Ecological impact assessment and optimizing artificial neural network for land cover classification under the billion tree tsunami project

Accelerating climate change and global warming are rapidly degrading natural ecosystems; in response, strategic afforestation serves as a nature-based, scalable approach to rebuild habitats, support biodiversity, and strengthen ecological resilience. Afforestation efforts, such as large-scale tree-planting initiative, is a key nature-based response in mitigating the effects of these challenges and Pakistan's Billion Tree Tsunami Project (BTTP) represents one of the most significant initiatives aimed at combating deforestation and promoting ecological restoration. However, accurately classifying forested areas remains a critical challenge for monitoring afforestation projects. Conventional classification approaches often lack the precision required for reliable land use and land cover (LULC) analysis, while previous ANN-based studies have suffered from inconsistent parameterization and a narrow focus on classification outcomes, overlooking the broader ecological perspective. This study aims to assess the effectiveness of the BTTP by employing Artificial Neural Network (ANN) classification techniques to analyze LULC changes before and after the project's implementation in Southern Khyber Pakhtunkhwa (KPK). By optimizing ANN classification approach the performance was systematically refined by adjusting various settings, with the study highlighting that a well-balanced configuration led to the best results, reaching higher classification accuracy while ensuring optimal computational time. Through careful tuning, the findings revealed that certain combinations of values consistently enhanced accuracy, emphasizing the importance of strategic parameter optimization. Land cover changes between 2013 (pre-afforestation) and 2024 (post-afforestation) were evaluated using Landsat imagery. The optimized ANN achieved an overall classification accuracy of 98.89 % and a Kappa Coefficient of 0.9775. Results showed a prominent increase in forest cover from 12.54 % in 2013 to 14.43 % in 2024, while barren land decreased from 80.32 % to 77.63 %. Unlike earlier works (e.g., Garhi Chadran), this research extends beyond classification by integrating ecological indicators, including NDVI, EVI and land surface temperature (LST). The analysis reveals an inverse relationship between vegetation cover and LST, validating the cooling effect of afforestation and its contribution to local climate regulation. By merging advanced ANN-based classification with ecological insights, this study demonstrates the positive ecological and climatic impacts of BTTP and provides policymakers with robust, evidence-based tools to guide future afforestation and climate adaptation initiatives. The findings underscore the BTTP's significant role in mitigating climate change, enhancing biodiversity and promoting sustainable land management, while also offering valuable insights for policymakers and environmental planners to improve and expand similar afforestation initiatives. • Reproducible, optimized ANN: Systematic hyperparameter grid search yielded a parsimonious feed-forward ANN (single hidden layer; tuned learning rate, momentum, RMS/stop criteria and iteration scheduling) with cross-validation and convergence/runtime diagnostics to ensure stable generalization. • Multi-indicator, field-anchored validation: ANN-derived LULC maps were validated with KP Forest Department field inventories and satellite indicators (NDVI, EVI, LST), confirming structural vegetation establishment and localized surface cooling. • Benchmarking and transferability: The optimized configuration consistently generalized across multiple test sites, delivering stable performance and predictable runtime–performance tradeoffs versus baseline methods. • Operational, policy-ready pipeline: Fully documented preprocessing, parameter grids and validation protocols form a transferable workflow for scalable afforestation monitoring (surviving canopy area, canopy closure, microclimate diagnostics).