Toward more robust extreme flood prediction by Bayesian hierarchical and multimodeling

Regional flood frequency analysis is commonly applied in situations where there exists insufficient data at a location for a reliable estimation of flood quantiles. We develop a Bayesian hierarchical modeling framework for a regional analysis of data from 203 large catchments in Norway with the generalized extreme value distribution as the underlying model. Generalized linear models on the parameters of the generalized extreme value distribution are able to incorporate location‐specific geographic and meteorological information and thereby accommodate these effects on the flood quantiles. A Bayesian model averaging component additionally assesses model uncertainty in the effect of the proposed covariates. The resulting regional model is seen to give substantially better predictive performance than the regional model currently used in Norway.

Section: Discussionmentioning

confidence: 88%

Bayesian Regional Flood Frequency Analysis for Large Catchments

Thorarinsdottir

Hellton

Steinbakk

et al. 2018

“…Third, the uncertainty in the frequency analysis associated with the types of distribution is generally small. Many studies have suggested that the major uncertainty in frequency analysis is contributed by the data themselves, while the choice of distribution plays a much less important role (Fassnacht & Records, ; Hosking & Wallis, ; Stedinger & Griffis, ; Yan & Moradkhani, ).…”

Section: Resultsmentioning

confidence: 99%

Next‐Generation Intensity‐Duration‐Frequency Curves for Hydrologic Design in Snow‐Dominated Environments

Yan

Sun

Wigmosta

et al. 2018

Self Cite

There is a renewed focus on the design of infrastructure resilient to extreme hydrometeorological events. While precipitation‐based intensity‐duration‐frequency (IDF) curves are commonly used as part of infrastructure design, a large percentage of peak runoff events in snow‐dominated regions are caused by snowmelt, particularly during rain‐on‐snow (ROS) events. In these regions, precipitation‐based IDF curves may lead to substantial overestimation/underestimation of design basis events and subsequent overdesign/underdesign of infrastructure. To overcome this deficiency, we proposed next‐generation IDF (NG‐IDF) curves, which characterize the actual water reaching the land surface. We compared NG‐IDF curves to standard precipitation‐based IDF curves for estimates of extreme events at 376 Snowpack Telemetry (SNOTEL) stations across the western United States that each had at least 30 years of high‐quality records. We found standard precipitation‐based IDF curves at 45% of the stations were subject to underdesign, many with significant underestimation of 100 year extreme events, for which the precipitation‐based IDF curves can underestimate water potentially available for runoff by as much as 125% due to snowmelt and ROS events. The regions with the greatest potential for underdesign were in the Pacific Northwest, the Sierra Nevada Mountains, and the Middle and Southern Rockies. We also found the potential for overdesign at 20% of the stations, primarily in the Middle Rockies and Arizona mountains. These results demonstrate the need to consider snow processes in the development of IDF curves, and they suggest use of the more robust NG‐IDF curves for hydrologic design in snow‐dominated environments.

“…Our approach could benefit from more sophisticated techniques to choose the family of conditional distributions and to handle situations of zero flow. Previous studies have tried to address structural uncertainty using techniques like Bayesian model averaging and composite likelihoods [ Wang et al ., ; Yan and Moradkhani , ], which could be explored more formally for the model developed here.…”

Section: Discussionmentioning

confidence: 99%

A hierarchical Bayesian model for regionalized seasonal forecasts: Application to low flows in the northeastern United States

Ahn

Palmer

2017

This study presents a regional, probabilistic framework for seasonal forecasts of extreme low summer flows in the northeastern United States conditioned on antecedent climate and hydrologic conditions. The model is developed to explore three innovations in hierarchical modeling for seasonal forecasting at ungaged sites: (1) predictive climate teleconnections are inferred directly from ocean fields instead of predefined climate indices, (2) a parsimonious modeling structure is introduced to allow climate teleconnections to vary spatially across streamflow gages, and (3) climate teleconnections and antecedent hydrologic conditions are considered jointly for regional forecast development. The proposed model is developed and calibrated in a hierarchical Bayesian framework to pool regional information across sites and enhance regionalization skill. The model is validated in a cross‐validation framework along with five simpler nested formulations to test specific hypotheses embedded in the full model structure. Results indicate that each of the three innovations improve out‐of‐sample summer low‐flow forecasts, with the greatest benefits derived from the spatially heterogeneous effect of climate teleconnections. We conclude with a discussion of possible model improvements from a better representation of antecedent hydrologic conditions at ungaged sites.