BLP3-SP: A Bayesian Log-Pearson Type III Model with Spatial Priors for Reducing Uncertainty in Flood Frequency Analyses

Dan Tian,Lei Wang

doi:10.3390/w14060909

Abstract

Gauge stations have uneven lengths of discharge records owing to the historical hydrologic data collection efforts. For watersheds with limited water data length, the flood frequency model, such as the Log-Pearson Type III, will have large uncertainties. To improve the flood frequency prediction for these watersheds, we propose a Bayesian Log-Pearson Type III model with spatial priors (BLP3-SP), which uses a spatial regression model to estimate the prior distribution of the parameters from nearby stations with longer data records and environmental factors. A Markov chain Monte Carlo (MCMC) algorithm is used to estimate the posterior distribution and associated flood quantiles. The method is validated using a case study watershed with 15 streamflow gauge stations located in the San Jacinto River Basin in Texas, US. The result shows that the BLP3-SP outperforms other choices of the priors for the Bayesian Log-Pearson Type III model by significantly reducing the uncertainty in the flood frequency estimation for the station with short data length. The results have confirmed that the spatial prior knowledge can improve the Bayesian inference of the Log-Pearson Type III flood frequency model for watersheds with short gauge period.

Highlights

A design flood is a hypothetical peak discharge graph representation of previous knowledge of precipitation frequency in an area, which is commonly used to evaluate the construction of dams, bridges, canals, and flood damage desistance systems
Spatial information expansion (SIE) is a technique used to employ the knowledge learned from nearby sites or sites from similar environments to substitute space from time [2–5], in order to improve the accuracy of the flood frequency estimate at the site of interest
This paper proposes a Bayesian Log-Pearson Type III model with spatial priors (BLP3-SP) that considers both spatial proximity and catchment attributes as the prior information to reduce the uncertainty in estimated flood frequency

Summary

Introduction

A design flood is a hypothetical peak discharge graph representation of previous knowledge of precipitation frequency in an area, which is commonly used to evaluate the construction of dams, bridges, canals, and flood damage desistance systems. Flood records do not fit any specific known statistical distributions. Bulletin 17C recommends the Log-Pearson Type III (LP3) distribution for design-flood prediction in the United States [1]. The limited length of gauged data is one of the major sources of the uncertainties of the predicted design floods. The longer the gauge records, the more accurately predicted design flood. Most areas are ungauged or recently gauged, leading to large uncertainty in flood frequency models. Spatial information expansion (SIE) is a technique used to employ the knowledge learned from nearby sites or sites from similar environments to substitute space from time [2–5], in order to improve the accuracy of the flood frequency estimate at the site of interest. The assumption is that the hydrological regime of nearby watersheds is similar, resulting in similar flood frequency distribution

Methods

Results

Discussion

Conclusion