BAYESIAN INVERSION OF EROSION MODELS WITH DETRITAL THERMOCHRONOMETRIC DATA

Detrital thermochronometric data have been recognized as a valuable and accessible source of information on catchment erosion, which in turn is relevant to problems of climate and tectonics. Direct inference of this information is not trivial, however, and only the most simple inverse problems have been addressed. We present a new approach to statistical inversion of erosion models with thermochronometric data, applicable to a wide range of erosion problems. This approach relies on the Bayesian interpretation of probability and uses a Markov chain Monte Carlo algorithm for inversion, affording flexibility in the choice of specific model parametrization and straight-forward assessment of uncertainty. Applying this methodology to a published apatite (U-Th)/He dataset from the Sierra Nevada we are able to estimate and validate long-term (10⁶ yr) erosion parameters based purely on detrital samples. We demonstrate that even small (<20 grains) samples can be used to reliably estimate long-term rates of exhumation. In addition, we demonstrate the ability to quantitatively assess spatial variations in short-term (10³ - 10⁴ yr) erosion rates. Additional applications of this modeling framework, such as constraining temporally variable exhumation rates using detrital data (equivalent to fiinding the “break in slope” in a bedrock vertical transect) are shown to be feasible.