Graduate Research Assistant Marquette University, Wisconsin, United States
Abstract Submission: Green stormwater infrastructure (GSI) plays a critical role in urban water management by reducing runoff, improving water quality, and enhancing environmental sustainability. Accurate classification of GSI is essential for monitoring and assessing the effectiveness of its ecosystem functions, yet existing methods to do so are often based upon subjective and time consuming in person inspections. The objective of this study is to use remote sensing to monitor GSI and classify the vegetation and land cover using machine learning for more effective GSI assessment. To do so, multispectral remote sensing data was collected from both drones and satellites at 17 green infrastructure sites in Milwaukee, WI. Using remote sensing data, unsupervised learning models were applied to categorize the land cover of GSI into condition-related classes such as healthy plants, unhealthy plants, dead plants, organic material, and inorganic materials. Specifically, the machine-learning techniques included, spectral, texture, and object-based methods. Preliminary results indicate that unsupervised learning algorithms effectively produce soft and hard clusters that can be applied to identify land use categories with different spectral reflectance. Additional data fusing techniques between drone and satellite images and comparison of unsupervised learning approaches will be presented. The main outcome of this study is a novel approach to assess the maintenance needs of GSI using remote sensing data, that can ultimately be used to reduce the costs associated with manual monitoring and gain a better understanding of how to optimize maintenance procedures.
