Can We Rely on Trend Test for Identifying Non-stationarity in Floods?
Abstract
Global-scale climate change and watershed-scale anthropogenic disturbances are expected to change stationary design flood estimates over time. We propose a hypothesis-testing framework to detect a significant change in design flood estimates between two periods. Design floods for various return periods are estimated from the non-stationary log-Pearson type-III (LP3) distribution using the parametric bootstrapping technique, and these design flood values are compared with the sampling distribution of design flood values estimated over the stationary period. We first tested this hypothesis testing framework for hypothetical scenarios of changes in flood distribution, including shifts in mean, trends in mean and variance, and changes in the skewness of the considered LP3 flood distribution. Later, we also applied the hypothesis testing framework on two basins showing a combination of changes in moments that are similar to the hypothetical flood change scenarios. We noted that a significant trend in the flood time series does not associate with the significant changes in design flood quantiles. Flood quantiles were found changing significantly without even the presence of a significant trend, and vice versa. We also found that changes in different moments of flood distribution impact design flood values corresponding to low and high return periods. As evaluating the trend in flood time-series only corresponds to estimating the changes in the conditional mean of the flood values, the proposed framework provides a more holistic view to evaluate the changes in design flood values considering all the three moments over the nonstationary period. We recommend the application of this hypothesis testing framework as an alternate approach to simple trend tests for evaluating the changes in design flood estimates between two periods.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H45H1477A