The world continues to witness increasing urban growth, which is exerting pressure on and altering the physical biodiversity, environment, and ecosystem (Abebe et al., 2022; Abutaleb et al., 2015; Yonghong Hu and Jia 2010) Global urban growth is driven by LULC change. In China’s Middle Reaches of the Yangtze River Urban Agglomeration (Hubei, Hunan, Jiangxi), expansion has reduced ecosystem service values, highlighting urbanization’s environmental impacts (Yangcheng Hu, Liu, and Li 2022). Clearly, temperature increases are becoming more extreme, with heatwaves reported more often, for longer durations, and at higher intensities (Tzyrkalli et al. 2024; Ntoumos et al. 2020). These changes affect climate by altering LST and rainfall, driving heat stress and related health problems (Alademomi et al., 2022; Alawamy et al., 2020; Arunab & Mathew, 2024). In developing areas, replacing vegetation covers and natural surfaces (Elagib 2011), the LULC significantly affects energy transformation (Rehman et al. 2022), and the spatiotemporal variation of Urban Heat Islands (UHI) in big cities (Shirani-Bidabadi et al. 2019). In most cases, the lack of proper planning of urban settlements leads to air and water pollution, weather changes, diseases, LULC changes, etc., both at the local and global levels (Ayua et al., 2023; Arunab & Mathew, 2024). These challenges are well recognized by scholars in the field, contributing to a growing interest in LULC research in recent years (Hassan et al. 2021). LST also impacts climate changes in the ecosystem and accelerates global warming (Rakib et al. 2020). Notwithstanding the changes in Land surface temperature, continuous observation in monitoring the relationship to land cover changes has become essential for appropriate management strategies and policy determination (Obiefuna et al. 2018). Calculating LST and its relationship with LULC is essential in addressing several climate change issues in urban areas (Farhan et al. 2024).Variations in seasonal precipitation explain some of the differences in urban–rural cooling (Chow and Roth 2006 ; Murphy et al. 2011). Studies show that LST–LULC variation alters precipitation and urban biodiversity, particularly in dense cities. Short-term variability is marked by seasonal day–night cycles with coherent spatial signals, and tropical nights exhibit strong temporal and spatial trends (Canton and Dipankar 2024). Knowing this relationship enables a better understanding of the magnitude and pattern of Urban Heat Island (UHI) (Marzban, Sodoudi, and Preusker 2018). The alterations of LST and LULC and their impact were examined by Rahman et al.,(2017) and Maimaitiyiming et al., (2014) observed the rise in LST from replacing vegetation with built-up areas drives UHI formation, largely due to human activity. Vegetation loss and waste heat emissions further intensify heat accumulation (Maimaitiyiming et al. 2014).

