Satellite-Based Urban Vegetation Loss Detection and Legal Permit Verification Using Super-Resolved Sentinel-2 Imagery

Abstract

Green spaces in cities like trees, parks, and vegetated land play a vital role in maintaining the ecological balance, moderating temperature, and ensuring urban biodiversity. However, there has been an increasing loss of these green spaces owing to rampant urbanization, making it difficult to monitor and enforce regulations concerning their loss. This study develops an automated satellite-based technique for detecting loss of urban vegetation and verifying if such loss occurs within the legal framework of approved permits. Multispectral imagery from Sentinel-2 sensors serves as the primary data source. However, to address its coarse spatial resolution of 10 meters, a pre-processing step that enhances image resolution using the S2DR3 deep learning model to around one meter resolution is performed. A semi-supervised dataset for training a machine learning algorithm to generate vegetation masks from images of selected spectral bands is developed through K-means clustering and spectral index analysis, followed by manual editing. The change detection process detects deforestation events using the two dates change detection method and morphological processing to remove artifacts. Finally, the detected deforested patches are converted to polygons and overlaid with the legally approved permit boundary polygons. According to the coverage ratio, areas can be categorized into either permitted or possibly illegal zones. This model will use remote sensing, artificial intelligence, and geographic information systems to enable its scalable implementation.