Estimation of rainfall intensity–duration–frequency curves at ungauged locations using quantile regression methods

Intensity–duration–frequency (IDF) curves of extreme rainfall are used extensively in infrastructure design and water resources management. In this study, a novel regional framework based on quantile regression (QR) is used to estimate rainfall IDF curves at ungauged locations. Unlike standard regional approaches, such as index-storm and at-site ordinary least-squares regression, which are dependent on parametric distributional assumptions, the non-parametric QR approach directly estimates rainfall quantiles as a function of physiographic characteristics. Linear and nonlinear methods are evaluated for both the regional delineation and IDF curve estimation steps. Specifically, delineation by canonical correlation analysis (CCA) and nonlinear CCA (NLCCA) is combined, in turn, with linear QR and nonlinear QR estimation in a regional modelling framework. An exhaustive comparative study is conducted between standard regional methods and the proposed QR framework at sites across Canada. Overall, the fully nonlinear QR framework, which uses NLCCA for delineation and nonlinear QR for estimation of IDF curves at ungauged sites, leads to the best results.     

Derivation of rainfall IDF relations by third-order polynomial normal transform

Establishing the rainfall intensity–duration–frequency (IDF) relations by the conventional method, the use of parametric distribution models has the advantage of automatic compliance of monotonicity condition of rainfall intensity and frequency. However, fitting rainfall data to a distribution separately by individual duration may possibly produce undulation and crossover of IDF curves which does not comply physical reality. This frequently occurs when rainfall record length is relatively short which often is the case. To tackle this problem this study presents a methodological framework that integrates the third-order polynomial normal transform (TPNT) with the least squares (LS) method to establish rainfall IDF relations by simultaneously considering multi-duration rainfall data. The constraints to preserve the monotonicity and non-crossover in the IDF relations can be incorporated easily in the LS-based TPNT framework. Hourly rainfall data at Zhongli rain gauge station in Taiwan with 27-year record are used to establish rainfall IDF relations and to illustrate the proposed methodology. Numerical investigation indicates that the undulation and crossover behavior of IDF curves can be effectively circumvented by the proposed approach to establish reasonable IDF relations.     

Assessment of design rainfall uncertainty for hydrologic engineering applications in Hong Kong

Hydrologic engineering designs and analyses often require the specification of design storm which involves rainfall amount, duration and hyetograph. In practice, the determination of design rainfall in hydrologic engineering applications involves the frequency analysis of extreme rainfalls of different durations and the establishment of rainfall hyetograph for the design event under consideration. Sampling errors exist in the estimation of rainfall depth (or intensity) quantiles from frequency analysis, which will be transmitted in the process of determining the design rainfall hyetograph. This paper presents a practical methodological framework based on the bootstrap resampling scheme to assess the uncertainty features associated with the magnitude of estimated rainfall depth/intensity quantiles and the corresponding design hyetographs. The procedure is implemented to quantify uncertainty of design rainfall hyetograph following the Stormwater Drainage Manual of Hong Kong involving the use of rainfall intensity–duration–frequency (IDF) model. Of particular interesting is that the bootstrap resampling scheme implemented herein is modified to handle unequal record period of annual maximum rainfall data series of different durations and to account for their intrinsic correlations. According to the adopted rainfall IDF model, the design rainfall hyetograph is a function of the IDF model coefficients. Due to the correlation among rainfall quantiles of different durations, the IDF coefficients are found to be strongly related in a nonlinear fashion which should not be ignored in the establishment of the design hyetographs.     

