Mobile QR Code QR CODE : Journal of the Korean Society of Civil Engineers

  1. 원광대학교 건설환경공학과 석사과정 (Wonkwang University․yangzhipeng325@gmail.com)
  2. 종신회원․교신저자․원광대학교 건설환경공학과 부교수, 공학박사 (Corresponding Author․Wonkwang University․yong_jung@wku.ac.kr)



홍수자료공간확장, 미계측유역, 극한 유량 범위, 일반화
Stream propagation method (SPM), Ungauged watersheds, Extreme flood range, Generalization

1. Introduction

The streamflow data is the substantial requirement for flood control design in hydraulic engineering. Flood disaster events have become the focus of more and more attention of the international community, governments and scientific circles. Flood protection infrastructure is built in small/medium-size watersheds, which require the largest flood records in the region(Tran et al., 2018). However, there are fewer discharge observations in small/medium-size watersheds, and most of the discharge data observations are obtained at hydrologically important locations or downstream (Kim et al., 2019). A number of works have been done on flood research, mainly focusing on flood deduction and forecast; flood management and detection; flood characteristic cause, risk and disaster analysis and so on. In flood design studies, there are also empirical studies using regional records of extreme floods to substantiate flood modeling and other related analyses (e.g., Biondić et al., 2007: Extreme flooding in the Danube Basin, Croatia; Furey and Gupta, 2005; Furey and Gupta, 2007: Data from the Goodwin Creek experimental watershed; Kovacs, 1988: 30 sites in South Africa; Maréchal et al., 1997: European Survey of Extreme Floods; Kim et al., 2019: Regional flood study in Korea). In other studies, China’s Xin’anjiang lumped hydrological model (Zhao and Liu, 1995) also used observational data for flood analysis.

Maximum flood records are limited for flood protection information in many developing countries. Using the method of river space propagation can solve the problem of loss of streamflow data in small/medium-size watersheds. Transfer streamflow data to an ungauged watershed using flow data with similar characteristics near the watershed (McDonnell and Beven, 2014; Merz and Blöschl, 2004; Oudin et al., 2008). For instance, the regression method finds relationships between streamflow data and watershed characteristics and generates flow data for ungauged watersheds (Heřmanovský et al., 2017; Wagener et al., 2004). To address this conundrum, a robust method, namely the streamflow propagation method (SPM, Kim et al., 2021), was proposed to generate ungauged streamflow data in small/medium-size watersheds for use as an improved regionalization method.

The purpose of this study is the generalization of representing the ranges of extreme floods for ungauged watersheds in Korea. This study generates ungauged flow data in watersheds of various scales using the SPM and optimize it in a rainfall-runoff conceptual model. At the same time, for the transmission of streamflow data in small/medium-size watersheds with no streamflow data, we carried out simulation comparisons in three different watersheds to observe the spatial distribution of streamflow in different sizes of watersheds using specific floods on a regional scale. When analyzing specific floods, we use three envelope curves to analyze spatial distribution of streamflow. Based on the obtained spatial distributions we tried to obtain the maximum specific flood extent within different watersheds to help manage small/medium-size watersheds. In addition, this paper explores the factors affecting the spatial characteristics of ungauged watersheds at different scales. The remainder of the paper is as follows: Section 2 provides the methodology; Section 3 explains the study area for selected storm events; Section 4 gives the generation of streamflow data with ranges of regional specific floods and range generalization; and Section 5 summarizes the findings.

2. Methodology

2.1 Streamflow Generation for Ungauged Watersheds

The lumped conceptual model (i.e. Storage Function Model, Kimura, 1961) is used to generate streamflow values in ungauged small/medium-size watersheds using the Streamflow Propagation Method (Kim et al., 2019) for this study. The two independent sets of parameters are distinguished such as physics-based parameters and time-varying (i.e. event-based) initial conditions. Physical parameters highly related to watershed characteristics are fixed for all separated smaller watersheds, while initial conditions are optimized to fit streamflow data at a selected downstream. The essential difference between this method and conventional regionalization methods is the separation of parameters and the concentration time of the simulation (i.e., not for prediction, but for past data generation). In the process of spatial propagation of streamflow data, the fixed parameters related to physical properties minimize the simulated streamflow data variances compared with the measured data. With these given conditions the error rates for each watershed will be minimally controlled for all separated smaller watersheds. For small to medium ungauged watersheds located upstream, streamflow data can be generated while we tried to match measured streamflow data at a selected downstream using optimizations of an event-based parameter with fixed physical parameters. For the SPM, as event-based initial conditions (Rsa: cumulative saturated rainfall, mm/h) was optimized based the generalized other parameters (K and P: constant parameters; and Tl: lag time, h). During the propagation process, we used the observed inflow value of CJD, SJD, and ADD as a criterion for optimization. The criterion of optimization is the Nash – Sutcliffe Efficiency (NSE, Nash and Sutcliffe, 1970).

