Automatic procedure for selecting flood events and identifying flood characteristics from daily streamflow data

Highlights

• We propose a generic standardized method for identifying the flood characteristics.

• An automatic procedure to select flood events and identify flood characteristics is constructed.

• We developed a graphical user interface (GUI) for this automatic procedure in MATLAB.

• The validation results show that our method has good applicability to watersheds with diverse characteristics.

Abstract

The selection of flood events and determination of flood characteristics (e.g. start and end dates, peak discharge, volume, and duration) are the first critical steps for flood analyses. To obtain the key flood information accurately, we used the automatic peak over threshold (POT) model for flood sampling and proposed an automatic approach to determine flood characteristics using the master recession curve analysis (MRC) method. We further developed a graphical user interface (GUI) and toolbox for this procedure in MATLAB. Model parameter estimation experiment (MOPEX) data from 423 stations were used to evaluate the proposed method. Our results suggest that the proposed procedure performs well for watersheds with diverse characteristics. The developed toolbox can be conveniently applied to other watersheds for flood sampling and the characterisation of flood events, thus helping reduce the uncertainty in subsequent flood analyses, such as multivariable flood frequency and trend analyses.

Introduction

Floods are serious natural hazards, causing significant damage and affecting millions of people worldwide (Kauffeldt et al., 2016; Bruijn et al., 2019; Qiu et al., 2017). Moreover, with the increase in atmospheric water-holding capacity due to global warming, extreme weather events, especially flood events, occur more frequently (Blöschl et al., 2019; Adhikari et al., 2010; Tanoue et al., 2016). To reduce flood damage and economic losses, it is essential to accurately investigate the variability of flood events and assess flood risks (Zeng et al., 2020). It is well known that the precise selection of flood events and identification of flood characteristics from daily streamflow data is the most critical steps for subsequent flood related research (e.g. multivariable flood frequency analysis, and trend analysis) (Karahacane et al., 2020). However, such work yet lacks comprehensive and feasible approaches (Karahacane et al., 2020), thus highlighting the importance of developing a framework for selecting flood events and identifying flood characteristics.

Flooding is a multivariate stochastic phenomenon generally described by variables such as flood peak, flood volume, and flood duration (Mediero et al., 2010). The determination of flood characteristics has not yet been comprehensively studied, although various simplified methods have been used to obtain them in previous studies (Vittal et al., 2015; Jeong et al., 2014; Nadarajah and Shiau 2005). Three main challenges should be considered for the accurate and objective identification of flood characteristics: the selection of suitable flood samples, accurate identification of the start and end dates of flood events, and extension of the recession process to separate flood events from the observed discharge.

The annual maximum series (AMS) approach and the POT method are widely used for flood sampling (S. Solari and Losada, 2012). The POT method, which can capture more information about flood processes extracted from the daily streamflow data and reduce the uncertainty of flood frequency analysis (Lang et al., 1999), has been widely used for flood risk estimation in the past few decades (Durocher et al., 2018; Durocher et al., 2019; Aissia et al., 2012). However, the flood samples obtained using the POT model largely depended on the selection of the threshold. Reliable threshold selection methods, such as the fixed quantile (Jeong et al., 2014; S. Solari and Losada, 2012), mean number of over-threshold events (Brunner et al., 2018, 2019), mean exceedance above the threshold (Davison and Smith, 1990; Lang et al., 1999), and the automatic threshold selection method (Liang et al., 2019; S. Solari and Losada, 2012) can result in a suitable flood sample. Automatic procedures, which determine the threshold automatically according to the degree of fit between the hypothetical distribution and the flood sample, have proven to be effective (Solari et al., 2017; Durocher et al., 2018; S. Solari and Losada, 2012). Durocher et al. (2018) compared the efficiency of different threshold selection methods and indicated that automatic threshold selection based on the goodness-of-fit (GOF) test can obtain a reasonable optimal threshold more objectively.

The determination of the start and end dates of flood events is the key to characterise the flood process. One of the widely used ways to address this issue is the conceptual graphical approach, in which the start date is usually marked by an abrupt increase in the hydrograph, and the end date can be determined by the flattening of the hydrograph's recession limb (Tosunoglu et al., 2020; Y. R. Liu et al., 2020; Sheng Yue, 2000). However, there is a certain degree of subjectivity involved in identifying the start and end dates. To reduce the influence of human decisions, a large number of relevant studies have used a simplified approach (Vittal et al., 2015; Brunner et al., 2019; Mediero et al., 2010). For example, Vittal et al. (2015) suggested that the intersections of the threshold line and flow hydrograph correspond to the start and end points of a flood event. To date, there is no objective method for addressing this problem without human intervention due to the complexity of flooding events.

Furthermore, to determine the flood volume and hydrograph, flood events need to be extracted from the flow hydrograph using baseflow separation methods (Smakhtin 2001; Sujono et al., 2004; Tallaksen 1995). Generally, simplified methods can be used to separate flood hydrographs. For example, the flood volume was simply estimated using a straight line to separate direct runoff from baseflow (Tosunoglu et al., 2020; Aissia et al., 2012; Yue, 2000). In addition, the master recession curve, a graphical method, has been widely used to describe the discharge-storage relationship of watersheds, and some researchers have proposed different functional models for obtaining the master recession curve, such as linear (Sujono et al., 2004), and power function relation (Carlotto and Chaffe 2019). Furthermore, this method allows the extraction of multiple flood events over a long period by extending the recession process (Beven et al., 2011; Lamb and Keith, 1997; Sujono et al., 2004); however, it only finds scarce use in characterising flood events (such as calculating flood volume) due to its complexity.

Both flood sampling and identification of flood characteristics are often performed manually, which is tedious and requires expertise. It also gives rise to large uncertainties in subsequent flood analyses. Moreover, the manual method would be very inefficient when applied to a large amount of data (Solari et al., 2017; Carlotto and Chaffe 2019; Arciniega-Esparza et al., 2017; Durocher et al., 2018). Therefore, it is necessary to develop an automatic generic procedure for selecting flood events and identifying flood characteristics with minimal human intervention that can be applied efficiently to extensive data.

Hence, we developed an automatic generic procedure to objectively select the threshold for flood sampling based on the POT model, and identify flood characteristics according to the master recession curve method, which is a generalised standardised approach that can be applied to all watersheds with diverse characteristics. We also construct a graphical user interface (GUI) for this procedure in MATLAB. We tested and validated the proposed procedure using MOPEX data (Duan et al., 2006). We expect to improve the selection of flood events and the determination of flood characteristics by using our automatic procedure, thereby reducing uncertainties in subsequent flood analyses, such as multivariable flood frequency and flood trend analyses.