Pooled frequency analysis for intensity–duration–frequency curve estimation

Rainfall intensity–duration–frequency (IDF) curves are used in the design of urban infrastructure. Their estimation is based on rainfall frequency analysis, usually performed on rainfall records from a single gauged station. However, available at‐site record length is often too short to provide accurate estimates for long return periods. In the present study, a general framework for pooled rainfall frequency analysis based on the index‐event model is proposed for IDF estimation at gauged stations. Pooling group formation is defined by the region of influence approach on the basis of the geographical distance similarity measure. Several pooled approaches are defined and evaluated by a procedure through which quantile estimation and uncertainty are assessed. Alternate approaches for the definition of a pooling group are based on different criteria regarding initial pooling group size (and the relationship between size and return period), approaches for assessing pooling group homogeneity, and the use of macroregions in pooling group formation. The proposed framework is applied to identify the preferred approach for pooled rainfall intensity frequency analysis in Canada. Pooled approaches are found to provide more precise estimates than the at‐site approach, especially for long return periods. Pooled parent distribution selection supported the use of the generalized extreme value distribution across the country. Recommendations for pooling group formation include increasing the pooling group size with increases in return period and identifying an appropriate trade‐off between pooling group homogeneity and size for long return periods.     

Regionalization of heavy rainfall to improve climatic design values for infrastructure: case study in Southern Ontario,Canada

Abstract

Southern Ontario, Canada, has been impacted in recent years by many heavy rainfall and flooding events that have exceeded existing historical estimates of infrastructure design rainfall intensity–duration–frequency (IDF) values. These recent events and the limited number of short-duration recording raingauges have prompted the need to research the climatology of heavy rainfall events within the study area, review the existing design IDF methodologies, and evaluate alternative approaches to traditional point-based heavy rainfall IDF curves, such as regional IDF design values. The use of additional data and the regional frequency analysis methodology were explored for the study area, with the objective of validating identified clusters or homogeneous regions of extreme rainfall amounts through Ward's method. As the results illustrate, nine homogeneous regions were identified in Southern Ontario using the annual maximum series (AMS) for daily and 24-h rainfall data from climate and rate-of-rainfall or tipping bucket raingauge (TBRG) stations, respectively. In most cases, the generalized extreme value and logistic distributions were identified as the statistical distributions that provide the best fit for the 24-h and sub-daily rainfall data in the study area. A connection was observed between extreme rainfall variability, temporal scale of heavy rainfall events and location of each homogeneous region. Moreover, the analysis indicated that scaling factors cannot be used reliably to estimate sub-daily and sub-hourly values from 24- and 1-h data in Southern Ontario.

Citation Paixao, E., Auld, H., Mirza, M.M.Q., Klaassen, J. & Shephard, M.W. (2011) Regionalization of heavy rainfall to improve climatic design values for infrastructure: case study in Southern Ontario, Canada. Hydrol. Sci. J. 56(7), 1067–1089.     

Regional flood frequency analysis using Bayesian generalized least squares: a comparison between quantile and parameter regression techniques

Regression‐based regional flood frequency analysis (RFFA) methods are widely adopted in hydrology. This paper compares two regression‐based RFFA methods using a Bayesian generalized least squares (GLS) modelling framework; the two are quantile regression technique (QRT) and parameter regression technique (PRT). In this study, the QRT focuses on the development of prediction equations for a flood quantile in the range of 2 to 100 years average recurrence intervals (ARI), while the PRT develops prediction equations for the first three moments of the log Pearson Type 3 (LP3) distribution, which are the mean, standard deviation and skew of the logarithms of the annual maximum flows; these regional parameters are then used to fit the LP3 distribution to estimate the desired flood quantiles at a given site. It has been shown that using a method similar to stepwise regression and by employing a number of statistics such as the model error variance, average variance of prediction, Bayesian information criterion and Akaike information criterion, the best set of explanatory variables in the GLS regression can be identified. In this study, a range of statistics and diagnostic plots have been adopted to evaluate the regression models. The method has been applied to 53 catchments in Tasmania, Australia. It has been found that catchment area and design rainfall intensity are the most important explanatory variables in predicting flood quantiles using the QRT. For the PRT, a total of four explanatory variables were adopted for predicting the mean, standard deviation and skew. The developed regression models satisfy the underlying model assumptions quite well; of importance, no outlier sites are detected in the plots of the regression diagnostics of the adopted regression equations. Based on ‘one‐at‐a‐time cross validation’ and a number of evaluation statistics, it has been found that for Tasmania the QRT provides more accurate flood quantile estimates for the higher ARIs while the PRT provides relatively better estimates for the smaller ARIs. The RFFA techniques presented here can easily be adapted to other Australian states and countries to derive more accurate regional flood predictions. Copyright © 2011 John Wiley & Sons, Ltd.     

