Long-Term Vegetation Dynamics in Managed Green Spaces under Suburban Land-Use Pressure: A Comparative Study of Two University Campuses in Bangladesh using Machine Learning

Semi-managed green spaces such as university campuses can play an important role in retaining vegetation within rapidly urbanizing landscapes, although their long-term land-cover changes remain relatively understudied. This study examines long-term changes in vegetation and built-up land at both the University of Chittagong (CU), Hathazari, and Jahangirnagar University (JU), Savar, Bangladesh, and compares these changes with land-use changes in their surrounding areas. Landsat observations spanning 1990-2025 were classified into four land-cover classes: vegetation, built-up area, bare land, and water. Random Forest and Support Vector Machine classifiers were used for land-cover classification and accuracy assessment. For CU, the 1990 classification produced an overall accuracy of 67.6% (Kappa = 0.46), mainly because historical reference data were limited and spectral mixing was greater in the earlier imagery. Subsequent CU classifications achieved overall accuracies of 84.2–91.5%, with Kappa values of 0.68-0.88. Vegetation cover at CU increased from 64.1% (291.9 ha) in 1990 to 88.9% (405.0 ha) in 2010 and was 87.5% (398.5 ha) in 2025. At JU, vegetation cover increased from 54.3% (154.4 ha) in 1990 to 84.9% (241.4 ha) in 2025, while classification accuracy ranged from 63.6% to 80.2% (Kappa = 0.68–0.81). The campus results were considered alongside published studies of surrounding Hathazari and Savar, where settlement and built-up areas have expanded and vegetation or forest cover has declined. In Hathazari, previous research reported a 146% increase in settlement area and a 43% decline in hill forest between 1977 and 2017. Studies of the wider Savar area reported a decline in vegetation cover from 42.81% to 32.15% and an increase in built-up area from 39.05% to 60.49% between 2011 and 2021, while another study reported a 9.19% increase in urban area between 2011 and 2022. The vegetation pattern observed at CU also coincides with documented tree-plantation initiatives on the campus beginning in 1982, although the effect of these initiatives cannot be established from land-cover data alone. Overall, these results show that vegetation cover within both campuses increased over the study period despite substantial development in the surrounding areas. The differences in the timing and extent of land-cover change between CU and JU provide a basis for examining how institutional land management and surrounding urban growth interact over time.