Identifying Data Sources and Physical Strategies Used by Neural Networks to Predict TC Rapid Intensification

Abstract The rapid intensification (RI) of tropical cyclones (TCs), defined here as an intensity increase of ≥30 kt (1 kt ≈ 0.51 m s −1 ) in 24 h, is a difficult but important forecasting problem. Operational RI forecasts have considerably improved since the late 2000s, largely thanks to better statistical models, including machine learning (ML). Most ML applications use scalars from the Statistical Hurricane Intensity Prediction Scheme (SHIPS) development dataset as predictors, describing the TC history, near-TC environment, and satellite presentation of the TC. More recent ML applications use convolutional neural networks (CNNs), which can ingest full satellite images (or time series of images) and freely “decide” which spatiotemporal features are important for RI. However, two questions remain unanswered: 1) Does image convolution significantly improve RI skill? 2) What strategies do CNNs use for RI prediction—and can we gain new insights from these strategies? We use an ablation experiment to answer the first question and explainable artificial intelligence (XAI) to answer the second. Convolution leads to only a small performance gain, likely because, as revealed by XAI, the CNN’s main strategy uses image features already well described in scalar predictors used by preexisting RI models. This work makes three additional contributions to the literature: 1) Neural networks (NNs) with SHIPS data outperform preexisting models in some aspects; 2) NNs provide well calibrated uncertainty quantification (UQ), while preexisting models have no UQ; 3) the NN without SHIPS data performs surprisingly well and is fairly independent of preexisting models, suggesting its potential value in an operational ensemble. Significance Statement Rapid intensification (RI) is a hazardous event and is difficult to forecast. RI forecasting could potentially be improved by machine learning (ML) models that ingest full satellite images—or even time series of images—centered on a tropical cyclone (TC). We investigate whether this improvement is actually realized and whether explainable ML can teach us anything new about RI precursors in satellite imagery.