Predicting residential septic system malfunctions for targeted drone inspections

Abstract

Septic system malfunctions can cause untreated sewage to pond in yards or contaminate drinking water wells leading to environmental and health problems. While most malfunction detections rely on reports by individuals, machine learning and remote sensing can be used to identify potentially failing systems. We propose a methodology that combines a machine learning technique implemented in Maxent with unmanned aerial system (UAS) mapping to create a priority queue for inspection and detecting malfunctions apparent in the collected imagery. We demonstrate the approach in Wake County, North Carolina, a County with 73,347 septic systems located within drinking water supply watersheds. The predictive modeling identified 102 systems with a 99.9% probability of failure. Four properties from the queue were mapped by UAS and the acquired imagery was visually analyzed in the visible spectrum for signs of malfunction. Our results suggest that the proposed approach can assist in the early identification of failing systems minimizing the environmental impacts and saving resource time and funds.