Key points are not available for this paper at this time.
Tourism plays a vital role in driving economic and social development, but poses growing threats to fragile ecosystems, necessitating sustainable ecotourism planning. This study introduces an integrated framework combining deep learning, machine learning, and multi-criteria decision analysis to assess ecotourism potential in India’s Sundarban Biosphere Reserve. Sixteen influencing factors were used to develop potential maps using the Analytical Hierarchy Process (AHP), eXtreme Gradient Boosting (XGBoost), and Deep Learning Neural Network (DLNN). The analysis incorporated multi-source spatial datasets, including Sentinel-2 (10 m), SRTM (30 m), IMD rainfall data (0.25°×0.25°), OpenStreetMap, and Survey of India topographic maps at a 10 m resolution. AHP identified 43.85% of the area as high or very high potential, while Model performance was evaluated using accuracy, Kappa index, MAE, RMSE, and ROC-AUC. The DLNN model outperformed others with an AUC of 0.91, followed by XGBoost (0.86) and AHP (0.79). XGBoost and DLNN classified only 4.63% and 6.22%, respectively, reflecting their more conservative, data-driven outputs. SHapley Additive exPlanations (SHAP) analysis revealed that proximity to tourist spots, mangrove forests, and rivers strongly influenced ecotourism suitability. Ground-truth validation confirmed the robustness of DLNN predictions. This study demonstrates the advantages of DL and ML techniques in capturing complex spatial patterns and overcoming the limitations of traditional expert-based models. The proposed framework offers an advanced tool for ecotourism planning and resource management, balancing development and conservation. It contributes methodologically and practically to the growing body of spatial decision-support systems in sustainable tourism.
Baidya et al. (Wed,) studied this question.