Supervised Classification of Stinknet in Maricopa County, Arizona

Abstract

Stinknet (Oncosiphon pilulifer) is an invasive plant species in Arizona that has rapidly expanded across Maricopa County since 2016, becoming a significant noxious weed. Management efforts have combined fieldwork with remote sensing techniques. Among these, supervised classification of high-resolution drone imagery using machine learning has proven effective; Maricopa County has applied drone-based classification since 2023 to guide its treatment program. Satellite imagery has also shown promise, though its coarser spatial resolution has led to more limited use. This study evaluates the prediction skill of satellite and aircraft-based stinknet classification in Cave Creek Regional Park, comparing two imagery sources—National Agriculture Imagery Program (NAIP) and PlanetScope—against existing drone-based classification results from spring 2023. NAIP imagery offers sub-meter resolution but is only captured after the optimal flowering period of stinknet, while PlanetScope provides coarser three-meter resolution imagery available year-round. Classification was performed in ArcGIS Pro for each imagery source using both the random forest and support vector machine methods. A confusion matrix comparing each classification to the drone-derived dataset was generated to assess relative accuracy. Both NAIP and PlanetScope classifiers demonstrated statistically significant prediction skill, and PlanetScope classifiers achieved significantly more accurate results than NAIP classifiers.