Assessing Soil Sealing Dynamics in the Metropolitan Area of Bari Using Sentinel-2, Sealed Surface Index, and Machine Learning Approaches

Soil sealing, the irreversible covering of natural soil with impermeable material, impacts environmental sustainability by degrading ecosystems and reducing water quality and infiltration capacity. Advanced geospatial detecting and monitoring approaches are, thus, needed to face this critical environmental challenge. This study assesses the effectiveness of two advanced geomatics approaches in detecting the dynamics of sealed surfaces in the metropolitan area of Bari from 2018 to 2024 using Sentinel-2 satellite imageries processed in Google Earth Engine: the sealed surface index and the machine learning-based Classification and Regression Trees algorithm. Twenty-eight Sentinel-2 imageries, acquired during February, May, July, and October, were analyzed. When necessary, a cloud cover masking process was employed before classification. After that, the filtered maps were classified with the two above-mentioned techniques generating four land cover classes: sealed surfaces, vegetation, bare land, and water bodies. Their accuracy was assessed against the training data gathered from the land use map provided by ISPRA. The outcomes demonstrated that incorporating machine learning significantly improved classification accuracy. The sealed surface class exhibited the highest classification accuracy, with Producer’s and User’s Accuracy consistently exceeding 90% and 85%, respectively. Conversely, bare soil proved more challenging to classify, leading to slightly lower accuracy values. These findings highlight the annual variations in sealed surfaces and their correlation with seasonal land-use dynamics. A notable increase in urban sealing, exceeding 2%, was observed between February 2018 and October 2024. This study underscores the importance of continuous monitoring of soil sealing for sustainable urban planning and environmental management.