 - Recent developments in deep learning have led to many new neural networks potentially applicable to weather forecasting. However, these techniques are always based on deterministic deep neural networks (DNN) and therefore prone to over-confident forecasts. This brings Bayesian deep learning (BDL) into our scope. In this study, we use Bayesian Long-Short Term Memory neural networks (BayesLSTM) to forecast output from the Lorenz 84 system with seasonal forcing, so as to examine if BDL is useful for weather forecast.

AB - Recent developments in deep learning have led to many new neural networks potentially applicable to weather forecasting. However, these techniques are always based on deterministic deep neural networks (DNN) and therefore prone to over-confident forecasts. This brings Bayesian deep learning (BDL) into our scope. In this study, we use Bayesian Long-Short Term Memory neural networks (BayesLSTM) to forecast output from the Lorenz 84 system with seasonal forcing, so as to examine if BDL is useful for weather forecast.

U2 - 10.5281/zenodo.4146850

DO - 10.5281/zenodo.4146850

M3 - Poster

T2 - ECMWF-ESA Workshop on Machine Learning for Earth System Observation and Prediction

Y2 - 5 October 2020 through 8 October 2020

ER -

Exploring Bayesian deep learning for weather forecasting with the Lorenz 84 system

Abstract

Workshop

Access to Document

Fingerprint

Cite this