Ecological drivers of soil carbon in Kashmir Himalayan forests: Application of machine learning combined with structural equation modelling

Soil carbon (SC) heterogeneity in mountain ecosystems is ascertained by a complex interdependency of topography, climate, edaphic features, and biotic elements, which may incite uncertainties in regional SC estimation. However, quantitative evaluations of the interplay between SC and these determinants as well as underlying possible link networks, are uncommon. Using the data set of SC along with soil properties at 0–10 and 10–20 cm depths from 135 plots under three coniferous forests, we aimed to ascertain SC heterogeneity and to elucidate how these interactions affect the SC storage, operating data-driven models (Least Absolute Shrinkage and Selection Operator [LASSO] regression and structural equation modeling [SEM]) to identify the dominant explanatory factors affecting the distribution of SC in Kashmir Himalayan forests. Average SC stocks at 0–10 cm and 10–20 cm depth intervals range from 32.41 Mg ha−1 in sub-alpine (SA) forest to 48.50 Mg ha−1 in mixed conifer (MC) forest. The findings show that SC declines significantly from 0 – 10 cm to 10–20 cm strata, consistent with other soil physico-chemical determinants other than bulk density. SEM renders better model fit (0–10 cm: R2 = 0.61; 10–20cm: R2 = 0.46) with lesser uncertainties compared to LASSO (0–10 cm: R2 = 0.55; 10–20cm: R2 = 0.37). Soil properties and topography play a key role in modulating SC stocks, with total nitrogen (TN), soil moisture (SM), and elevation being principal drivers with contrasting effects on SC storage, while climate and vegetation parameters are of lesser influence. The relative effect of majority of explanatory drivers reduces with depth while that of temperature increases. Our analyses indicate that shifts in floristic composition could have long-lasting implications on soil structure and C storage, providing valuable data for C sink management.