The study area as shown in Figure 1 focused on Lubigi Wetland, located in northwestern Kampala, Uganda. The wetland formed part of the greater Lake Victoria Basin and provided ecological services such as flood regulation, water purification, and habitat for biodiversity [9] [11] . However, the wetland had been increasingly encroached upon by settlements, agriculture, and industrial activities [13] [14] . The methodology employed is shown in Figure 2 .

Satellite imagery from Landsat (1999, 2010) and Sentinel-2 (2022) were used to analyze land cover dynamics. Landsat images with a spatial resolution of 30 m were resampled to 10 m to match Sentinel-2 resolution. Ancillary data included population statistics from Uganda Bureau of Statistics and secondary literature on wetland dynamics [9] [26] .

Preprocessing included geometric and radiometric corrections. Land cover maps were generated using the Maximum Likelihood Classification algorithm, which had been widely applied in wetland and land use studies [3] [6] [16] . Five land cover classes were identified: wetlands, built-up areas, cropland, forest, and water bodies.

Accuracy of classification was evaluated using ground control points and high-resolution imagery. A confusion matrix and kappa coefficient were generated to determine classification reliability [27] .

The LULC projection process for 2030 used a transition sub-model structure. The analysis was conducted using logistic regression models and a Multi-Layer Perceptron (MLP) neural network in the Land Change Modeler [23] [24] . Driver variables incorporated into the MLP included road networks, settlement distribution, and elevation, which were selected to represent accessibility, demographic pressure, and physical constraints on land cover change. This structured approach enabled the identification of land cover transitions and persistence between 1999 and 2010, thereby supporting 2030 projections. The main findings show that at least 1905 cells changed land cover classes during this time, but a significant number of cells (25,468) remained in the same land cover class. The sample size for the analysis was 1,905, with a 50% split between training and testing data.

Ecosystem services were assessed by linking land cover categories with provisioning, regulating, supporting, and cultural functions [4] [5] [18] . Changes in land cover were analyzed to infer impacts on wetland services such as flood regulation, biodiversity, and carbon storage.

The classified land cover maps for 1999, 2010, and 2022 illustrated significant transformations within the Lubigi Wetland ( Figure 3 and Figure 4 ). Between 1999 and 2022, built-up areas expanded from 10.74 km² to 122.5 km², representing a 1040% increase. Conversely, wetlands declined from 453.54 km² to 228.96 km², a loss of nearly 50%. Cropland expanded significantly, while forest and water bodies remained relatively stable.

The land cover change statistics are summarized in Table 1 , which shows the change in each category. Wetlands were the most affected, highlighting the scale of anthropogenic pressure. These results were consistent with earlier studies that identified urban expansion and agriculture as major drivers of wetland degradation in Uganda [11] [14] [15] .

The Land Use and Land Cover (LULC) classes were merged into three categories, namely: Anthropogenic (Built-up Area and Cropland), Forestland, and Wetland (Waterbody and Wetland). This reclassification was undertaken to reduce model complexity and improve predictive performance by minimizing confusion among spectrally similar classes, while still preserving the ecological and socio-economic relevance of key land cover types. Similar aggregation of LULC classes has been applied in previous modelling studies to enhance robustness and interpretability [23] . Grouping Built-up and Cropland into “Anthropogenic” emphasized human-driven land transformations, whereas Forestland and Wetlands were retained as distinct categories due to their ecological importance and vulnerability to degradation. Using this simplified classification, the Multi-Layer Perceptron (MLP) neural network and Markov Chain model projected continued wetland decline and urban expansion by 2030 ( Figure 5 ). Built-up areas were expected to reach 279 km², while wetlands would shrink further to 200 km². The validation demonstrated strong overlap between observed and predicted land cover, with the model achieving an accuracy of 87.64%. This performance aligned with reported applications of similar predictive approaches in land use modelling [6] [24] .

These projections indicated that without intervention, wetland loss in Kampala would accelerate, intensifying ecological vulnerability.

Regression analysis results ( Table 2 ) revealed a strong positive correlation (r = 0.985, p < 0.05) between population growth and the expansion of built-up areas, based on four paired data points derived from population figures and land cover extent (km²) for the analyzed years. This indicates demographic pressure as a key driver of wetland degradation [9] [26] . However, the small sample size (n = 4) represents a limitation, and the results should therefore be interpreted with caution. Infrastructure development, road construction, and informal settlements along the Kampala-Mukono corridor further contributed to the observed wetland losses [13] [14] .

Ecosystem services were assessed by linking land cover categories with ecosystem functions using a qualitative matrix approach developed from the literature [4] . The decline in wetland cover had already undermined critical ecosystem services. Flood regulation capacity was reduced, increasing the frequency and severity of urban flooding ( Figure 6 ) [11] [12] . Water purification and groundwater recharge functions were diminished, contributing to declining water quality in Kampala [4] [18] . The loss of wetland vegetation also reduced carbon sequestration potential, undermining climate change mitigation [1] [2] [12] . Biodiversity support functions were threatened, with field-based studies reporting declines in bird and aquatic species populations [15] . In addition, the degradation of Lubigi Wetland diminished cultural ecosystem services. Recreational opportunities and access to natural green spaces have declined, cultural heritage and spiritual values associated with wetlands are eroding, and the role of wetlands as outdoor learning environments has been weakened, undermining both social well-being and community engagement in conservation [4] .

The observed trends in Lubigi were consistent with global wetland loss patterns, where urbanization and agricultural expansion remained the primary drivers [2] [5] [6] . Comparable studies in Greater Cairo [6] and papyrus swamps in East Africa [5] reported similar declines in wetland ecosystem services.

These results reinforced the importance of predictive modelling in anticipating environmental change and informing sustainable urban planning [3] [16] [17] .