How to make 137Cs erosion estimation more useful: An uncertainty perspective

- Erosion estimates are extremely sensitive to random spatial variation of 137Cs.
- Spatial variation of 137Cs is a key contributor to overall estimation uncertainty.
- 137Cs technique can only be used to estimate mean erosion rate for a uniform area.
- A large sample size is vital to yield reliable mean erosion estimates.

Although the 137Cs technique has been widely used to provide quantitative soil redistribution estimates since 1980s, no systematic sensitivity and uncertainty analyses have been carried out to evaluate the conversion models. The objectives of this study are to (1) perform sophisticated sensitivity and uncertainty analyses for the three widely-used models to characterize the potential sources of uncertainty including spatial variations on both reference and measuring sites and (2) explore ways to minimize uncertainty and to improve soil redistribution estimation. The normalized sensitivity showed that soil redistribution estimates were extremely sensitive to 137Cs reference and sample inventories, and less sensitive, to the same degree, to bulk density, tillage depth, and particle size correction factor. Uncertainty analysis showed that the spatial variabilities on both reference and measuring sites were predominant contributors to overall uncertainty of soil erosion estimation, followed by particle size correction factor P, with negligible contributions from bulk density and tillage depth, showing that close attention must be paid to 137Cs spatial variability and factor P. In the presence of substantial random spatial variation in 137Cs distribution, the 137Cs technique is not suitable for estimating point soil redistribution rates as is widely perceived in the literature, because part of the random variation in 137Cs distribution is not a result of soil redistribution. Fortunately, the random variation can be overcome statistically by increasing independent sample numbers on both reference and measuring sites and by interpreting soil redistribution rates in terms of mean value for a uniform landform unit or contour transect (slope position) because the random errors tend to cancel out if averaged over a uniform area.