Machine Learning for Model Error Inference and Correction

Abstract

Model error is one of the main obstacles to improved accuracy and reliability in numerical weather prediction (NWP) and climate prediction conducted with state-of-the-art, comprehensive high-resolution general circulation models. In a data assimilation framework, recent advances in the context of weak-constraint 4D-Var have shown that it is possible to estimate and correct for a large fraction of systematic model error which develops in the stratosphere over short forecast ranges. The recent explosion of interest in machine learning/deep learning technologies has been driven by their remarkable success in disparate application areas. This raises the question of whether model error estimation and correction in operational NWP and climate prediction can also benefit from these techniques. In this work, we aim to start to give an answer to this question. Specifically, we show that artificial neural networks (ANNs) can reproduce the main results obtained with weak-constraint 4D-Var in the operational configuration of the IFS model of the European Centre for Medium-Range Weather Forecasts (ECMWF). We show that the use of ANN models inside the weak-constraint 4D-Var framework has the potential to extend the applicability of the weak-constraint methodology for model error correction to the whole atmospheric column. Finally, we discuss the potential and limitations of the machine learning/deep learning technologies in the core NWP tasks. In particular, we reconsider the fundamental constraints of a purely data-driven approach to forecasting and provide a view on how to best integrate machine learning technologies within current data assimilation and forecasting methods. Plain Language Summary Model error is one of the main obstacles to improved accuracy and reliability in current numerical weather prediction and in climate prediction. Recent advances in data assimilation at the European Centre for Medium-Range Weather Forecasts (ECMWF) indicate that it is possible to estimate and correct for a large fraction of systematic model error in the stratosphere. The question we address here is whether machine learning techniques can be used alone and in conjunction with standard data assimilation methods to improve on those results. We show that it is indeed possible to extend current data assimilation capabilities in operational state-of-the-art forecast systems using machine learning tools, and we discuss the potential and limitations of future applications of these ideas to other core NWP tasks.

Learning earth system models from observations: machine learning or data assimilation?

“…Weak-constraint DA [62] is similar, in that it does not improve the forward model, but estimates a spatial field of model errors. ML could be equally applicable to learning this kind of model error [102]. However, in weak-constraint DA, it can be hard to separate these errors from errors in the state, if they occur on similar spatial scales [63].…”

Deep Learning for Geophysics: Current and Future Trends

“…Researchers used DL to estimate ground-level PM2.5 or PM10 levels by using satellite observations and station measurements (Li et al, 2017;Shen et al, 2018;Tang et al, 2018). DL also helps improve the accuracy of weather forecasting, which is a long-standing challenge in atmospheric science (Bonavita & Laloyaux, 2020;Scher & Messori, 2021). The tracks of typhoons were predicted with a GAN based on satellite images (Rüttgers et al, 2019).…”

“…With multiple realizations of dropout, the results are collected, and the variance is computed as the uncertainty. DL with uncertainty estimation in inference is reported in areas such as volcano-seismic monitoring (Bueno et al, 2019), geomagnetic storm forecasting (Tasistro-Hart et al, 2020), weather forecasting (Scher & Messori, 2021;Bonavita & Laloyaux, 2020), soil moisture predictions (Fang, Kifer, et al, 2020) and earthquake locations estimation (Mousavi & Beroza, 2020b).…”

Combining data assimilation and machine learning to infer unresolved scale parametrization

“…With spatially dense and noise-free data, this approach has been based on sparse regression [9], echo state networks [10,11], recurrent neural networks (NN) [12], residual neural network [13] or convolutional neural networks [14,15]. The challenging problem of partial and/or noisy observations has been addressed using dedicated NN architecture [16] or in combination with DA methods [17][18][19][20][21].…”