Design flood estimation for global river networks based on machine learning models

Design flood estimation is a fundamental task in hydrology. In this research, we propose a machine-learning-based approach to estimate design floods globally. This approach involves three stages: (i) estimating at-site flood frequency curves for global gauging stations using the Anderson–Darling test and a Bayesian Markov chain Monte Carlo (MCMC) method; (ii) clustering these stations into subgroups using a K-means model based on 12 globally available catchment descriptors; and (iii) developing a regression model in each subgroup for regional design flood estimation using the same descriptors. A total of 11 793 stations globally were selected for model development, and three widely used regression models were compared for design flood estimation. The results showed that (1) the proposed approach achieved the highest accuracy for design flood estimation when using all 12 descriptors for clustering; and the performance of the regression was improved by considering more descriptors during training and validation; (2) a support vector machine regression provided the highest prediction performance amongst all regression models tested, with a root mean square normalised error of 0.708 for 100-year return period flood estimation; (3) 100-year design floods in tropical, arid, temperate, cold and polar climate zones could be reliably estimated (i.e. <±25 % error), with relative mean bias (RBIAS) values of −0.199, −0.233, −0.169, 0.179 and −0.091 respectively; (4) the machine-learning-based approach developed in this paper showed considerable improvement over the index-flood-based method introduced by Smith et al. (2015, https://doi.org/10.1002/2014WR015814) for design flood estimation at global scales; and (5) the average RBIAS in estimation is less than 18 % for 10-, 20-, 50- and 100-year design floods. We conclude that the proposed approach is a valid method to estimate design floods anywhere on the global river network, improving our prediction of the flood hazard, especially in ungauged areas.

Sampling uncertainty of UK design flood estimation

The UK standard for estimating flood frequencies is outlined by the flood estimation handbook (FEH) and associated updates. Estimates inevitably come with uncertainty due to sampling error as well as model and measurement error. Using resampling approaches adapted to the FEH methods, this paper quantifies the sampling uncertainty for single site, pooled (ungauged), enhanced single site (gauged pooling) and across catchment types. This study builds upon previous progress regarding easily applicable quantifications of FEH-based uncertainty estimation (Kjeldsen 2015, 2021; Dixon 2017). Where these previous studies have provided simple analytical expressions for quantifying uncertainty for single site and ungauged design flow estimates, this study provides an easy-to-use method for quantifying uncertainty for enhanced single site estimates.

Nonstationary Design Flood Estimation in Response to Climate Change, Population Growth and Cascade Reservoir Regulation

The hydrologic data series are nonstationary due to climate change and local anthropogenic activities. The existing nonstationary design flood estimation methods usually focus on the statistical nonstationarity of the flow data series in the catchment, which neglect the hydraulic approach, such as reservoir flood regulation. In this paper, a novel approach to comprehensively consider the driving factors of non-stationarities in design flood estimation is proposed, which involves three main steps: (1) implementation of the candidate predictors with trend tests and change point detection for preliminary analysis; (2) application of the nonstationary flood frequency analysis with the principle of Equivalent Reliability (ER) for design flood volumes; (3) development of a nonstationary most likely regional composition (NS-MLRC) method, and the estimation of a design flood hydrograph at downstream cascade reservoirs. The proposed framework is applied to the cascade reservoirs in the Han River, China. The results imply that: (1) the NS-MLRC method provides a much better explanation for the nonstationary spatial correlation of the flood events in Han River basin, and the multiple nonstationary driving forces can be precisely quantified by the proposed design flood estimation framework; (2) the impacts of climate change and population growth are long-lasting processes with significant risk of flood events compared with stationary distribution conditions; and (3) the swift effects of cascade reservoirs are reflected in design flood hydrographs with lower peaks and lesser volumes. This study can provide a more integrated template for downstream flood risk management under the impact of climate change and human activities.

A Comparative Flood Frequency Analysis of High-Flow between Annual Maximum and Partial Duration Series at Sungai Langat Basin

Flood frequency analysis should consider small and frequent floods. Despite the complexities in partial duration series implementation, it can give a better flood estimation in a way that it does not exclude any significant high flow events, even if it is not the highest event of the year. This study employs the streamflow data recorded at Kajang station, Sungai Langat, Malaysia over a 36-year period spanning from 1978 to 2013. The paper attempts to conduct flood frequency analysis using two approaches, annual maximum and partial duration series. The optimal threshold value is selected to be 48.7 m3/s, where the dispersion index stabilizes at around 1, DI = 1 . The results have shown that generalized extreme value (GEV) distribution describes the annual maximum data while the lognormal (LN3) and generalized Pareto (GPA) distribution is chosen as the best fit distribution at Kajang station for a partial duration series. There is a slight difference between estimated streamflow magnitude when using GPA and LN3 for selected return periods, while a considerable difference was observed when using annual maximum at a higher return period. As a conclusion, PDS gives more relevant magnitude estimation rather than AMS. Flood frequency plays an important role in understanding the nature and magnitude of high flow, which in turn can assist relevant agencies in the design of hydrological structures and reduce flood impacts.

Effects of Probability-Distributed Losses on Flood Estimates Using Event-Based Rainfall-Runoff Models

Probability distributions of initial losses are investigated using a large dataset of catchments throughout Australia. The variability in design flood estimates caused by probability-distributed initial losses and associated uncertainties are investigated. Based on historic data sets in Australia, the Gamma and Beta distributions are found to be suitable for describing initial loss data. It has also been found that the central tendency of probability-distributed initial loss is more important in design flood estimation than the form of the probability density function. Findings from this study have notable implications on the regionalization of initial loss data, which is required for the application of Monte Carlo methods for design flood estimation in ungauged catchments.

Design flood estimation with varying record lengths in Norway under stationarity and nonstationarity scenarios

In traditional flood frequency analysis, a minimum of 30 observations is required to guarantee the accuracy of design results with an allowable uncertainty; however, there has not been a recommendation for the requirement on the length of data in NFFA (nonstationary flood frequency analysis). Therefore, this study has been carried out with three aims: (i) to evaluate the predictive capabilities of nonstationary (NS) and stationary (ST) models with varying flood record lengths; (ii) to examine the impacts of flood record lengths on the NS and ST design floods and associated uncertainties; and (iii) to recommend the probable requirements of flood record length in NFFA. To achieve these objectives, 20 stations with record length longer than 100 years in Norway were selected and investigated by using both GEV (generalized extreme value)-ST and GEV-NS models with linearly varying location parameter (denoted by GEV-NS0). The results indicate that the fitting quality and predictive capabilities of GEV-NS0 outperform those of GEV-ST models when record length is approximately larger than 60 years for most stations, and the stability of the GEV-ST and GEV-NS0 is improved as record lengths increase. Therefore, a minimum of 60 years of flood observations is recommended for NFFA for the selected basins in Norway.