(1)
${NSE}=$$1-\left[\dfrac{\Sigma_{i=1}^{n}\left(Y_{i}^{obs}-Y_{i}^{s im}\right)^{2}}{\Sigma_{i=1}^{n}\left(Y_{i}^{obs}-Y^{obs mean}\right)^{2}}\right]$

where Yi is the observed and simulated values and Yobs mean is the mean value of the observations.

2.2 Ranges of maximum streamflow

After streamflow data is generated for ungauged small/ medium-size watersheds, the flow distribution of river watersheds of different sizes is observed through the specific flood distribution analysis. The specific flood distribution analysis typically analyzes the streamflow of each rainfall event. In this study, we applied this method to show the possible maximum streamflow ranges of watersheds of different sizes. The range of possible maximum streamflow distribution are expressed by envelope curve equations such as: the Creager envelope (2) (Creager et al., 1945), the Francou-Rodier envelope (3) (Biondić et al., 2007), the Kovacs regional maximum flood (4) (Kovacs, 1988), and the Korean regional maximum flood (KRMF) (5) (Kim et al., 2019) equations. These equations are displayed in Eqs. (2)~(5).

(2)
${q}=0.503{C}(0.3861{A})^{({a A}^{{b}}-1)}$

where q is the specific flood (m³/s·km²); C is the Creager coefficient; A is Area (km²); and a and b are constant variables.

(3)
$\dfrac{Q}{Q_{0}}=\dfrac{A^{1\dfrac{k}{10}}}{A_{o}}$

where Q is maximum discharge (m³ s-1); A is catchment area (km²); K is the Francou-Rodier coefficient; Q0= 106 m³ s-1; and A0 = 108 km²

(4)
${Q}=10^{6}\left(\dfrac{{A}}{10^{8}}\right)^{1-0.1{K}}$

where Q is the streamflow (m³/sec); A is the area (km²); and K is the coefficient representing the characteristics of watersheds.

(5)
${q}=11.25\dfrac{1}{{A}^{0.25}}$

In the formula, q is the specific flood (m³/s·km²), and A is the area (km²).

2.3 Spatial Characteristics and Generalization

According to the specific flood distribution analysis, the Creager envelope equation, the Francou-Rodier envelope, the Kovacs regional maximum flood equation and the KRMF equation analysis compare time-flood plots with area-watershed plots for the CJD, SJD, and ADD watersheds. After individual watershed analysis, CJD, SJD, and ADD information were added together for generalization. For generalization, an analysis of variance (ANOVA) test was used to verify its results to prove our conjecture about possible generalization with all information together. We use the RMSE (Root Mean Square Error) to calculate the RMSE-maximum and RMSE-minimum values for a trend line of maximum and minimum of specific flood using three envelop curves. The RMSE equation looks like this:

(6)
$R M S E(a,\: P)=\sqrt{\dfrac{\Sigma_{i=1}^{n}\left(p i-a_{i}\right)^{2}}{n}}$

where a is actual target, p is predicted target, n is select the amount of data.

3. Study Area

The topography of watersheds varies and consists of different terrain types. South Korea is a country with numerous mountains and with planes. For this study we selected three different watersheds as shown in Fig. 1 (i.e., Chungju Dam, Seomjin Dam, and Andong Dam watersheds). The Chungju Dam (CJD) watershed located at the upper reaches of the Southern Han River. The CJD watershed covers an area of about 6648 km² and percent of mountain is about 41.43 %. The slope of 55.1 % of CJD watershed is 40 %, and 8.3 % of the watershed has a slope of more than 80 %. The CJD watershed is divided into 35 small watersheds. To determine the spatial characteristics of flood in a large watershed, we need spatially distributed streamflow data. The observations from 12 streamflow stations are not enough to reflect the spatial characteristics of flood in the area. We used three different water level gauges (i.e., Youngwol (YW), Youngwol 1 (YW1), Youngchun (YC)) as verification points. For the propagation process, the CJD inflow is the main criteria for the propagation process.