Bayesian estimation of intensity–duration–frequency curves and of the return period associated to a given rainfall event

Intensity–duration–frequency (IDF) curves are used extensively in engineering to assess the return periods of rainfall events and often steer decisions in urban water structures such as sewers, pipes and retention basins. In the province of Québec, precipitation time series are often short, leading to a considerable uncertainty on the parameters of the probabilistic distributions describing rainfall intensity. In this paper, we apply Bayesian analysis to the estimation of IDF curves. The results show the extent of uncertainties in IDF curves and the ensuing risk of their misinterpretation. This uncertainty is even more problematic when IDF curves are used to estimate the return period of a given event. Indeed, standard methods provide overly large return period estimates, leading to a false sense of security. Comparison of the Bayesian and classical approaches is made using different prior assumptions for the return period and different estimation methods. A new prior distribution is also proposed based on subjective appraisal by witnesses of the extreme character of the event.     

Entropy estimation of hydrological extremes

The method of Relative Entropy with Fractile constraints (REF method) is explained and applied to model extreme compound hydrological phenomena, such as extreme sea levels under storm conditions. Also presented is a simple method of Tail Entropy Approximation (TEA), which amounts to a correction of traditional statistical estimates for extreme observations.Distribution assumptions are necessary but downplayed in the REF method, relegating the prior distribution to the role of an extrapolation function. The estimates are objective in an information-theoretical sense. They also satisfy a strict requirement of self-consistency that is generally not satisfied by standard statistical methods: invariance under monotonic transformations of the random variable.Historical records of storm surge levels in the Netherlands and annual maximum tidal heights for Sheerness, UK, are used as examples. Comparison is made with distributions obtained using other methods.It is concluded that the tail entropy approximation provides simple, objective estimates of extremes in the tail beyond the range of observations.     

Using regional regression within index flood procedures and an empirical Bayesian estimator

Studies have illustrated the performance of at-site and regional flood quantile estimators. For realistic generalized extreme value (GEV) distributions and short records, a simple index-flood quantile estimator performs better than two-parameter (2P) GEV quantile estimators with probability weighted moment (PWM) estimation using a regional shape parameter and at-site mean and L-coefficient of variation (L-CV), and full three-parameter at-site GEV/PWM quantile estimators. However, as regional heterogeneity or record lengths increase, the 2P-estimator quickly dominates. This paper generalizes the index flood procedure by employing regression with physiographic information to refine a normalized T-year flood estimator. A linear empirical Bayes estimator uses the normalized quantile regression estimator to define a prior distribution which is employed with the normalized 2P-quantile estimator. Monte Carlo simulations indicate that this empirical Bayes estimator does essentially as well as or better than the simpler normalized quantile regression estimator at sites with short records, and performs as well as or better than the 2P-estimator at sites with longer records or smaller L-CV.     

Assessing the relationships between elevation and extreme precipitation with various durations in southern Taiwan using spatial regression models

A spatially autocorrelated effect exists in precipitation of a mountainous basin. This study examines the relationship between maximum annual rainfall and elevation in the Kaoping River Basin of southern Taiwan using spatial regression models (i.e. geographically weighted regression (GWR), simultaneous autoregression (SAR), and conditional autoregression (CAR)). Results show that the GWR, SAR, and CAR models can improve spatial data fitting and provide an enhanced estimation for the rainfall–elevation relationship than the ordinary least squares approach. In particular, GWR achieves the most accurate estimation, and SAR and CAR achieve similar performance in terms of the Akaike information criterion. The relationship between extreme rainfall and elevation for longer duration is more concise than that for short durations. Results show that the spatial distribution of precipitation depends on elevation and that rainfall patterns in study area are heterogeneous between the southwestern plain and the eastern mountain area. Copyright © 2011 John Wiley & Sons, Ltd.     