Evidence suggests that both quantitative and qualitative approaches are widely used to analyze changes in LULC patterns (Naikoo et al. 2020). A quantitative assessment of LULC change dynamics is the most effective way to manage and understand the landscape transformation (Naikoo et al. 2020). Weather stations are the traditional tool for UHI studies but face limits from installation and site constraints, restricting spatial coverage and the range of UHI types analyzed (Se Woong Kim and Brown 2021). Remote sensing has been the most common method used, especially in Africa, due to its efficient and faster approach for monitoring spatiotemporal data variation of UHI over a long period (Ayua et al., 2023; Govil et al., 2020). Recent studies show that Landsat images are widely used for estimating LST and LULC, and it is now a well-established method (Galodha and Gupta 2021; S Guha, Govil, and Diwan 2020; Subhanil Guha and Govil 2021). Landsat imagery provides a high-resolution method for estimating LST that is relevant for both small- and large-scale research (Sajan et al. 2023). Studies of UHI using remote sensing are very few in Africa, with most studies highly concentrated in Asia, Europe, and North and South America (Yadav and Singh 2024). Remote sensing has been identified as one of the most critical techniques to estimate LST since the 1970s (Cetin et al. 2024; Devendran and Banon 2022), and for monitoring the spatiotemporal data variation of the UHI effect over a long period (Ayanlade 2016). Using remote sensing techniques to identify growth patterns in urban areas and land cover changes helps determine LST variations and evaluate LST on a global scale (Malik, Shukla, and Mishra 2019). To overcome the limitations of previous models for LULC and LST relations classification, an artificial intelligence approach is required to examine the relationships between LST and LULC (M. Kim, Kim, and Kim 2022). Ensemble ML models like Random Forest and XGBoost effectively estimate LST using remote sensing indices (e.g., NDVI, NDBI), offering higher accuracy and interpretability than traditional statistical methods (H. Li et al. 2021; Suthar et al. 2024; Yin et al. 2021). While ML models have been applied for LST and UHI estimation in Africa, their use remains limited in Sub-Saharan regions due to data constraints, scarce high-resolution monitoring, and limited adoption of explainable ML approaches (Faye et al., 2022). The use of machine learning algorithms can enhance interpretation so that humans can understand and make predictions (S W Kim and Brown 2021). A study in Porto Alegre, Rio Grande do Sul, Brazil, showed the spatial and temporal evolution of the relation between vegetation and thermal dynamics through linear regression (Kaiser et al. 2022). The issue of UHI has received considerable critical attention in recent years due to specific numerical indicators, particularly an increase in air temperature in specific areas within cities (Kafy et al. 2020; Galodha and Gupta 2021) in which LST differences occur between urban and rural regions (Subhanil Guha and Govil 2021; Hassan et al. 2021). The cause of UHIs has been associated with factors, such as having densely populated buildings in an urban area and different types of building materials that absorb heat transmitted to the environment through radiation of high waves (S W Kim and Brown 2021; He et al. 2021). Apart from health risks, the joint economic costs of urban impacts from the UHI effect and climate change have been estimated to be 2.6 times those without the UHI effect (Yunfei Li et al. 2020). Similarly, other studies also found that landscape metrics and land cover composition had a substantial impact on the UHI effect (Bagyaraj et al. 2023). UHI intensity typically peaks at night, as urban surfaces release the heat accumulated during the day, resulting in reduced nocturnal cooling compared to vegetated areas (Yunfei Li et al. 2020). Urban Heat Islands (UHI) are classified into Atmospheric Heat Islands (AHI) in the atmosphere and Surface Heat Islands (SHI) at ground level. LST is commonly used to study UHI, as it is vital for assessing surface energy balance (Galodha and Gupta 2021; He et al. 2021). Mitigating UHI requires optimizing urban form through planning, green infrastructure, cool roofs, reflective pavements, and supportive policies (Yadav and Singh 2024). Also, (Yadav and Singh 2024) suggest renewable energy and low-carbon practices aid UHI mitigation and urban sustainability, yet urban morphology’s role remains underexplored in developing cities (He et al. 2021). The UHI effect has been documented since 1833, first noted by meteorologist Luke Howard (Stewart 2011). Studies have also established that cities vary, and their surface behavior patterns vary concerning heat release, absorption, evaporation, and radiation (Hassan et al. 2021). New works are being conducted to introduce some thermal comfort indices for measuring the UHI effect and intensity (S Guha et al. 2018). Despite progress made empirically and theoretically, studying the concept of increasing urban temperature and its impact on urban ecosystems over a century, it still lacks understanding (Marzban, Sodoudi, and Preusker 2018). A significant segment of the existing literature examined larger West African countries (Oluwafemi E Adeyeri et al. 2024; Athukorala and Murayama 2020; O E Adeyeri, Akinsanola, and Ishola 2017; Awuh et al. 2019; Ishola, Okogbue, and Adeyeri 2016; Herrmann and Brandt 2013), The Gambia (GBA) is highly vulnerable to climate variability due to its small size, narrow geography, and reliance on climate-sensitive sectors. Rapid urbanization in the Greater Banjul Area has driven deforestation, wetland loss, and coastal reclamation, amplifying UHI effects and risks from flooding, heat stress, and sea-level rise. Limited resources and weak governance further constrain adaptation, underscoring the need to link LULC and LST in vulnerability assessments. GBA provides key insights into how small West African states face climate and urban pressures, yet UHI predictive modeling remains largely unexplored in the region. Reviews such as Najah et al.(2025) on Despite advances in ML for UHI studies, applications are concentrated in Asia, Europe, and North America. West African cities face data and methodological gaps, with little ML use for linking LST prediction to heat risk. In The Gambia, no study has analyzed LULC–LST dynamics or projected thermal risks using interpretable ML for policy and adaptation. Therefore, this study offers the following objectives:

This study advances from environmental monitoring to actionable urban resilience planning, addressing a key knowledge gap in the Global South where data and institutional limits hinder adaptation. By linking urbanization, governance, and environmental vulnerability, it delivers both a replicable framework and policy-oriented guidance for climate-resilient development in vulnerable coastal African cities.

