²ÝÝ®ÊÓƵ

Adverse weather conditions, particularly, high winds, can have a highly adverse impact on small unmanned aircraft system (sUAS) operations. These conditions can vary significantly within a small area (particularly, in an urban environment); thus, hyperlocal weather predictions are often necessary in order to determine whether a particular sUAS route will be safe to fly. The General Urban area Microclimate Predictions tool (GUMP) seeks to provide such predictions through the use of machine learning (ML) models and computational fluid dynamics (CFD) simulations. Specifically, ML models are trained to ingest mesoscale forecasts from the National Oceanic and Atmospheric Administration (NOAA) and output refined forecasts for some specific location, typically, a weather station that serves as a source of ground truth data during training. At the same time, CFD simulations over 3D models of structures (e.g., buildings) are utilized to extend the refined forecast to other points within the area of interest surrounding the location. Because it is difficult to perform such simulations in real-time, they are executed offline under a wide range of boundary conditions, generating a broad set of resulting wind flow fields. During deployment, GUMP retrieves the wind flow field that is most consistent with the ML model’s forecast. The wind flow field can be converted into an intuitive risk map for sUAS operators through the use of appropriate thresholds on wind velocities. I addition to NASA, additional partners on this project are Intelligent Automation Inc. and AvMet.