Highlighted progress on creating machine learning three-dimensional hurricane wind field from the surface to upper levels. The purpose of this project is to provide 3D wind information to hurricane models, much like Tail Doppler Radar (TDR) based winds, but for all hurricane cases. Use of TDR in the HWRF model has resulted in measurable improvements in forecasts. However, TDR is only available when NOAA research aircraft are conducting hurricane reconnaissance, which typically represents a few percent (<10%) of the hurricane cases even in the Atlantic. The algorithm is developed using single-field principal component analysis, which is a regression using storm characteristics (center, intensity, translation speed, and heading) and principal components (spatial patterns of variance) of the infrared image. Inputs to the algorithm are tropical cyclone characteristics (all which are available in real time) and a single infrared image (i.e., 25 principal components). The initial version of the retrieval algorithm captures the overall structure of the radius of maximum wind and other critical wind radii as shown in the figure below. Future work will entail model updates based on additional comparisons with other independent data like atmospheric motion vectors, tail Doppler winds, and additional cases.