While machine learning (ML) models have shown considerable promise for modelling complex environmental phenomena such as floods, their adoption in tropical flood prediction is often constrained by limited interpretability. This study addresses this challenge by developing an integrated SHapley Additive exPlanations (SHAP) enhanced framework that increases the transparency and usability of flood susceptibility mapping based on multispectral satellite data. The proposed model combines key remote-sensing-based spectral indices, including the Curve Number, Normalized Difference Water Index (NDWI), Automated Water Extraction Index (AWEI), and Normalized Difference Vegetation Index (NDVI), with topographical variables such as slope and elevation, to generate predictions that are both accurate and interpretable. A supervised learning approach using the Light Gradient Boosting Machine (LightGBM) algorithm was implemented due to its high predictive quality and computational efficiency. Model training and validation employed stratified sampling on a dataset comprising Landsat 7 surface reflectance imagery, SRTM elevation data, and historical flood records from the Global Flood Database. Feature selection was performed in two stages: first, reducing redundancy via correlation analysis, and second, applying SHAP-based attribution to quantify the contribution of each variable. The final model retained four primary predictors such as AWEI, Near-Infrared (NIR) reflectance, slope, and elevation, with SHAP analysis providing both global and local explanations that revealed strong relationships between surface water content, terrain variation, and flood occurrence. The tuned LightGBM model achieved an accuracy of 92% and an AUC of 0.950, while spatial analysis of the SHAP outputs identified high-risk flood zones consistent with historical patterns. These insights demonstrate the potential of SHAP not only for enhancing interpretability but also for supporting targeted flood risk mitigation strategies, offering actionable information for urban planners and disaster management agencies. This work advances the application of explainable artificial intelligence in environmental modelling, emphasizing the importance of pairing predictive performance with model transparency.