Comparative evaluation of Poisson partial duration models for mixed flood populations

Simple homogeneous formulations of two extreme value partial duration flood models are compared to more sophisticated compound formulations in terms of asymptotic performance of quantile estimates. The compound model formulations were developed to model flood series resulting from mixed climatological processes. It was found that only in the case of marked nonhomogeneity in the data samples did the compound formulation of the models offer significant advantages in terms of variance of quantile estimates. However, the estimates from the homogeneous model were significantly biased in the negative direction. This negative bias of quantile estimates from the simple model was even more pronounced when the more sophisticated Weibull model was used as the base.     

Comparative evaluation of Poisson partial duration models for mixed flood populations

Simple homogeneous formulations of two extreme value partial duration flood models are compared to more sophisticated compound formulations in terms of asymptotic performance of quantile estimates. The compound model formulations were developed to model flood series resulting from mixed climatological processes. It was found that only in the case of marked nonhomogeneity in the data samples did the compound formulation of the models offer significant advantages in terms of variance of quantile estimates. However, the estimates from the homogeneous model were significantly biased in the negative direction. This negative bias of quantile estimates from the simple model was even more pronounced when the more sophisticated Weibull model was used as the base.     

Simple generalization approach for intensity–duration–frequency relationships

Rainfall intensity–duration–frequency (IDF) relationships describe rainfall intensity as a function of duration and return period, and they are significant for water resources planning, as well as for the design of hydraulic constructions. In this study, the two‐parameter lognormal (LN2) and Gumbel distributions are used as parent distribution functions. Derivation of the IDF relationship by this approach is quite simple, because it only requires an appropriate function of the mean of annual maximum rainfall intensity as a function of rainfall duration. It is shown that the monotonic temporal trend in the mean rainfall intensity can successfully be described by this parametric function which comprises a combination of the parameters of the quantile function a(T) and completely the duration function b(d) of the separable IDF relationship. In the case study of Aegean Region (Turkey), the IDF relationships derived through this simple generalization procedure (SGP) may produce IDF relationships as successfully as does the well‐known robust estimation procedure (REP), which is based on minimization of the nonparametric Kruskal–Wallis test statistic with respect to the parameters θ and η of the duration function. Because the approach proposed herein is based on lower‐order sample statistics, risks and uncertainties arising from sampling errors in higher‐order sample statistics were significantly reduced. The authors recommend to establish the separable IDF relationships by the SGP for a statistically favorable two‐parameter parent distribution, because it uses the same assumptions as the REP does, it maintains the observed temporal trend in the mean additionally, it is easy to handle analytically and requires considerably less computational effort. Copyright © 2012 John Wiley & Sons, Ltd.     

Selecting the best IDF model by using the multifractal approach

In this work, the multifractal properties of hourly rainfall data recorded at a location in Southern Spain have been related to the scale properties of the corresponding intensity–duration–frequency (IDF) curves. Four parametric models for the IDF curves have been fitted to the quantiles of rainfall obtained using the generalized Pareto frequency distribution function with the extreme data series obtained for the same place. The scaling of the rainfall intensity moments has been analysed, and the empirical moments scaling exponent function has been obtained. The corresponding values of q₁ and γ₁ have been empirical and theoretically calculated and compared with some characteristics of the different IDF models. Thus, the scaling behaviour of IDF curves has been analysed, and the best model has been selected. Copyright © 2012 John Wiley & Sons, Ltd.     

A spatial temporal downscaling approach to development of IDF relations for Perth airport region in the context of climate change

ABSTRACT

Downscaling of climate projections is the most adapted method to assess the impacts of climate change at regional and local scales. This study utilized both spatial and temporal downscaling approaches to develop intensity–duration–frequency (IDF) relations for sub-daily rainfall extremes in the Perth airport area. A multiple regression-based statistical downscaling model tool was used for spatial downscaling of daily rainfall using general circulation models (GCMs) (Hadley Centre’s GCM and Canadian Global Climate Model) climate variables. A simple scaling regime was identified for 30 minutes to 24 hours duration of observed annual maximum (AM) rainfall. Then, statistical properties of sub-daily AM rainfall were estimated by scaling an invariant model based on the generalized extreme value distribution. RMSE, Nash-Sutcliffe efficiency coefficient and percentage bias values were estimated to check the accuracy of downscaled sub-daily rainfall. This proved the capability of the proposed approach in developing a linkage between large-scale GCM daily variables and extreme sub-daily rainfall events at a given location. Finally IDF curves were developed for future periods, which show similar extreme rainfall decreasing trends for the 2020s, 2050s and 2080s for both GCMs.

