Probabilistic rainfall thresholds for flood forecasting: evaluating different methodologies for modelling rainfall spatial correlation (or dependence)

Rainfall threshold (RT) method is one of the evolving flood forecasting approaches. When the cumulative rainfall depth for a given initial soil moisture condition intersects the threshold rainfall curve, the peak discharge is expected to be equal or greater than the threshold discharge for flooding at the target site. Besides the total rainfall depth, spatial and temporal distribution of rainfall impacts the flood peak discharge and the time to peak. To revisit a previous study conducted by the authors, in which spatially independent rainfall pattern was assumed, the spatial distribution of rainfall was simulated following a Monte Carlo approach. The structure of the spatial dependence among sub‐watersheds' rainfalls was taken into account under three different scenarios, namely independent, bivariate copula (2copula) and multivariate Gaussian copula (MGC). For each set of generated random dimensionless rainfalls, the probabilistic RT curves were derived for dry moisture condition. Results were evaluated with both historical and simulated events. For the simulated events, threshold curves were assessed by means of categorical statistics, such as hit rate, false rate and critical success index (CSI). Results revealed that the best performance based on the CSI criterion corresponded to 50% curve in 2copula and MGC scenarios as well as 90% curve in the independent scenario. The recognition of 50% curve in 2copula and MGC scenarios is in agreement with our expectations that the mean probable curve should have the best performance. Moreover, the proposed inclusion of spatially dependent rainfall scenario improved the performance of RT curves by about 25% in comparison with the presumed spatially uniform rainfall scenario. Copyright © 2011 John Wiley & Sons, Ltd.     

Radar rainfall estimation for flash flood forecasting in small urban watersheds

Radar rainfall estimation for flash flood forecasting in small, urban catchments is examined through analyses of radar, rain gage and discharge observations from the 14.3 km² Dead Run drainage basin in Baltimore County, Maryland. The flash flood forecasting problem pushes the envelope of rainfall estimation to time and space scales that are commensurate with the scales at which the fundamental governing laws of land surface processes are derived. Analyses of radar rainfall estimates are based on volume scan WSR-88D reflectivity observations for 36 storms during the period 2003–2005. Gage-radar analyses show large spatial variability of storm total rainfall over the 14.3 km² basin for flash flood producing storms. The ability to capture the detailed spatial variation of rainfall for flash flood producing storms by WSR-88D rainfall estimates varies markedly from event to event. As spatial scale decreases from the 14.3 km² scale of the Dead Run watershed to 1 km² (and the characteristic time scale of flash flood producing rainfall decreases from 1 h to 15 min) the predictability of flash flood response from WSR-88D rainfall estimates decreases sharply. Storm to storm variability of multiplicative bias in storm total rainfall estimates is a dominant element of the error structure of radar rainfall estimates, and it varies systematically over the warm season and with flood magnitude. Analyses of the 7 July 2004 and 28 June 2005 storms illustrate microphysical and dynamical controls on radar estimation error for extreme flash flood producing storms.     

Modelling bivariate extreme precipitation distribution for data‐scarce regions using Gumbel–Hougaard copula with maximum entropy estimation

A new method of parameter estimation in data scarce regions is valuable for bivariate hydrological extreme frequency analysis. This paper proposes a new method of parameter estimation (maximum entropy estimation, MEE) for both Gumbel and Gumbel–Hougaard copula in situations when insufficient data are available. MEE requires only the lower and upper bounds of two hydrological variables. To test our new method, two experiments to model the joint distribution of the maximum daily precipitation at two pairs of stations on the tributaries of Heihe and Jinghe River, respectively, were performed and compared with the method of moments, correlation index estimation, and maximum likelihood estimation, which require a large amount of data. Both experiments show that for the Ye Niugou and Qilian stations, the performance of MEE is nearly identical to those of the conventional methods. For the Xifeng and Huanxian stations, MEE can capture information indicating that the maximum daily precipitation at the Xifeng and Huanxian stations has an upper tail dependence, whereas the results generated by correlation index estimation and maximum likelihood estimation are unreasonable. Moreover, MEE is proved to be generally reliable and robust by many simulations under three different situations. The Gumbel–Hougaard copula with MEE can also be applied to the bivariate frequency analysis of other extreme events in data‐scarce regions.     

