HiRO-ACEv2: Simulating extreme weather at 3 km resolution for a fraction of the cost

This post introduces HiRO-ACEv2, an updated system for simulating extreme weather at 3 km resolution. Key changes include a new downscaling model (HiROv2) capable of predicting multiple high-resolution fields simultaneously (beyond just rainfall) and improved skill in predicting extreme events. The training methodology for HiROv2 was modified to enhance its ability to reproduce intense rain, wind, and pressure values, achieving within 3% accuracy for rare events. The system also demonstrates improved skill in estimating tropical cyclone energy and generating realistic storm structures. An updated ACE model was also trained, showing good alignment with X-SHiELD for hurricanes, average climate patterns, and distributions of weather variables over a 10-year simulation. The combined HiRO-ACEv2 system generated a full year of global 3 km data with statistics closely matching X-SHiELD, at approximately 1/300th the cost. Future work includes training on new climate datasets and adding more output fields.