Soil erodibility and its scale-specific controls along a southeast-northwest transect on the Tibetan Plateau

•MEMD was used to decompose K and related 13 environmental variables.•The decomposed scales of the variables were 113.7, 193.4, 319.7 and 669.6 km.•Effects of the influencing factors on K were different at different spatial scales.•Prediction model of K considering the scale effect had higher accuracy. Soil erodibility (K) is a key factor in the prediction of soil erosion. K on the Tibetan Plateau is affected by both soil properties and the alpine environment, but the intensity and scale of the influences of soil and environment variables on K have rarely been studied. This study aimed to identify the scale-specific factors controlling K and their relationships along a 1970-km transect on the plateau using multivariate empirical-mode decomposition. The transect was established in the central part of Tibet Autonomous Region, and 60 sampling points were set at an approximate fixed interval to investigate the properties of surface soil and associated environmental factors. Mean K of the transect was 0.0340 t hm2 h MJ-1 mm-1 hm-2, with a range of 0.0138-0.0464. K and 13 other soil and environmental variables were synchronously decomposed into four intrinsic mode functions (IMFs) and one residual. The spatial scales of IMF1-4 were 114, 193, 320 and 670 km, respectively. The proportion of variation of IMF1 in the total variation for all factors except mean annual precipitation (MAP) ranked first or second, indicating that most of the spatial variation of the variables mainly occurred at small scales. Sand, silt and clay contents had the same correlations with K at the sampling scale and each decomposed scale, and correlations of K with other environmental variables varied with scale due to the different main process scales of the variables and K. The accuracy of the prediction model established considering the scale effect was higher than the accuracy of the model without considering this effect (R2adj = 0.976 vs 0.627). This study identified the main factors influencing K at different spatial scales along a long transect on the Tibetan Plateau and established a K-prediction model considering the effect of scale, which provides a basis for accurately predicting erosion in alpine regions.