Spatial Estimations of Soil Properties for Physically-based Soil Erosion Modelling in the Three Gorges Reservoir Area, Central China

Abstract

Soils present a central medium for processes between the environmental spheres, and therefore play a key role in the functioning of terrestrial ecosystems. However, soil erosion as a natural force of landscape evolution adversely affects the capacity of soils to support ecosystem services. Moreover, inadequate agricultural practices, deforestation, and construction activities amplify natural soil loss rates and transform soil erosion to a major threat for managed ecosystems worldwide. Particularly, the Three Gorges Reservoir Area in China is highly susceptible to soil erosion by water. This is attributable to unfavorable environmental conditions, such as rainfall events of high intensity and steep slope inclinations in areas of extensive, but small-scale crop cultivation. Moreover, in the course of the impoundment of the Yangtze River in the area of the Three Gorges, resettlements and accompanied deforestation reinforced the risk of hazardous soil erosion, which attenuates soil productivity and threatens the functioning of the reservoir. Therefore, conservation measures to stabilize steep sloping surfaces have been implemented to mitigate the hazardous effects of soil erosion. However, to assess the conservation measures an efficient tool is required to identify spatial soil erosion patterns in small, mountainous, and data scarce catchments within the Three Gorges Reservoir Area. The present thesis aims to provide an efficient modelling framework that facilitates a detailed quantification of sediment reallocations due to erosive rainfall-runoff events. Therefore, Digital Soil Mapping techniques based on Latin Hypercube Sampling and Random Forest regression were applied to derive spatially distributed data on soil properties and to furnish a physically- and event-based soil erosion model. The soil sampling design was optimized to address the difficult terrain, an integrative use of legacy soil samples, and a reduced sample set size. Furthermore, the present thesis introduces a spatial uncertainty measure, which was used to identify areas for additional sampling to further refine initially processed soil property maps. In addition, continuous data on rainfall, runoff, and sediment yields were obtained to identify erosive rainfall-runoff events and to calibrate the physically-based soil erosion model EROSION 3D. Evaluation of the hypercube sampling design was conducted by comparing it to a simulated Latin Hypercube design without constraints in terms of operability and efficiency adjustments. Using the optimized sample set size of n = 30, the proposed sample design adequately reproduced the variation of terrain parameters, which served as proxies on the target soil properties of coarse, medium, and fine topsoil sand contents. Furthermore, the validity of the approach was assessed by estimating the spatial distribution of the target soil properties and validating the results independently. The results show convincing accuracies with R²-values between 0.59 and 0.71. The adequacy of the uncertainty-guided sampling for refining initial mapping approaches was evaluated by comparing the refined maps of topsoil silt and clay contents to the initial and further mapping approaches that exclusively used random samples from the entire study area. For the comparative analysis, the quality of the approaches was assessed by independent, bootstrap-, and cross-validation. The refined mapping approach performs best, showing a reduced spatial uncertainty of 31% for topsoil silt and 27% for topsoil clay compared to the initial approaches. Using independent validation, the accuracy increases by similar proportions, showing an accuracy of R² = 0.59 for silt and R² = 0.56 for clay. The EROSION 3D model runs were evaluated using the measured sediment yields. The model performs well for large events (sediment yield > 1 Mg) with an average individual model error of 7.5%, while small events show an average error of 36.2%. The focus of analysis was led on the large events to evaluate reallocation patterns. Soil losses occur on approximately 11.1% of the study area with an average soil loss rate of 49.9 Mg ha-1. Soil loss mainly occurs on crop rotation areas with a spatial proportion of 69.2% for ‘corn-rapeseed’ and 69.1% for ‘potato-cabbage’. Deposition occurs on 11% of the study area. Forested areas (9.7%), infrastructure (41%), cropland (corn-rapeseed: 13.6%, potato-cabbage: 11.3%), and grassland (18.4%) are affected by deposition. Since the vast majority of annual sediment yields (80.3%) were associated to a few large erosive events, the modelling framework can be recommended to identify sediment reallocations and to assess conservation measures in small catchments in the Three Gorges Reservoir Area.