Machine Learning for Model Error Inference and Correction

Bonavita, Massimo; Laloyaux, Patrick

doi:10.1002/essoar.10503695.1

Cited by 7 publications

(8 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, in the current version of WC‐4DVar the model error control vector does not contain a humidity variable. This is based on earlier findings that no significant large‐scale biases were present in the short‐range humidity forecasts, which was confirmed in more recent studies (Bonavita and Laloyaux, 2020). Background departures for humidity‐sensitive instruments are generally improved (both in the mean and the standard deviation) in the WC‐4DVar experiments, with the notable exception of conventional humidity observations (from radiosonde and aircraft) in the lower troposphere.…”

Section: Discussionsupporting

confidence: 82%

“…Another way to deal with systematic model errors is through the use of Artificial Neural Networks (ANN), either on their own or through hybrid ANN‐WC4DVar configurations (Bonavita and Laloyaux, 2020). The results presented in this work indicate at least two interesting avenues for further development.…”

Section: Discussionmentioning

confidence: 99%

“…This idea has a long history in the meteorological literature, for example, Leith (1978), and has been more recently revived in both simplified and operational analysis and modelling frameworks (e.g. Carrassi and Vannitsem, 2010; Mitchell and Carrassi, 2015; Piccolo and Cullen, 2015; Bonavita and Laloyaux, 2020; Crawford et al ., 2020). These more recent works are typically based on an offline model error estimation framework; a database of analysis increments (analysis−background fields) is used to estimate tendency corrections that can be used during the cycling of the assimilation system or in free forecast mode (note that while the estimation procedure is offline, the estimated corrections are usually flow‐ and/or time‐dependent).…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Exploring the structure of time‐correlated model errors in the ECMWF data assimilation system

Bonavita

2021

Quart J Royal Meteoro Soc

Self Cite

View full text Add to dashboard Cite

Model errors are increasingly seen as a fundamental performance limiter in both Numerical Weather Prediction and Climate Prediction simulations run with state‐of‐the‐art Earth‐system digital twins. This has motivated recent efforts aimed at estimating and correcting the systematic, predictable components of model error in a consistent data assimilation framework. While encouraging results have been obtained with a careful examination of the spatial aspects of the model error estimates, less attention has been devoted to the time correlation aspects of model errors and their impact on the assimilation cycle. In this work we employ a Lagged Analysis Increment Covariance (LAIC) diagnostic to gain insight in the temporal evolution of systematic model errors in the ECMWF operational data assimilation system, evaluate the effectiveness of the current weak constraint 4D‐Var algorithm in reducing these types of errors and, based on these findings, start exploring new ideas for the development of model error estimation and correction strategies in data assimilation.

show abstract

Section: Discussionsupporting

confidence: 82%

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Exploring the structure of time‐correlated model errors in the ECMWF data assimilation system

Bonavita

2021

Quart J Royal Meteoro Soc

Self Cite

View full text Add to dashboard Cite

show abstract

“…Pathak et al [65] and Watson [66] used ML coupled with a dynamical system model for improved accuracy and forecasting horizons of chaotic systems. Bonavita & Laloyaux [67] used ML to extend current data assimilation capabilities in operational state-of-the-art forecasting systems. Farchi et al [68] used ML to correct model error in data assimilation and forecasting.…”

Section: Physics-informed Machine Learning: Objectives Approaches Applicationsmentioning

confidence: 99%

Physics-informed machine learning: case studies for weather and climate modelling

Kashinath

Mustafa

Albert

et al. 2021

Phil. Trans. R. Soc. A.

298

143

View full text Add to dashboard Cite

Machine learning (ML) provides novel and powerful ways of accurately and efficiently recognizing complex patterns, emulating nonlinear dynamics, and predicting the spatio-temporal evolution of weather and climate processes. Off-the-shelf ML models, however, do not necessarily obey the fundamental governing laws of physical systems, nor do they generalize well to scenarios on which they have not been trained. We survey systematic approaches to incorporating physics and domain knowledge into ML models and distill these approaches into broad categories. Through 10 case studies, we show how these approaches have been used successfully for emulating, downscaling, and forecasting weather and climate processes. The accomplishments of these studies include greater physical consistency, reduced training time, improved data efficiency, and better generalization. Finally, we synthesize the lessons learned and identify scientific, diagnostic, computational, and resource challenges for developing truly robust and reliable physics-informed ML models for weather and climate processes. This article is part of the theme issue ‘Machine learning for weather and climate modelling’.

show abstract

“…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].…”

Section: Learning New Earth System Physicsmentioning

confidence: 99%

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

Geer

2021

Phil. Trans. R. Soc. A.

119

View full text Add to dashboard Cite

Recent progress in machine learning (ML) inspires the idea of improving (or learning) earth system models directly from the observations. Earth sciences already use data assimilation (DA), which underpins decades of progress in weather forecasting. DA and ML have many similarities: they are both inverse methods that can be united under a Bayesian (probabilistic) framework. ML could benefit from approaches used in DA, which has evolved to deal with real observations—these are uncertain, sparsely sampled, and only indirectly sensitive to the processes of interest. DA could also become more like ML and start learning improved models of the earth system, using parameter estimation, or by directly incorporating machine-learnable models. DA follows the Bayesian approach more exactly in terms of representing uncertainty, and in retaining existing physical knowledge, which helps to better constrain the learnt aspects of models. This article makes equivalences between DA and ML in the unifying framework of Bayesian networks. These help illustrate the equivalences between four-dimensional variational (4D-Var) DA and a recurrent neural network (RNN), for example. More broadly, Bayesian networks are graphical representations of the knowledge and processes embodied in earth system models, giving a framework for organising modelling components and knowledge, whether coming from physical equations or learnt from observations. Their full Bayesian solution is not computationally feasible but these networks can be solved with approximate methods already used in DA and ML, so they could provide a practical framework for the unification of the two. Development of all these approaches could address the grand challenge of making better use of observations to improve physical models of earth system processes. This article is part of the theme issue ‘Machine learning for weather and climate modelling’.

show abstract

Machine Learning for Model Error Inference and Correction

Cited by 7 publications

References 25 publications

Exploring the structure of time‐correlated model errors in the ECMWF data assimilation system

Exploring the structure of time‐correlated model errors in the ECMWF data assimilation system

Physics-informed machine learning: case studies for weather and climate modelling

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

Contact Info

Product

Resources

About