Comparing the use of the traditional and repeated-sampling-approach of the 137Cs technique in soil erosion estimation

Two approaches have been used in the literature to estimate soil erosion using 137Cs measurements. The traditional-approach requires only one set of samples and uses a single reference 137Cs level, determined off-site, for the whole site. The repeated-sampling-approach requires two sets of samples and uses point-specific values, the 137Cs inventories of the first set of samples, as the reference 137Cs levels. No study has been carried out to investigate the advantages and limitations of these two approaches. In this study, a hay-pasture site, located in Ontario, Canada, with a short period of corn (Zea mays L.) production was examined. Total erosion was estimated using both the traditional- and repeated-sampling-approach. Water and tillage erosion were assessed using established models. The sum of water and tillage erosion served as the model-predicted total soil erosion, assuming negligible wind erosion. The site was classified into five landform elements and the 137Cs-estimated and model-predicted erosion rates were grouped into the landform elements. It was determined that when the reference 137Cs level is accurate, the traditional-approach can produce accurate soil erosion rates, providing that errors due to the conversion models are negligible. However, when there is a large error in the reference 137Cs level, the estimated erosion rates are biased. The repeated-sampling-approach eliminates the bias of reference 137Cs level but has higher random errors. However, the random errors can be reduced by taking multiple samples and using the spatially averaged erosion rate for data interpretation. Landscape segmentation based on topography provides a useful landform division for describing erosion processes and, therefore, practical units for grouping the sampling points. In addition, repeated-sampling-approach also has many advantages due to a shortened time period. In particular, the uncertainties of simple conversion models can be greatly reduced.