Analog forecasting of extreme-causing weather patterns using deep learning

Abstract

Numerical weather prediction (NWP) models require ever-growing computing time/resources, but still, have difficulties with predicting weather extremes. Here we introduce a data-driven framework that is based on analog forecasting (prediction using past similar patterns) and employs a novel deep learning pattern-recognition technique (capsule neural networks, CapsNets) and impact-based auto-labeling strategy. CapsNets are trained on mid-tropospheric large-scale circulation patterns (Z500) labeled $0-4$ depending on the existence and geographical region of surface temperature extremes over North America several days ahead. The trained networks predict the occurrence/region of cold or heat waves, only using Z500, with accuracies (recalls) of $69\%-45\%$ $(77\%-48\%)$ or $62\%-41\%$ $(73\%-47\%)$ $1-5$ days ahead. CapsNets outperform simpler techniques such as convolutional neural networks and logistic regression. Using both temperature and Z500, accuracies (recalls) with CapsNets increase to $\sim 80\%$ $(88\%)$, showing the promises of multi-modal data-driven frameworks for accurate/fast extreme weather predictions, which can augment NWP efforts in providing early warnings.

Leveraging Spatial Patterns in Precipitation Forecasts Using Deep Learning to Support Regional Water Management

“…The same structure is used for all models within each class, as shown in Table 1. These combinations of layers and dimensions are similar to prior studies using deep learning with weather data for different applications over similar‐sized domains (Chattopadhyay et al., 2020; Chapman et al., 2019; Larraondo et al., 2019). In general, preliminary experiments suggest that model performance for higher lead times and quantiles tends to improve with more layers and parameters, though with increased risk of overfitting, which occurs when an increase in model complexity reduces performance on the test data.…”

“…Prediction tasks with image inputs are typically addressed with convolutional neural networks (CNNs), because they can detect patterns across scales, shapes, and locations relevant for prediction (Schultz et al., 2021). Recent studies have focused on advancing CNNs and other DL models that learn directly from spatial data, rather than manually extracted features (e.g., Chattopadhyay et al., 2020; Liu et al., 2016). Many of these efforts have centered around short‐term precipitation forecasting or nowcasting experiments (Castro et al., 2021; Hernández et al., 2016; Larraondo et al., 2019; Shi et al., 2017).…”

The Summertime Pacific‐North American Weather Regimes and Their Predictability

“…The cluster patterns are obtained by minimizing the Euclidian distance between iteratively updated cluster's pattern and their matching daily Z500 anomaly fields. It should be mentioned that there are a variety of methods that have been used for clustering weather patterns and it is not clear if any is better than the rest (e.g., see Ashok et al., 2017; Chattopadhyay, Hassanzadeh, & Pasha, 2020; Ghil & Robertson, 2002; Mo & Ghil, 1988; Robertson & Mechoso, 2003; Sahai et al., 2017; Straus, 2018). At least one recent study (Bao & Wallace, 2015) concluded that the flow regimes derived from SOM analysis are more distinctive and more robust than those obtained using Ward's method, which is a type of hierarchical clustering technique.…”

Self‐Supervised Classification of Weather Systems Based on Spatiotemporal Contrastive Learning

“…Therefore, CWS has been broadly applied in weather forecasts and statistical climatology. In ensemble forecasts, CWS is utilized to generate a set of pre‐defined circulation types and simplify the ensemble forecasts, where each weather pattern of ensemble members is assigned to the closest matching type, to a sequence of circulation type probabilities (Chattopadhyay et al., 2020; Neal et al., 2016; Ohba et al., 2018), hence reducing computational complexity while maximizing useful information. In climatology, classification helps examine climate‐scale changes in the frequency of circulation types (Gibson et al., 2016; Luong et al., 2020; Lynch et al., 2006).…”