Soil loss by water erosion assessment uncertainties – experiences from South Africa

The Universal Soil Loss Equation (USLE) was initially developed to support the implementation of conservation measures to minimize soil loss by water erosion, i.e. sheet and rill erosion, on a local scale and in the context of agricultural land use. The approach was refined over decades and became the Revised Universal Soil Loss Equation (RUSLE).  At the same time the scope of applications has grown considerably. Due to its rather simple structure and relatively low demand for input data, it has been used for the assessment of soil loss from water erosion for ever growing spatial entities, i.e. regional scale catchments or whole countries. Recently this has been applied on a global scale in order to identify global hotspots of soil erosion. This coherent approach for a global comparison is most welcome against the background of the large number of country-specific assessments which are rather difficult to compare.

However, there are two issues of concern. First, one needs to remember that RUSLE-derived soil loss assessments do not account for gully erosion, which might not be linearly scaled with sheet and rill erosion. Furthermore, information on the uncertainties of RUSLE based erosion assessments are not frequently reported. This especially concerns the impact of specific combinations of varying unique input data on the results.

In this contribution we compare in high spatial resolution the results of two RUSLE-based soil loss by water erosion assessments conducted for six 100 by 100 km large study sites in South Africa. The first assessment was conducted about two decades ago and was based on the then available data covering the whole of South Africa. The second assessment is a revision, which includes the latest input data for rainfall erosivity, soil erodibility, the topographic factor as well as land cover and management. When compared, the results of the current soil loss estimates are an order of magnitude lower than the previous estimates. Does this difference represent a temporal trend or just the inherent uncertainties reflecting different input data and slightly different data processing? This is the question discussed in this contribution.