Estimation and spatial interpolation of rainfall intensity distribution from the effective rate of precipitation

Great emphasis is being placed on the use of rainfall intensity data at short time intervals to accurately model the dynamics of modern cropping systems, runoff, erosion and pollutant transport. However, rainfall data are often readily available at more aggregated level of time scale and measurements of rainfall intensity at higher resolution are available only at limited stations. A distribution approach is a good compromise between fine-scale (e.g. sub-daily) models and coarse-scale (e.g. daily) rainfall data, because the use of rainfall intensity distribution could substantially improve hydrological models. In the distribution approach, the cumulative distribution function of rainfall intensity is employed to represent the effect of the within-day temporal variability of rainfall and a disaggregation model (i.e. a model disaggregates time series into sets of higher solution) is used to estimate distribution parameters from the daily average effective precipitation. Scaling problems in hydrologic applications often occur at both space and time dimensions and temporal scaling effects on hydrologic responses may exhibit great spatial variability. Transferring disaggregation model parameter values from one station to an arbitrary position is prone to error, thus a satisfactory alternative is to employ spatial interpolation between stations. This study investigates the spatial interpolation of the probability-based disaggregation model. Rainfall intensity observations are represented as a two-parameter lognormal distribution and methods are developed to estimate distribution parameters from either high-resolution rainfall data or coarse-scale precipitation information such as effective intensity rates. Model parameters are spatially interpolated by kriging to obtain the rainfall intensity distribution when only daily totals are available. The method was applied to 56 pluviometer stations in Western Australia. Two goodness-of-fit statistics were used to evaluate the skill—daily and quantile coefficient of efficiency between simulations and observations. Simulations based on cross-validation show that kriging performed better than other two spatial interpolation approaches (B-splines and thin-plate splines).     

A dependence modelling study of extreme rainfall in Madeira Island

The dependence between variables plays a central role in multivariate extremes. In this paper, spatial dependence of Madeira Island's rainfall data is addressed within an extreme value copula approach through an analysis of maximum annual data. The impact of altitude, slope orientation, distance between rain gauge stations and distance from the stations to the sea are investigated for two different periods of time. The results obtained highlight the influence of the island's complex topography on the spatial distribution of extreme rainfall in Madeira Island.     

The implications of regional variations in rainfall for reconstructing rainfall patterns using tree rings

In Australia, multidecadal periods of floods and droughts have major economic consequences. Due to the short duration of Australian instrumental precipitation records, it is difficult to determine the patterns of these multidecadal periods. Proxy records can be used to create long‐term rainfall reconstructions for regions that are lacking instrumental data. However, the spatial extent over which single‐site proxy records can be applied is poorly understood. Southeast Queensland (SEQ) is an area where tree rings can be used to reconstruct long‐term rainfall patterns, but their regional representation is unknown. In this study, the spatial variability in rainfall across SEQ is investigated from 1908 to 2007 using 140 instrumental rainfall stations. Pearson correlation analysis between stations is used to create groups at the r = 0.80, 0.85, and 0.90 correlation levels, and then annual deviations from the mean are determined. These patterns indicate that rainfall is not uniform across SEQ but can be broken into 2 main spatially consistent groups. Each of these groups is broken down into several subgroups with higher correlation levels. Long‐term streamflow records are found to be correlated to rainfall patterns local to the streamflow stations, indicating that analysis of extreme events should consider spatial precipitation variability. Finally, the only currently available proxy rainfall reconstruction for the region, a 140‐year Toona ciliata tree ring width record from Lamington National Park, is compared to rainfall groups at different correlation levels across all of SEQ. The correlation between the reconstruction and the rainfall station groupings is best for the groups within which the tree‐ring record is spatially located, and this correlation improves as rainfall group correlation increases. Correlation is nearly nonexistent for groupings located at a distance from the tree‐ring site. These results demonstrate the importance of assessing the spatial variability of precipitation so that the spatial applicability of proxy records can be assessed.     

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.     

Modelling bivariate rainfall distribution and generating bivariate correlated rainfall data in neighbouring meteorological subdivisions using copula

The rainfall patterns of neighbouring meteorological subdivisions of India are similar because of similar climatological and geographical characteristics. Analysing the rainfall pattern separately for these meteorological subdivisions may not always capture the correlation and tail dependence. Furthermore, generating the multivariate rainfall data separately may not preserve the correlation. In this study, copula method is used to derive the bivariate distribution of monsoon rainfall in neighbouring meteorological subdivisions. Different Archimedean copulas are used for this purpose and the best copula is selected based on nonparametric test and tail dependence coefficient. The fitted copula is then applied to derive the bivariate distribution, joint return period and conditional distribution. Bivariate rainfall data is generated with the fitted copula and it is observed with the increase of sample size, the generated data is able to capture the correlation as well as tail dependence. The methodology is demonstrated with the case study of two neighbouring meteorological subdivisions of North‐East India: Assam and Meghalaya meteorological subdivision and Nagaland, Manipur, Mizoram and Tripura meteorological subdivision. Copyright © 2010 John Wiley & Sons, Ltd.     

