A curb cut is a ramp that connects the sidewalk to a street crossing, making it accessible for physically disabled pedestrians. We employed ArcGIS Pro and machine learning algorithms in Python to classify a dataset of curb cut spectral signatures using Sentinel-2 imagery with 10-meter resolution. Initially, multispectral visible and near-infra-red (NIR) Sentinel-2 sensors and machine learning geoprocessing tools in ArcGIS Pro were employed for generating signatures from curb cuts manually identified using Google Earth in Hillsborough County, Florida. We then interpolated these capture points using a Python-modified Bayesian Maximum Likelihood Estimator to create a cross-county signature map. The resulting layer was overlaid onto a zip-code gridded land use land cover (LULC) map and analyzed using a semi-parametric eigendecomposition eigen-spatial filtering approach. The hot/cold spot residuals represented independent curb cut clustering propensities. Higher sub-resolution satellite signals can optimize the identification of LULC classifiable, zip-code gridded capture point signatures, which can improve the predictive mapping of other curb cut geolocations, such as those in school parking lots and homeless shelters. A real-time satellite mapping system can utilize sub-meter resolution data in a mobile application for retrieving a ranked list of visually similar curb cut geolocations.

Keywords: Curb cuts, curb ramps, satellite data, ArcGIS, Python, eigenvectors, Hillsborough County.