This study applies an integrated framework of remote sensing and spatial statistics to examine LULC impacts on LST and UHI in the Greater Banjul Area (GBA). Multi-temporal Landsat and Sentinel imagery captured spatial–temporal LULC changes and LST variations. Supervised ML algorithms classified LULC, while LST from thermal bands delineated UHI hotspots. Temporal analysis revealed UHI evolution with urban expansion and ecological change. Accuracy was assessed using RMSE, bias, and sample size. Random Forest, XGBoost, and SHAP-based feature importance supported prediction, while future thermal scenarios and a Heat Stress Index map highlighted population exposure risk. Figure 2 overall depicts the methodology adopted for the study.

This methodology outlines data acquisition, training, and analysis linking LULC, LST, NDVI, and NDBI to UHI. Landsat 5 (1990, 2006, 2010; 30 m) and Sentinel-2 (2020; 10 m) imagery classified seven LULC categories: water bodies, trees, flooded vegetation, crops, built-up areas, bare ground, and rangeland. Bands B1–B7 (Landsat) and B2, B3, B4, B8 (Sentinel) were analyzed. Preprocessing in ArcGIS 10.8.2 included TOA correction and cloud masking. Training samples validated via Google Earth Pro supported supervised maximum likelihood classification. LULC classes were mapped in ArcMap using standardized symbology. The description of the different classes is elaborated in Table 2 below.

Figure 13 presents a comparative evaluation of Random Forest (RF) and XGBoost models for Land Surface Temperature (LST) prediction. The scatter plots (left panel) demonstrate that 90% (RF) and 83% (XGBoost) of the variance align with the 1:1 reference line, confirming strong predictive performance. However, XGBoost predictions (orange) exhibit tighter clustering along the ideal line, indicating reduced variance and higher precision compared to RF predictions (blue). The residual distribution plots (right panel) for both models are centered near zero, suggesting minimal systematic bias. XGBoost residuals show a sharper, narrower peak with most errors within ± 2°C, whereas RF residuals display broader dispersion. Quantitatively, XGBoost yielded an RMSE of 1.972°C, MAE of 1.475°C, and R² of 0.917. In comparison, RF produced an RMSE of 2.373°C, MAE of 1.769°C, and R² of 0.880. These metrics confirm that XGBoost offers superior accuracy and precision in LST prediction.

Between 1990–2020, GBA’s built-up areas expanded by 187.87 ha (36.2%) as cropland, water bodies, bare land, and rangeland declined; flooded vegetation rose slightly, and tree cover fell. Mean LST increased from 28.19 ℃ (1990) to 37.14 ℃ (2010), then dropped to 35.68 ℃ (2020), with ranges widening from 15–32.54 ℃ to 24–49.9 ℃. The findings contrast with (Tanoori, Soltani, and Modiri 2024), findings show the urban core remains cooler year-round than its surroundings. In GBA, LST stayed high from 1990–2020, with UHI rising from 6–23°C (1990) to 24–48°C (2010) before declining to 15–40°C (2020). Spatial patterns in LST and UHI (Fig. 6, 7) indicate vegetation loss and urban expansion, especially in the south, highlighting urbanization’s impact on land cover and temperature. XGBoost outperformed Random Forest in LST prediction with lower RMSE (1.97°C vs. 2.37°C), lower MAE (1.48°C vs. 1.77°C), and higher R² (0.92 vs. 0.88). In Table 6, Results show accurate, consistent predictions, with LULC as the dominant predictor (82–90%), NDVI moderately important (65–70%), and “Year” (< 5%) and NDBI (< 1%) minimal. Spatial and environmental variables thus outweigh temporal factors in determining LST.

From 2020–2040, most land uses show warming, with mean LST rising most in Parks (+ 1.92°C), Dumpsites (+ 1.83°C), and Vacant Lands (+ 1.76°C), linked to urban infill and vegetation loss. Industrial and transport zones show slight declines (− 0.15 to − 0.23°C), likely from reflective materials. HSI analysis indicates rising thermal stress: Residential (+ 12.7%), Commercial (+ 13.8%), and Transport (+ 15.4%), while Protected Areas remain stable (0.42–0.44), reflecting vegetation’s cooling role. Spatial-temporal results show a strong LST–LULC relationship, with R² = 0.74.. This also confirmed the findings from (Obiefuna et al. 2018), A 1984–2015 assessment showed rapid urbanization driving LST increases, while wetlands and vegetated areas helped moderate temperatures, supporting these findings. A strong negative LST–NDVI correlation (40.15% in 2006; Fig. 8, 11) shows vegetation’s cooling role, while a rising LST–NDBI correlation (R² = 0.26 in 2020; Fig. 10) reflects urbanization-driven UHI, especially in Banjul, Kanifing, and Brikama.

