Harnessing GIS, remote sensing, and machine learning for sustainable management and carbon sequestration of non-timber forest products in Gujarat, India

This study focuses on the sustainable management of non-timber forest products (NTFPs) in the Narmada, Dang, and Panchmahal districts of Gujarat, India, emphasizing carbon sequestration and carbon credits. NTFPs such as medicinal plants, fruits, nuts, and resins play a crucial role in the local economy and biodiversity conservation. Accurate mapping and assessment of these resources are essential for implementing sustainable management and conservation strategies. Advanced spatial analysis techniques, including geographic information systems (GIS), remote sensing, and logistic regression models, were employed to analyze the spatial distribution of NTFPs. High-resolution Sentinel-2 satellite imagery and field survey data were integrated to create detailed spatial maps while, logistic regression models evaluated environmental factors like soil type, elevation, and climatic conditions affecting NTFPs distribution. The study identified that environmental variables such as litter cover, elevation, and NTFPs type are critical in determining the distribution of NTFPs, with NTFPs type accounting for 68% of the variance in distribution patterns. The logistic regression model and GIS expert system predicted NTFPs distribution with 65.43% and 70.37% accuracy, respectively. The GIS expert system demonstrated a higher specificity rate (47.05%) compared to logistic regression (35.29%), indicating its superior ability to predict NTFP absence. Both models were validated using a sample size of 81, with error matrices generated for comparative analysis. Additionally, the research explored the carbon sequestration potential of NTFPs and their implications for carbon credits. The study recommend machine learning algorithms, particularly the random forest (RF) model, for aboveground carbon stock estimation across different NTFPs regions. The RF model showed superior performance with an R² value exceeding 0.6 across the study areas, compared to multivariate stepwise regression, which had R² values below 0.4. The RF model's accuracy was validated through a comparison with actual field data, achieving a root mean square error of 24.72 t hm⁻² in NTFPs regions. Carbon stocks were observed to range from 50 to 250 t hm⁻², depending on the region's ecological characteristics and topographical variations. The research findings highlight the importance of integrating local ecological knowledge with spatial data and advanced modelling techniques to enhance the sustainable management of NTFPs. The potential for carbon credits offers a financial incentive for conserving NTFPs, promoting sustainable harvesting, and ensuring long-term biodiversity conservation. This study provides a replicable framework for NTFPs management and carbon stock estimation, applicable to similar ecological settings worldwide. The integration of GIS, remote sensing, and machine learning methodologies presents a robust approach for policymakers and stakeholders aiming to optimize NTFPs conservation strategies while advocating carbon credit opportunities for economic development.