Assessing Vegetation Loss in Emerging Industrial Zones through Spatio-Temporal ANN Classification of Sentinel-2 Data

Abstract

Background: Land-use change has been developed as a critical environmental issue, primarily driven by population growth and the expansion of industrial zones. However, specific studies that examine the dynamics of land-cover change surrounding newly developed industrial areas, such as the Kendal Industrial Park (KIP), remain limited in the Indonesian context. Objective: This study aims to analyze land-cover changes in the vicinity of the KIP during the 2019–2023 period and to quantify the proportions of the transitions. Methods: This study adopted Sentinel-2 Level-2A surface reflectance imagery from the harmonized COPERNICUS/S2_SR collection. Land cover was classified using an Artificial Neural Network (ANN) with a binary sigmoid function and evaluated through standard metrics, followed by land-use change analysis across multiple buffer radii and village scales. Results: ANN showed consistently strong performance across all scenarios, achieving accuracy and AUC values of 0.83–0.85. Furthermore, the 75:25 train–test split provided the most balanced and generalizable results across replications. Spatio-temporal analysis indicated higher vegetation density within the 0.5–1 km core radius, a sharp decline in the 2–4 km transition zone, and a slight recovery at 6–7 km. Kumpulrejo Village also recorded the highest vegetation loss of –3.55% due to rice field conversion into industrial and residential areas. Generally, consistent 2019–2023 patterns identified the transition zone as the most vulnerable area to vegetation degradation. Conclusion: ANN effectively captures spatio-temporal land-cover dynamics, where industrial expansion drives vegetation loss, particularly within the 2–4 km transition zone. The results provide empirical evidence of the trade-off between economic growth and environmental sustainability, addressing a gap by examining newly developed industrial regions. Future studies should compare ANN with other deep learning models and integrate socio-economic data to better explain land-conversion drivers and enhance generalizability. Keywords: Industrial Development Impact, Kendal Industrial Park (KIP), Land Use Change, Spatio-Temporal Classification, Vegetation Loss