A simple model for spatial rainfall fields

Spatial rainfall amounts accumulated over short to medium periods of time, say a few days, tend to have a probabilistic structure with very distinctive features. Some of these that are specially relevant for the purpose of spatial modeling are the presence of mixed sampling distributions, right skewed distributions conditional on rainfall occurrence, and a complex spatial association structure. The goal of this work is to construct a family for the bivariate distributions of spatial rainfall fields that incorporates these distinctive features. It is based on the separate modeling of spatial occurrence of rainfall and the spatial distribution of positive rainfalls. The main properties of the bivariate distributions are derived, and some properties of the random field realizations are illustrated through simulation. Some limitations of the proposed model are also discussed.     

Frequency analysis of droughts using the Plackett copula and parameter estimation by genetic algorithm

This study aims to model the joint probability distribution of drought duration, severity and inter-arrival time using a trivariate Plackett copula. The drought duration and inter-arrival time each follow the Weibull distribution and the drought severity follows the gamma distribution. Parameters of these univariate distributions are estimated using the method of moments (MOM), maximum likelihood method (MLM), probability weighted moments (PWM), and a genetic algorithm (GA); whereas parameters of the bivariate and trivariate Plackett copulas are estimated using the log-pseudolikelihood function method (LPLF) and GA. Streamflow data from three gaging stations, Zhuangtou, Taian and Tianyang, located in the Wei River basin, China, are employed to test the trivariate Plackett copula. The results show that the Plackett copula is capable of yielding bivariate and trivariate probability distributions of correlated drought variables.     

Geospatial–temporal dependence among weekly precipitation extremes with applications to observations and climate model simulations in South America

A quantification of the spatio-temporal dependence among precipitation extremes is important for investigating the properties of intense storms as well as flood or flash-flood related hazards. Extreme value theory has been widely applied to the hydrologic sciences and hydraulic engineering. However, rigorous approaches to quantify dependence structures among extreme values in space and time have not been reported in the literature. Previous researchers have quantified the dependence among extreme values through the concept of (pairwise bivariate) tail dependence coefficients. For estimation of the tail dependence coefficients, we apply a recently developed method [Kuhn G. On dependence and extremes. PhD thesis (Advisor: C. Klüppelberg), Munich University of Technology, 2006] which utilized the multivariate tail dependence function of a subclass of elliptical copulas. This study extends the previous approach in the context of space and time by considering pairs of spatial grids in South America and quantifying the dependence among precipitation extremes based on the time series at each spatial grid. In addition, Kendall’s τ is used to estimate the pairwise copula correlation (for an elliptical copula) of precipitation between all grids in South America. The geospatial–temporal dependence measures are applied to precipitation observations from 1940 to 2005 as well as simulations from the Community Climate System Model version 3 (CCSM3) for 1940–2099. New insights are obtained regarding the spatio-temporal dependence structures for precipitation over South America both with regard to correlation as well as tail dependence.     

Extension of observed flood series by combining a distributed hydro-meteorological model and a copula-based model

Long flood series are required to accurately estimate flood quantiles associated with high return periods, in order to design and assess the risk in hydraulic structures such as dams. However, observed flood series are commonly short. Flood series can be extended through hydro-meteorological modelling, yet the computational effort can be very demanding in case of a distributed model with a short time step is considered to obtain an accurate flood hydrograph characterisation. Statistical models can also be used, where the copula approach is spreading for performing multivariate flood frequency analyses. Nevertheless, the selection of the copula to characterise the dependence structure of short data series involves a large uncertainty. In the present study, a methodology to extend flood series by combining both approaches is introduced. First, the minimum number of flood hydrographs required to be simulated by a spatially distributed hydro-meteorological model is identified in terms of the uncertainty of quantile estimates obtained by both copula and marginal distributions. Second, a large synthetic sample is generated by a bivariate copula-based model, reducing the computation time required by the hydro-meteorological model. The hydro-meteorological modelling chain consists of the RainSim stochastic rainfall generator and the Real-time Interactive Basin Simulator (RIBS) rainfall-runoff model. The proposed procedure is applied to a case study in Spain. As a result, a large synthetic sample of peak-volume pairs is stochastically generated, keeping the statistical properties of the simulated series generated by the hydro-meteorological model. This method reduces the computation time consumed. The extended sample, consisting of the joint simulated and synthetic sample, can be used for improving flood risk assessment studies.     

