Structural Forecasting for Short-Term Tropical Cyclone Intensity Guidance

Abstract

Geostationary satellite (GOES) imagery provides a high temporal resolution window into tropical cyclone (TC) behavior. We present a new method for probabilistic forecasting not only of TC intensity but also TC convective structure as revealed by GOES. This paper describes a prototype model based solely on observed infrared imagery and past operational intensity estimates. These structural forecasts simulate the spatio-temporal evolution of the radial profiles of cloud-top temperatures over the subsequent 12 hours. Our structural forecasting pipeline applies a Deep Autoregressive Generative Model (PixelSNAIL) to create probabilistic forecasts of the evolution of these functions over time. A standard convolutional neural network (trained for ‘nowcasting’, or predicting the current state based on past states and current features) is then applied to simulated structural trajectories to predict intensities at lead times of 6 to 12 hours. Intensity guidance from our prototype model achieves a marginally higher error than the National Hurricane Center’s official forecasts. However, our prototype model does not account for environmental factors such as vertical wind shear and sea surface temperature. We demonstrate that it is possible to reasonably predict short-term evolution of TC convective structure as depicted by infrared imagery, producing interpretable structural forecasts that may be valuable for TC operational guidance.