Urban density–LST links are shaped by surface emissivity. LST–NDVI correlation was moderate in 1990 (R² = 0.17), peaked in 2006 (R² = 0.40), then weakened with urbanization. Built-up land rose from minimal in 1990 to 53% in 2020, as vegetation and cropland declined, tree cover fell (8%→2%), and LST increased from 15–32.54°C to 24–49.9°C. The findings are consistent with Njoku and Tenenbaum (2022) and Awuh et al.(2019) demonstrating that built-up areas are major contributors to LST variations in the GBA. Mukherjee and Singh (Mukherjee and Singh 2020) confirmed LST increased, resulting in the formation of UHI. A recent study by Valentin Loembet Makaya et al. (2025) investigated In Dakar (1986–2023), LULC changes strongly influenced LST and UHI. Rising built-up areas and declining bare soil intensified UHI, while vegetation and water bodies lowered surface temperatures, confirming their cooling role. A study by Ayanlade et al. (2017) In Lagos (2002–2013), rapid urban expansion and vegetation loss intensified the UHI effect. Seasonal and diurnal LST analysis using MODIS data (Singh et al., 2017), Unlike ground data, remote sensing captures spatial variability, showing built-up areas as heat sources and vegetation as cooling sinks. LULC changes have altered urban thermal conditions, raising surface temperatures and impacting microclimate and comfort. In 2010, built-up growth strongly correlated with rising LST (R² = 0.74) from Das and Angadi (Das and Angadi 2020) showing similar results in a correlation between LST and NDBI.

Moreover, the findings agree with previous studies, including Njoku and Tenenbaum (2022), Govil et al. (2020), Subhanil Guha and Govil (2021), Das and Angadi (2020), and Kafy et al., (2020). Vegetation and water bodies showed low temperatures, while built-up and sparse areas were hotter. GBA’s rapid urban expansion mirrors Lagos, where population grew from 9.3M (2006) to 21M (2018, 7%). GBA grew 5.1% (2013–2020), with immigration and housing demand, especially in the south, driving built-up growth. Population rise is thus a key factor in urban expansion [1,56]. Following the present findings by Kimuku and Ngigi (2017) reported that between 1989 and 2015 in Nakuru County, Kenya, established that an increase in settlement in urban areas correlated with increased LST of the region. Rising built-up areas increase LST and intensify UHI, as impervious surfaces absorb heat by day and release it at night, making urban zones warmer than rural surroundings (Farid et al. 2022). Similarly (X. Li et al. 2021) This study assessed LULC change and SUHI in Kampala (1995–2017). Built-up land doubled from 12,133 ha (1995) to 25,389 ha (2016). High-temperature areas expanded from 22,910 ha (2003) to 27,900 ha, while average daytime SUHI intensity fell slightly from 2.2°C (2003) to 1.9°C (2017), though spatial variability increased.

The climate characteristics between the urban and rural areas as a primary indicator of UHI, exacerbating rapid urbanization, which is attributed to an increase in Land surface temperature (Yadav and Singh 2024). In GBA, stack profile analysis (Fig. 6 and Fig. 7) shows Dense urbanization in central and northern areas drives high LST and strong UHI, while less developed or vegetated regions show lower heat concentration, as revealed in a study by Cetin et al., (2024), whose findings show vegetated areas are directly linked with lower surface temperatures. Low UHI appears in the Atlantic west, while wetland-dominated northeast zones show persistently high UHI. UHI rose from low (1990) to high (2020), reflecting urbanization’s thermal impact. Rising temperatures heighten drought, rainfall, and food insecurity risks. In Brikama, poor services, weak land management, outdated zoning, and unregulated land sales drive uncontrolled growth, ecological loss, and climate vulnerability. Our findings are consistent with Arunab & Mathew (2024) whose study used spatially enhanced models, where XGBoost outperformed RF in LST prediction (R² = 0.917, RMSE = 1.97°C vs. R² = 0.880, RMSE = 2.37°C), consistent with earlier findings (R² = 0.871, RMSE = 0.48°C). Performance was driven by spatial predictors, with LULC (82–90%) and NDVI dominating, while temporal features contributed minimally, underscoring land cover and vegetation as key LST controls.