Data‐based analysis of bivariate copula tail dependence for drought duration and severity

In recent decades, copula functions have been applied in bivariate drought duration and severity frequency analysis. Among several potential copulas, Clayton has been mostly used in drought analysis. In this research, we studied the influence of the tail shape of various copula functions (i.e. Gumbel, Frank, Clayton and Gaussian) on drought bivariate frequency analysis. The appropriateness of Clayton copula for the characterization of drought characteristics is also investigated. Drought data are extracted from standardized precipitation index time series for four stations in Canada (La Tuque and Grande Prairie) and Iran (Anzali and Zahedan). Both duration and severity data sets are positively skewed. Different marginal distributions were first fitted to drought duration and severity data. The gamma and exponential distributions were selected for drought duration and severity, respectively, according to the positive skewness and Kolmogorov–Smirnov test. The results of copula modelling show that the Clayton copula function is not an appropriate choice for the used data sets in the current study and does not give more drought risk information than an independent model for which the duration and severity dependence is not significant. The reason is that the dependence of two variables in the upper tail of Clayton copula is very weak and similar to the independent case, whereas the observed data in the transformed domain of cumulative density function show high association in the upper tail. Instead, the Frank and Gumbel copula functions show better performance than Clayton function for drought bivariate frequency analysis. Copyright © 2012 John Wiley & Sons, Ltd.     

Drought hotspot analysis and risk assessment using probabilistic drought monitoring and severity–duration–frequency analysis

Drought hotspot identification requires continuous drought monitoring and spatial risk assessment. The present study analysed drought events in the agriculture‐dominated mid‐Mahanadi River Basin in Odisha, India, using crop water stress as a drought indicator. This drought index incorporated different factors that affect crop water deficit such as the cropping pattern, soil characteristics, and surface soil moisture. The drought monitoring framework utilized a relevance vector machine model‐based classification that provided the uncertainty associated with drought categorization. Using the proposed framework, drought hotspots are identified in the study region and compared with indices based on precipitation and soil moisture. Further, a bivariate copula is employed to model the agricultural drought characteristics and develop the drought severity–duration–frequency (S–D–F) relationships. The drought hotspot maps and S–D–F curves are developed for different locations in the region. These provided useful information on the site‐specific drought patterns and the characteristics of the devastating droughts of 2002 and 2012, characterized by an average drought duration of 7 months at several locations. The site‐specific risk of short‐ and long‐term agricultural droughts are then investigated using the conditional copula. The results suggest that the conditional return periods and the S–D–F curves are valuable tools to assess the spatial variability of drought risk in the region.     

Analysis and modeling of spatial correlation structure in small-scale rainfall in Central Oklahoma

Spatial correlation structure in small-scale rainfall is analyzed based on a dense cluster of raingauges in Central Oklahoma. This cluster, called the EVAC PicoNet, consists of 53 gauges installed in 25 measurement stations covering an area of about 3 km by 3 km. Two raingauges are placed in 24 stations and five in the central station. Three aspects of the estimated spatial correlation functions are discussed: dependence on time-scale ranging from 1 min to 24 h, inter-storm variability, and dependence on rainfall intensity. The results show a regular dependence of the correlogram parameters on the averaging time-scale, large differences of the correlograms in the individual storms, and the dominance of storms with high spatial variability on the average large sample characteristics. The authors also demonstrate and discuss the ambiguities in correlation estimates conditioned on rainfall intensities. The findings of this study have implications for raingauge network design, rainfall modeling, and conclusive evaluation of radar and satellite estimates of rainfall.     

