Using cesium-137 to quantify sediment source contribution and uncertainty in a small watershed

• Overland source and sediment are main uncertainty contributors.
• 30 to 50 samples are needed to obtain reliable proportion estimate.
• Proportion of < 63 μm may be scaled up to the proportion of whole soil.
• MC simulation underestimates standard error of mean proportion estimate.
• Weighting adjustment to mixing model often bias mean proportion estimate.

Knowledge of sediment provenance is critical for precision conservation planning and calibration of soil erosion models. The objectives are to evaluate the ability of 137Cs to apportion sediment source contributions, quantify uncertainty of the estimates, and estimate desirable sample numbers. We collected 50 surface soil samples from overland in a watershed (15.6 km2), 28 subsoil samples from gully bank, and 43 sediment samples in channels. The 137Cs activity was measured by γ spectrometry. Proportion means were calculated by solving a linear mixing model. Uncertainty was estimated by the first-order approximation and Monte Carlo (MC) simulation. The 137Cs tracer was conservative in the fraction of < 63 μm. The 137Cs activities differed between sources in the whole watershed at P = 0.05. The mean proportions of the < 63 μm fraction (relative error) predicted for the watershed using the mixing model were 0.42 (35%) for the overland source and 0.58 (25%) for the gully source, with a 95% confidence interval of ± 0.145 for both. Overland source contributed most uncertainty to the proportion estimates, followed by sediment, with minimum from gully. The MC simulation predicted the same mean proportions, but with relative errors being < 3% for both. Compared with the first-order approximation, MC underestimated uncertainty of the means due to the large sample number used in simulation. In general, 30 to 50 samples are needed for each source and sediment to generate reliable estimates using 137Cs. Estimated source proportions of the fine sediment may be converted to proportional contributions of total soil erosion by adjusting by particle sizes and sediment delivery ratio. No adjustment or weighting should be made directly to the mixing model unless conservativeness of the tracer is violated.