Department of Physical Sciences and Information Technology, Agricultural Engineering College and Research Institute, Tamil Nadu Agricultural University (TNAU), Coimbatore 641 003, Tamil Nadu, India
Auto-generated profile, based on publicly available information
M Sankavi
Department of Physical Sciences and Information Technology, Agricultural Engineering College and Research Institute, Tamil Nadu Agricultural University (TNAU), Coimbatore 641 003, Tamil Nadu, India
Forest fires pose a significant threat to ecosystems, biodiversity and human livelihoods, necessitating the development of advanced predictive models for timely intervention. Machine learning (ML) techniques have emerged as effective tools for predicting forest fires by utilizing various environmental, meteorological and topographical datasets. This review explores recent advancements in ML-based fire prediction, highlighting key methodologies such as Random Forest, Support Vector Machines, Neural Networks and hybrid models. The combination of remote sensing data, real-time observation and cloud computing has notably improved the prediction accuracy. Moreover, the growth of Explainable AI (XAI) has enhanced the interpretability of ML models. Despite advancements, challenges such as data imbalances, model generalization issues and computational constraints persist. Future research must focus on refining ML algorithms to improve regional adaptability, integrating climate change forecasts and establishing real-time early warning systems. By bridging the gap between theoretical advancements and practical applications, ML-driven forest fire prediction models can significantly contribute to mitigating the devastating impact of forest fire and enhancing global fire management strategies.
Rubber plantations (Hevea brasiliensis (Willd. ex A.Juss.) Müll.Arg.) are expanding tropical carbon (C) stores, storing an estimated 160–300 mg C ha-1 of total ecosystem C at maturity, but precise biomass and soil organic carbon (SOC) assessment remains difficult due to age disparities, historical land-use variations and data gaps. This review systematically analyses trends and research gaps in biomass accumulation of rubber systems. PRISMA 2020 guidelines were used to examine 104 studies in Scopus published between 1999 and 2025. The literature revealed a marked increase in research studies in China and Thailand post-2013 using a bibliometric approach presented by Biblioshiny. The most important methods are species-specific allometry (coefficient of determination (R2) > 0.9), age-based growth curves, remote sensing (RS) regressions (R2 0.2–0.7) and machine learning (ML) models like Random Forest/XGBoost (R2 0.8–0.97, root mean square error (RMSE) 7–40 mg/ha), which outperform simpler parametric models. Deep learning (DL) yields the highest accuracies (RMSE ~6 mg ha-1). Gaps persist in standardised protocols, South Indian datasets (e.g., Tamil Nadu/Western Ghats) and integrated lifecycle carbon accounting, including latex/SOC. Forest-to-rubber conversion incurs net losses, while cropland-to-rubber/agroforestry offers gains. The authors recommend adopting harmonised Tier 2/3 IPCC models, multi-decadal RS archives for reducing emissions from deforestation and forest degradation (REDD+) and ML extension to Indian smallholders. Policy integration with offset markets requires avoiding deforestation to realise climate-smart benefits.