Analog Forecasting of Extreme‐Causing Weather Patterns Using Deep Learning

Abstract

Numerical weather prediction models require ever‐growing computing time and resources but, still, have sometimes difficulties with predicting weather extremes. 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 an impact‐based autolabeling strategy. Using data from a large‐ensemble fully coupled Earth system model, CapsNets are trained on midtropospheric 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. Using both surface temperature and Z500, accuracies (recalls) with CapsNets increase to ∼80% (88%). In both cases, CapsNets outperform simpler techniques such as convolutional neural networks and logistic regression, and their accuracy is least affected as the size of the training set is reduced. The results show the promises of multivariate data‐driven frameworks for accurate and fast extreme weather predictions, which can potentially augment numerical weather prediction efforts in providing early warnings.

Predicting clustered weather patterns: A test case for applications of convolutional neural networks to spatio-temporal climate data

“…Here, we show the performance of CNNs for a problem that can be of practical importance: predicting the evolution of spatio-temporal climate/environmental patterns in the context of the cluster indices. Such cluster-based data-driven forecasting using machine learning methods or other techniques has been of rising interest in recent years 8,12,20,41 . Clustered precipitation or surface temperature patterns provide geographically cohesive regions of interest while clustered Z500 patterns often have connections with modes of climate variability.…”

“…Note that here we attempt to predict Z500 only from knowing the earlier Z500 pattern. Including more variables, e.g., geopotential heights at other pressure levels, sea surface temperature, etc., might improve the prediction accuracy, especially at longer leads (see the discussion in Chattopadhyay et al 20 ). We leave this to future work.…”

“…Classifying, identifying, and predicting specific patterns or key features in spatio-temporal climate and environmental data are of great interest for various purposes such as finding circulation regimes and teleconnection patterns [1][2][3][4][5] , identifying extreme-causing weather patterns [6][7][8][9][10][11][12] , studying the effects of climate change [13][14][15][16] , understanding ocean-atmosphere interaction 8,17,18 , weather forecasting 8,12,19,20 , and investigating air pollution transport 21,22 , just to name a few. Such classifications/identifications and predictions are often performed by employing empirical orthogonal function (EOF) analysis, clustering algorithms (e.g., K-means, hierarchical, self-organizing maps 1,3,[23][24][25][26][27][28][29] ), linear regression, or specifically designed indices, such as those used to identify atmospheric blocking events.…”

“…In recent years, applications of machine learning methods for accelerating and facilitating scientific discovery have increased rapidly in various research areas. For example, in climate science, neural networks have produced promising results for parameterization of convection and simulation of clouds [32][33][34][35][36] , and forecasting of weather/ climate variability and extremes and weather forecasting 12,20,[37][38][39][40][41][42] ; also see the recent Perspective by Reichstein et al 43 . A class of supervised deep learning architectures, called convolutional neural networks (CNN), has 1.…”

Improving data-driven global weather prediction using deep convolutional neural networks on a cubed sphere

“…Deep NNs have also been used to identify extreme weather and climate patterns in observed and modeled atmospheric states [24,19,20], to predict extreme weather events [e.g., 15], and to provide operational guidance and risk assessment for severe weather [26]. [21] developed deep NNs to extract spatial patterns in precipitation from gridded atmospheric fields, while [6] showed that deep NNs can skillfully predict extreme heat patterns several days ahead with relatively minimal input information. Another machine-learning effort has focused on the improvement of physics parameterizations in GCMs for both weather forecasting and climate prediction [4,32].…”