Findings show weak planning coordination, outdated laws, and unclear mandates, with unregulated land sales and absent housing policy driving ecological loss and climate risk. Brikama holds land but lacks services, Banjul and Kanifing are saturated (Touray 2024). The Tourism Development Area (TDA) is experiencing growing urbanization. GBA’s low-density form offers potential for sustainable planning. Rapid urbanization in Banjul and Kanifing intensifies UHI, raising climate risks like drought, extreme rainfall, food insecurity, and heat-related health issues (Son et al., 2020, Suthar et al., 2024). Addressing these issues requires comprehensive reforms: updated zoning for mixed-use growth, a national housing policy for climate-resilient infrastructure, and a UHI strategy with cool roofs and green spaces. Policies must embed sustainable land management and restrict high-risk coastal development. Stronger institutions, better local coordination, revised land laws, and tools like remote sensing and AI can improve monitoring and guide planning (Tanoori et al., 2024, Touray, 2024). Urban ecology measures—green corridors, trees, parks, and buffers—can reduce temperatures and improve comfort. Public awareness, stronger land laws, and stricter regulation will enhance management. Sustainable planning, adaptive policies, and vegetation expansion can help GBA balance growth with ecology, building resilience and equity.

Researchers in underdeveloped countries face challenges from rapid population growth, limited infrastructure, weak regulations, and difficulty obtaining timely, accurate data (Devendran and Banon 2022). The main limitation is inconsistent decadal data, as no high-quality imagery exists for 2000–2005, limiting temporal comparison. Further gaps, short datasets, and seasonal NDVI variability reduce LST correlation reliability (S Guha, Govil, and Diwan 2020). Long-term UHI effects in GBA need deeper study. In African cities, UHIs pose health, productivity, and income risks for street vendors lacking shade and water, increase evaporative demand and water stress, and worsen thermal discomfort in poorly ventilated spaces. They also affect rental markets, with cooling amenities shaping pricing and occupancy. Addressing these challenges requires integrating spatial, social, and economic analysis.

Rapid urban growth and landscape loss in GBA have raised LST and intensified UHI. Climate-sensitive planning, zoning for reflective roofs and pavements, and wetland protection are vital. Green roofs and urban forestry in hotspots can cut localized heat and enhance resilience.

This study analyzed urbanization, Land Use Land Cover (LULC) changes, and the Urban Heat Island (UHI) effect in the Greater Banjul Area (GBA) from 1990–2020. Built-up areas expanded by 36% (187.9 ha), driven by population growth and unregulated development, while mean Land Surface Temperature (LST) rose from 28.19°C to 35.68°C. Strong UHI effects were observed in Banjul and Kanifing. LST distributions showed an evolving UHI pattern: 23°C in 1990, 13–34°C in 2006, peaking at 24–48°C in 2010, then slightly decreasing to 15–40°C in 2020. The weakening NDVI–LST relationship (R² = 0.07 by 2020) indicates declining vegetation cooling and intensifying urban heat. Machine learning models complemented geospatial analysis. XGBoost outperformed Random Forest (R² = 0.917 vs. 0.880; RMSE = 1.97°C vs. 2.37°C), confirming the dominant role of LULC and NDVI. This aligns with studies elsewhere, supporting ML integration with spatial data for environmental monitoring.

As the first comprehensive UHI study in GBA, it establishes a baseline for assessing urbanization’s impacts in The Gambia. By combining ML with the Heat Stress Index (HSI), it advances UHI analysis beyond temperature-based approaches, linking environmental dynamics with human comfort. The framework informs climate-smart planning, nature-based solutions, and health-sensitive zoning, addressing gaps in African UHI research. Future work should extend time-series and cross-city analysis, integrate socio-economic vulnerabilities, and develop decision-support tools. The study underscores how weak governance, outdated policies, and ecological loss intensify risks, and advocates adaptive planning with green infrastructure, equitable land distribution, and climate-responsive design to improve resilience and sustainability in GBA and similar regions.

CRediT authorship contribution statement