Spatial heterogeneity and environmental controllers of soil organic carbon stocks in a boreal forest

The knowledge of spatial heterogeneity and environmental controllers of soil organic carbon (SOC) stocks is essential for upscaling and predicting SOC dynamics under changing land use and climatic conditions.  This study investigated the spatial variability and intrinsic and extrinsic controllers of SOC stocks in a boreal forest catchment (320 ha) at the International Institute for Sustainable Development Experimental Lakes Area in Ontario, Canada. Forty-seven surface soil (0-30 cm) samples, representative of the spatial variability of topography, surface water flow patterns and vegetation distribution, were obtained within the catchment. Air dried soil samples were sieved to separate gravel (>2 mm) and fine-earth (<2 mm) fractions and were analyzed for SOC concentration using the loss-on-ignition method. Core sample method was used to determine the soil bulk density. SOC concentrations in surface soils showed a large spatial variability (1.2% to 50.4%, CV= 111.3%). Thick organic soil layers in the wetlands of the sub-catchment showed the highest SOC concentrations. The surface soil SOC stocks ranged between 14.5 to 240.5 Mg ha-1 with an average stock of 101.5 Mg ha-1. Spatial autocorrelations of SOC stocks were modelled by calculating relevant variograms. The variability of SOC stocks (sill = 834) was dominated by the random variability (nugget=275) whereas the variability of SOC concentration (sill = 2.5) was dominated by the spatially structured variability (nugget = 0). We found a strong spatial autocorrelation of the SOC concentrations within the catchment, but the SOC stocks were less spatially correlated. This was largely due to the heterogeneity in the thickness of the surface soil layer (10 cm - 30 cm) and in the gravel content (0-28.9%). We found that a large over-estimation of SOC stocks (52.5%) could result if these intrinsic factors are not considered. Extrinsic controllers were generally not significantly related to the SOC stock; Spearman’s rank correlation analysis on the entire dataset showed non-significant relationships between the SOC stock and extrinsic controllers, namely NDVI (r = 0.04) elevation (r = 0.2), slope (r = -0.1) and topographic indices, stream power index (r = -0.1), relative position index (r=-0.2) and plan curvature (r = -0.1). However, regression tree analysis revealed local-scale effects of aspect, NDVI, elevation, and distance to ridge on the SOC stocks. Many forest soil databases lack information of gravel content and soil depth. Thus, upscaling boreal forest SOC stocks without these two key intrinsic controllers can lead to higher uncertainties in  SOC stock estimates. Further, the impacts of extrinsic controllers may vary across heterogenous landscapes. Machine learning-based digital soil mapping techniques such as Random Forest models are more appropriate for incorporating local-scale impacts of extrinsic controllers when upscaling SOC stocks of boreal forest soils.