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.

翻译：地球静止卫星(GOES)图像为热带气旋(TC)行为提供了一个高时间分辨率窗口。我们提出了一种新的方法,不仅对三角气旋强度进行概率预测,而且还对三角对流结构进行预测。本文描述了一个完全基于观测红外图像和以往操作强度估计的原型模型。这些结构预测模拟了云顶温度辐射剖面在随后12小时内的瞬时演化。我们的结构预测管道采用了深自动递增生成模型(PixelSNAIL),以创造这些功能在一段时间内演变的概率预测。一个标准的革命神经网络(经过“播送”培训,或根据过去和当前特征预测当前状况)随后被用于模拟结构轨迹,以预测在6至12小时内云顶温度的辐射强度变化。我们原型模型的强度指导比国家飓风中心的官方预测略高一点。然而,我们的原型模型没有考虑到诸如垂直风沙滩和海平面温度调整等环境因素的短期模型。我们证明,它有可能通过可合理预测的海面温度变图解。