Abstract
<jats:p>This study quantifies land cover change in the City of Zadar between 2017 and 2025 using PlanetScope imagery and Machine Learning classification. Spectral bands were complemented with spectral indices (NDVI, VARI, Brightness, BSI) and a NIR based texture measure to improve class separability. Random Forest and XGBoost algorithms were evaluated using stratified five fold cross validation, while a transfer experiment was conducted to assess temporal robustness. Epoch specific models achieved high classification accuracy (ACC ≈ 0.94–0.95; ROC AUC ≈ 0.96–0.98), whereas direct model transfer from 2017 to 2025 resulted in reduced performance (ACC = 0.75), confirming the presence of temporal domain shift. At the city scale, impervious surfaces expanded proportionally from 27% to 29%, accompanied by a slight decline and fragmentation of urban green spaces. Hexagon based spatial analysis identified spatially coherent zones of impervious surface growth, indicating structurally organized urban expansion. The results suggest continued conversion of green areas. Future research should explore multi-annual or sub annual satellite time series and advanced domain adaptation approaches to improve model generalization across temporal domains.</jats:p>