The influence of extreme events on basin-scale erosion rates

Climate, especially precipitation, influences basin-scale erosion rates through a complex combination of weathering, erosion, transport, and vegetation cover. Surprisingly, the magnitude of the erosion rate is only weakly correlated with annual precipitation. We hypothesized that not just the mean annual precipitation value, but also the higher-order statistical moments of precipitation would be influential on these processes, and thereby improve the correlation with basin erosion. In particular, prevalence of extreme events - for example, the proportion of rain that falls in the highest 1 in 1000 days - may be key to predicting erosion rates in headwater basins. To test this hypothesis, we compiled daily rainfall data from weather stations around the continental US, and extracted mean, standard deviation, skewness, kurtosis, and proportion of precipitation falling above the 90^th, 99^th, 99.9^th, and 99.99^th percentile values for the period 1950-2000. We also used the NOAA Atlas 14-point precipitation frequency estimates from the NOAA Hydrometeorological Design Studies Center to estimate shorter duration event magnitude and frequency. We paired this climate data with nearby basin erosion rates measured with cosmogenic ¹⁰Be from the database compiled by Mishra et al. (2019) (Mishra A. K., Placzek C., and Jones R., 2019, Coupled influence of precipitation and vegetation on millennial-scale erosion rates derived from ¹⁰Be. PLOS ONE, 14 e0211325). Because of the known strong influence of slope on erosion rate, we limited our analysis to basins with mean slope ≤ 15°. Within the range of climates in the continental US, erosion rates decreased with increasing mean precipitation, likely due to the influence of increased vegetation. Mishra et al. (2019) previously showed that mean precipitation was a weak predictor of erosion rate. In our regression analysis, including the higher moments and the rare-event magnitudes did not increase our ability to predict erosion rate, contrary to our hypothesis. The reason appears to be that all moments and quantiles of the precipitation we tested were highly correlated with mean precipitation, hence they contributed limited new information to the regression. This implies that extreme events are a smaller concern in landscape evolution than we hypothesized.