Editor M.C. Acreman; Associate editor S. Kanae     

Rainfall frequency analysis using a mixed GEV distribution: a case study for annual maximum rainfalls in South Korea

Extreme rainfalls in South Korea result mainly from convective storms and typhoon storms during the summer. A proper way for dealing with the extreme rainfalls in hydrologic design is to consider the statistical characteristics of the annual maximum rainfall from two different storms when determining design rainfalls. Therefore, this study introduced a mixed generalized extreme value (GEV) distribution to estimate the rainfall quantile for 57 gauge stations across South Korea and compared the rainfall quantiles with those from conventional rainfall frequency analysis using a single GEV distribution. Overall, these results show that the mixed GEV distribution allows probability behavior to be taken into account during rainfall frequency analysis through the process of parameter estimation. The resulting rainfall quantile estimates were found to be significantly smaller than those determined using a single GEV distribution. The difference of rainfall quantiles was found to be closely correlated with the occurrence probability of typhoon and the distribution parameters.     

From marginal to conditional probability functions of parameters in a conceptual rainfall–runoff model: an event-based approach

ABSTRACT

A parameter estimation strategy for a conceptual rainfall–runoff (CRR) model applied to a storm sewer system in an urban catchment (Chassieu, Lyon, France) is proposed on the basis of event-by-event Bayesian local calibrations. The marginal distribution formed by locally-estimated parameters is divided into conditional functions, clustering the event-based parameters based on their transferability to similar rainfall events. The conditional functions showed to be consistent with an observed bimodality in the marginal representation, reflecting two different hydrological conditions mainly related to the magnitude of the rainfall intensities (high or low). The improvements achieved by expressing the parameter probability functions into a conditional form are shown in terms of accuracy (Nash-Sutcliffe efficiency criterion), precision (average relative interval length) and reliability (percentage of coverage) for simulating flow rate in 255 and 110 calibration/verification events.     

Bayesian comparison of different rainfall depth–duration–frequency relationships

Depth–duration–frequency curves estimate the rainfall intensity patterns for various return periods and rainfall durations. An empirical model based on the generalized extreme value distribution is presented for hourly maximum rainfall, and improved by the inclusion of daily maximum rainfall, through the extremal indexes of 24 hourly and daily rainfall data. The model is then divided into two sub-models for the short and long rainfall durations. Three likelihood formulations are proposed to model and compare independence or dependence hypotheses between the different durations. Dependence is modelled using the bivariate extreme logistic distribution. The results are calculated in a Bayesian framework with a Markov Chain Monte Carlo algorithm. The application to a data series from Marseille shows an improvement of the hourly estimations thanks to the combination between hourly and daily data in the model. Moreover, results are significantly different with or without dependence hypotheses: the dependence between 24 and 72 h durations is significant, and the quantile estimates are more severe in the dependence case.     

A multiscaling‐based intensity–duration–frequency model for extreme precipitation

Rainfall intensity–duration–frequency (IDF) curves are a standard tool in urban water resources engineering and management. They express how return levels of extreme rainfall intensity vary with duration. The simple scaling property of extreme rainfall intensity, with respect to duration, determines the form of IDF relationships. It is supposed that the annual maximum intensity follows the generalized extreme value (GEV) distribution. As well known, for simple scaling processes, the location parameter and scale parameter of the GEV distribution obey a power law with the same exponent. Although, the simple scaling hypothesis is commonly used as a suitable working assumption, the multiscaling approach provides a more general framework. We present a new IDF relationship that has been formulated on the basis of the multiscaling property. It turns out that the GEV parameters (location and scale) have a different scaling exponent. Next, we apply a Bayesian framework to estimate the multiscaling GEV model and to choose the most appropriate model. It is shown that the model performance increases when using the multiscaling approach. The new model for IDF curves reproduces the data very well and has a reasonable degree of complexity without overfitting on the data.