Scale- and location-specific multivariate controls of topsoil organic carbon density depend on landform heterogeneity

Abstract

Spatial variability of surface soil organic carbon density (SOCD) and its controls may be scale-specific and localized depending on the landforms which drive distribution of environmental factors such as soil moisture and soil texture. We hypothesize that SOCD variation dominates at larger scales and is controlled by less number of environmental variables at more homogeneous landforms. The objective of this study was to test this hypothesis by revealing the scale-location specific multivariate relationships between surface SOCD and environmental factors at three transects (i.e., upstream, midstream, and downstream) with different landform heterogeneities in a typical basin using bivariate wavelet coherency and multiple wavelet coherence. Results indicated that variations in SOCD was dominated at large scales (>10.6 km) across the entire transect (i.e., global) at the more homogeneous midstream transect, while those were more localized and mainly occurred at small (<2.6 km, upstream) or middle scales (2.6–10.6 km, downstream) at the other two transects. Soil water content was the best and could significantly explain the SOCD variations in 58% of the scale-location domains with significant (68% significance level) variations of SOCD at the midstream transect. However, more (two and seven at upstream and downstream, respectively) variables were needed but did not obviously increase (64% at downstream) or even decreased (23% at upstream) the percentage of scale-location domains where SOCD variations were significantly explained. This study verified that variations in SOCD and its scale-location specific controls were associated with landform heterogeneity. This study also indicated the outperformance of wavelet methods over traditional correlation analyses in terms of untangling the controls of multivariate factors on SOCD variation at different locations and scales. Meanwhile, a new index (percentage area of both significant coherence and significant SOCD variations relative to domains with significant SOCD variation, PASCSV) was developed, and was proved to be better than a previously used index (percentage area of significant coherence relative to all scale-location domains, PASC) in identifying the best factors in controlling SOCD variations in multiple location-scale domains.