Copula‐based uncertainty modelling: application to multisensor precipitation estimates

The multisensor precipitation estimates (MPE) data, available in hourly temporal and 4 km × 4 km spatial resolution, are produced by the National Weather Service and mosaicked as a national product known as Stage IV. The MPE products have a significant advantage over rain gauge measurements due to their ability to capture spatial variability of rainfall. However, the advantages are limited by complications related to the indirect nature of remotely sensed precipitation estimates. Previous studies confirm that efforts are required to determine the accuracy of MPE and their associated uncertainties for future use in hydrological and climate studies. So far, various approaches and extensive research have been undertaken to develop an uncertainty model. In this paper, an ensemble generator is presented for MPE products that can be used to evaluate the uncertainty of rainfall estimates. Two different elliptical copula families, namely, Gaussian and t‐copula are used for simulations. The results indicate that using t‐copula may have significant advantages over the well‐known Gaussian copula particularly with respect to extremes. Overall, the model in which t‐copula was used for simulation successfully generated rainfall ensembles with similar characteristics to those of the ground reference measurements. Copyright © 2010 John Wiley & Sons, Ltd.     

Copula-based mixed models for bivariate rainfall data: an empirical study in regression perspective

A comprehensive parametric approach to study the probability distribution of rainfall data at scales of hydrologic interest (e.g. from few minutes up to daily) requires the use of mixed distributions with a discrete part accounting for the occurrence of rain and a continuous one for the rainfall amount. In particular, when a bivariate vector (X, Y) is considered (e.g. simultaneous observations from two rainfall stations or from two instruments such as radar and rain gauge), it is necessary to resort to a bivariate mixed model. A quite flexible mixed distribution can be defined by using a 2-copula and four marginals, obtaining a bivariate copula-based mixed model. Such a distribution is able to correctly describe the intermittent nature of rainfall and the dependence structure of the variables. Furthermore, without loss of generality and with gain of parsimony this model can be simplified by some transformations of the marginals. The main goals of this work are: (1) to empirically explore the behaviour of the parameters of marginal transformations as a function of time scale and inter-gauge distance, by analysing data from a network of rain gauges; (2) to compare the properties of the regression curves associated to the copula-based mixed model with those derived from the model simplified by transformations of the marginals. The results from the investigation of transformations’ parameters are in agreement with the expected theoretical dependence on inter-gauge distance, and show dependence on time scale. The analysis on the regression curves points out that: (1) a copula-based mixed model involves regression curves quite close to some non-parametric models; (2) the performance of the parametric regression decreases in the same cases in which non-parametric regression shows some instability; (3) the copula-based mixed model and its simplified version show similar behaviour in term of regression for mid-low values of rainfall. An erratum to this article can be found at     

One-minute rain rate distribution in Nigeria derived from TRMM satellite data

Data from the Tropical Rainfall Measuring Mission (TRMM) satellite sensors, the Microwave Imager (TMI, 3A12 V6) and other satellite sources (3B43 V6) have been used to derive the thunderstorm ratio β, total rain accumulation M, and 1-min rainfall rates, R1min, for 37 stations in Nigeria, for 0.001–1% of an average year, for the period 1998–2006. Results of the rain accumulations from the TRMM satellite (1998–2006) were compared with the data collected from 14 ground stations in Nigeria for the period 1991–2000. The two data sets are reasonably positively correlated, with correlation coefficients varying from 0.64 to 0.99. Deduced 1-min rainfall rates compared fairly well with the previous ground data of Ajayi and Ezekpo (1988. Development of climatic maps of rainfall rate and attenuation for microwave applications in Nigeria. The Nigerian Engineering 23(4), 13–30) with correlation coefficients varying from 0.17 to 0.97 in all 37 stations. The agreement was much better when compared with the International Telecommunications Union Radio communication Study group 3 digital maps with correlation coefficients varying from 0.84 to 0.98 in 23 locations; however there were negative correlation coefficient (of 0.2 in 7 stations) in the Middle Belt and a weak positive coefficient (of 0.09 in 6 stations) in the South South. Regionally the inferred mean annual 1-min rainfall rates are the highest in the South-East region with values between 111 and 125 mm/h throughout the 9 years, followed by the South-South region (105–124 mm/h). The lowest rainfall rate and rainfall accumulation occur in the North-West region (60–86 mm/h) followed, in ascending order, by the North-East (66–95 mm/h) region, the Middle-Belt region (76–102 mm/h) and the South-West region (77–110 mm/h). The present results were also compared with 9 tropical stations around the world and there was positive correlation between the results. The present results will be very useful for satellite rain attenuation modeling in the tropics and subtropical stations around the world.It is useful to note that one country, particularly one as large as Nigeria, can have significant variations in its rainfall characteristics for a variety of reasons, and this is borne out by the results presented.