Investigating environmental justice in stormwater utility fees at the parcel scale using an unmanned aircraft systems and machine learning.

Stormwater runoff is an increasing issue that coincides with urbanization. Impervious surfaces create a platform for nonpoint source pollution. Many cities in Texas use stormwater utility fees (SWU) to fund infrastructures to manage stormwater. While some cities calculate SWU by flat rates, others determine the total impervious surface on each residential property based on its impervious area. Geospatial information system methods are needed to calculate impervious surfaces using aerial images to determine the amount of impervious area. Many cities calculate SWU by multiplying a property's total area by a percentage factor and comparing the values to a tier system. However, this method creates a false representation of the actual impervious surface of each property. This study proposes and compares two machine learning methods for delineating impervious surfaces on residential units: (1) deep learning and (2) segmentation and supervised classification. Deep learning is a type of machine learning that uses artificial neural networks to perform tasks without a programming requirement. In this study, Mask R-CNN, a Region-based Convolutional Neural Network, is used to segment objects from aerial images to delineate impervious surfaces. Segmentation and supervised classification use pixel-based classification to differentiate between impervious and pervious surfaces. These methods are applied to imagery from an unmanned aircraft system in Corpus Christi, Texas. The images are processed to determine a property's actual total impervious surface. To validate the accuracy of the two methods, impervious surface was manually digitized to create a ground truth reference. The results from the two methods were compared to the ground truth, and error matrices were calculated. The overall accuracy for both methods was calculated above 90%. With high accuracy detection, an impervious percentage can be determined for each parcel, providing a proportional calculation between a property’s total impervious area and a homeowner’s SWU.