Sai Ravela, Joaquin Salas, Anamitra Saha, A Machine Learning Approach to Surface Roughness Length Estimation

In this project, we estimate surface roughness length (z0), a critical boundary layer parameter. Our results are provided at 1km spatial resolution and are meant for general use. The inputs to our surface roughness length estimator uses Land Use Land Cover and digital elevation maps (GTOPO30-1km or SRTM-30m), obtained from Earth Explorer, along with a large number of  tables gleaned from literature (GWA, 1, 2, 3, Davenport Roughness Classification [also see here], FEMA HAZUS) to obtain training inputs. The current learning algorithm used is based on extreme gradient boosting. To obtain a roughness map (with same grid as USGS LULC tiff) use this link.(ESSG CGC Z0.mat v0.1)

Our main objective is to improve near-surface wind-field estimation, where surface roughness is a critical parameter, to project cyclone-induced wind hazard and risk maps (10m  sustained 2min winds and 3s gusts at 10, 50 and 100 return year periods) for the MIT Climate Grand Challenges project. Details of the gradient windfield and boundary layer model, including terrain-induced speed modifications will be described in a separate entry. We are also interested in using the estimates for overland flows in inundation mapping.