A Comparative Study of High-level Classification Algorithms for Land Use and Land Cover Classification and Periodic Change Analysis Over Transboundary Ruvu River Basin, Tanzania

Understanding the declination rate for different land use and land cover (LULC) classes in the basin is crucial for effective land management and long-term resource sustainability. This study focused on the transboundary Ruvu River Basin (RRB), which spans Tanzania and Kenya, where ground-based data is limited due to restricted data-sharing policies between the nations. The machine learning algorithms of Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbours (KNN), and classical Maximum Likelihood (ML) technique were compared in their classification performance for the years 1987, 2001, 2012, and 2023. The classification results showed marginal variations in assessment metrics. For this case, all algorithms achieved overall accuracy ranging from 92.56 to 98.91% and kappa values from 87.17 to 98.59% during the review period. In general, RF outperformed others. Specifically, it achieved the highest overall accuracy ranging from 98.59 to 98.91%, and kappa from 97.70 to 98.38%. These high scores could be attributed to its ability to handle large datasets effectively. Between 1987 and 2023, forest and shrubland decreased by 12.83% and 21.57%, while agriculture increased by 45.94% with low to moderate annual rates. At the same time, bareland and settlements showed a substantial increase of 126.36% and 898.06%, respectively. Based on conversion matrix results, 99.25%, 37.88%, and 31.59% of the settlement, agricultural, and shrubland areas were converted to other LULC types. It can be concluded that human interventions dynamically modified the natural landscapes in the basin. These findings may be useful for managing the basin and its natural resources.