Assessing Decadal Predictability in an Earth‐System Model Using Explainable Neural Networks

Abstract

We show that explainable neural networks can identify regions of oceanic variability that contribute predictability on decadal timescales in a fully coupled Earth‐system model. The neural networks learn to use sea‐surface temperature anomalies to predict future continental surface temperature anomalies. We then use a neural‐network explainability method called layerwise relevance propagation to infer which oceanic patterns lead to accurate predictions made by the neural networks. In particular, regions within the North Atlantic Ocean and North Pacific Ocean lend the most predictability for surface temperature across continental North America. We apply the proposed methodology to decadal variability, although the concept is generalizable to other timescales of predictability. Furthermore, while our approach focuses on predictable patterns of internal variability within climate models, it should be generalizable to observational data as well. Our study contributes to the growing evidence that explainable neural networks are important tools for advancing geoscientific knowledge.

Predicting Slowdowns in Decadal Climate Warming Trends With Explainable Neural Networks

“…However, these simulations can be computationally expensive to run. Alternatively, recent progress in machine learning has shown promising results for decadal climate applications, especially when combined with explainability methods (e.g., Gordon et al., 2021; G. Liu et al., 2021; Toms et al., 2021). Motivated by this new line of research, we explore the utility of a relatively shallow ANN for predicting temporary decadal warming slowdowns of GMST using upper OHC variability.…”

“…Deep learning methods, such as neural networks, have the ability to extract and leverage nonlinear patterns across data‐intensive spatial fields, which make them promising tools for revealing new insights and sources of predictability in climate science (Barnes, Mayer, et al., 2020; Irrgang et al., 2021; Reichstein et al., 2019; Sonnewald et al., 2021). Recent work has demonstrated the utility for neural networks in identifying climate modes, teleconnections, and forecasts of opportunity for a wide variety of timescales (e.g., Gibson et al., 2021; Gordon et al., 2021; Ham et al., 2019; J. Liu et al., 2021; Mayer & Barnes, 2021; Nadiga, 2021; Tang & Duan, 2021; Toms et al., 2021; Wu & Hsieh, 2004). Further, a growing number of explainable artificial intelligence (XAI) methods have been adapted for applications in weather and climate science (McGovern et al., 2019; Toms et al., 2020), which can retrospectively trace the decisions of neural networks and assist scientists in comparing the attribution of input features to known physical mechanisms in the Earth system.…”

“…Recent work has demonstrated the utility for neural networks in identifying climate modes, teleconnections, and forecasts of opportunity for a wide variety of timescales (e.g., Gibson et al, 2021;Gordon et al, 2021;Ham et al, 2019;J. Liu et al, 2021;Mayer & Barnes, 2021;Nadiga, 2021;Tang & Duan, 2021;Toms et al, 2021;Wu & Hsieh, 2004). Further, a growing number of explainable artificial intelligence (XAI) methods have been adapted for applications in weather and climate science (McGovern et al, 2019;Toms et al, 2020), which can retrospectively trace the decisions of neural networks and assist scientists in comparing the attribution of input features to known physical mechanisms in the Earth system.…”

Using an Explainable Machine Learning Approach to Characterize Earth System Model Errors: Application of SHAP Analysis to Modeling Lightning Flash Occurrence

“…Here, we include SHAP values for both cases where the XGBoost classifier is correct and incorrect. The choice of if the analysis should be conditioned on correct machine learning model predictions varies across applications (e.g., Lundberg et al, 2018;Toms et al, 2021). For comparison, we repeat the analysis in this section using only cases where the classifier is correct, and find consistent results (See Supporting Information S1).…”

Explainable Machine Learning Reveals Capabilities, Redundancy, and Limitations of a Geospatial Air Quality Benchmark Dataset

“…With their SHAP value analysis, they gain process understanding at individual air pollution measurement sites. Toms et al [33] apply an explainable neural network as a tool to identify patterns of Earth system predictability. Their neural network is trained to predict decadal oceanic variability and explained it by applying layer-wise relevance propagation [30].…”