Identification of eco-hydrological drivers controlling carbon exchanges of evergreen needle-leaf and deciduous broad-leaf ecosystems of Himalaya using machine learning classifiers

In the Indian Himalayan region, limited availability of eddy covariance data constrains our understanding of the environmental drivers regulating carbon sequestration in forest ecosystems. The present study addresses this gap by applying data-driven modelling approaches to predict daytime carbon exchanges across two contrasting Himalayan forest types: an evergreen needle-leaf forest in Uttarakhand (KKM) and a deciduous broad-leaf forest in Assam (KNP), India. Using high-frequency (hourly) eddy covariance measurements and associated micrometeorological parameters, i.e. rainfall, net radiation, air temperature, soil moisture, soil temperature, and relative humidity for the period of March 2020 to Dec, 2022 and Jan, 2016 to Dec, 2018, four machine learning classifiers (Naïve Bayes, K-nearest neighbors, support vector machine, and decision tree) were evaluated. A total of 21 experiments were performed individual or in combination of parameters. Model robustness was ensured through 100-fold bootstrapping, and performance was assessed using receiver operating characteristic (ROC) curves and area under the curve (AUC) metrics. Results showed that the decision tree classifier consistently outperformed other models in predicting daytime carbon fluxes at both sites, while Naïve Bayes performed particularly well when rainfall was the predictor. At KKM, rainfall, soil moisture, soil temperature, and relative humidity emerged as key drivers, highlighting the critical role of moisture availability. At KNP, rainfall, soil moisture, air temperature, and net radiation were dominant, especially during pre-monsoon and monsoon periods. These findings have important implications for nature-based solutions (NbS) affecting carbon sequestration; hence, the study supports targeted NbS strategies. Moisture conservation and soil health are vital in needle-leaf forests, whereas maximizing early-season moisture and maintaining thermal regimes enhance carbon uptake in broad-leaf systems.