The size of Seomjin Dam (SJD) watershed is 4,896.5 km² including 38.65 % mountainous areas. The SJD watershed is divided into 34 small watersheds. We used four different water level gauges (i.e., H_ Jwapo Bridge (JP), S_ Im Woon-gyo (IW), Y- Seongsu (SS)) as verification points. The SJD inflow is the main information we want to match as the CJD watershed.

The Andong Dam (ADD) watershed is located in the upstream of Nakdong river with a drainage area (23,384 km²). The mountain coverage is about 67.84 % which is larger than others in this study. In this study, the ADD watershed was divided into 55 small watersheds without streamflow information. Three water level gauges were selected for the validation purpose (i.e.: S_Yangsam Bridge (YS), H_ Docheongyo Bridge (DC), H_ Punghori (PH)). Based on the data availability and the streamflow representativeness, the SPM verification points were chosen. The separated smaller watersheds are based on the DEM analysis at COSFIM simulator provide by K-water. This study selected the 19 largest precipitation events from 2001 to 2019.

Fig. 1. Chungju Dam, Seomjin Dam, and Andong Dam watersheds. Diagrams of the Relationship between Selected Water Gauges and Dams (Schematics of Watersheds from the Korean Research File:Water Environment Information System, 2022)
../../Resources/KSCE/Ksce.2022.42.5.0627/fig1.png

4. Results and discussion

4.1 Generation of Streamflow Data

Fig. 2(a) is a comprehensive graph of the comparison between observed and simulated values using the streamflow propagation method (SPM) for the CJD,SJD and ADD watersheds. We obtained the observed and simulated values of heavy streamflow events from 2001 to 2019 from the observation stations identified in the three major watersheds (described in Chapter 2). Fig. 2(b) is a graph of the peak streamflow data collected by various observation stations in the CJD, SJD, and ADD watersheds. Fig. 2(c) is box plots of the optimized and verified NSE values of observation stations after SPM process in the three major watersheds. In Fig. 2(c), four box plots are a group according to the order of the CJD, SJD, and the ADD watershed. The first box plots, (i.e., Dark color) in each group are the optimization points, and the remaining points are data propagated points.

In Fig. 2(a), the data for all watersheds generally show an upward trend in both simulated and observed values along a linear trend. The streamflow data of the CJD (blue), ADD (green), and SJD (red) watersheds are distributed in the range of CJD<14000 m³/s, SJD<6000 m³/s and ADD <3000 m³/s, respectively. These various distributions are based on the different size and weather coalitions may produce these differences.

Fig. 2(b) is a graph of the peak streamflow data collected by each observation station in the CJD, SJD and ADD watersheds. Through the distribution of each peak point in the figure, the distribution range of the peak flow data of each watershed can be known. The distribution ranges of the CJD station optimization point and the other three verification points are less than 10000 m³/s. The distribution range of peak data in the SJD watershed is <2500 m³/s. The distribution range of peak data in the ADD watershed is <6000 m³/s. At the same time, the distribution of the peak data of the verification points in each watershed is roughly consistent with the peak data distribution of the optimization points. Fig. 2(c) shows that in different areas within different watersheds, the data of the optimization points can reflect the general situation of the data of the verification points. By comparing the peak data of each watershed, we found that the size and distribution of the peak data in each watershed also have an impact on the size of the NSE value of the optimization and verification points in each watershed. The finalized mean NSE values of the CJD, SJD, and ADD watersheds were 0.85,0.83, and 0.84, respectively. By the comparison of the box plots of the verification points, the SJD watershed is the most stable compared with other watersheds based on the variances size. Due to the different geographical locations and area sizes of the three watersheds, the NSE values at the verification points for the SPM in each watershed were fluctuated.

Fig. 2. (a) A Composite Plot of the Observed-Simulated Values for the CJD, SJD and ADD Watersheds; (b) A Graph of the Maximum Peak Data of Streamflow Data Collected by Various Observation Stations in the CJD, SJD, and ADD Watersheds from 2001 to 2019; (c) NSEs of Each Observation Station
../../Resources/KSCE/Ksce.2022.42.5.0627/fig2.png

4.2 Ranges of Regional Specific Floods

The envelope curve method is used to establish the regional relationship between the ranges of maximum streamflow data and the size of watersheds for ungauged watersheds. We assume that the hydrological climate of the study area is stable, and the size of the watershed area is the main control feature for the scale of the streamflow data. Figs. 3~5 shows the maximum regional streamflow in the CJD, SJD, and ADD watersheds. The maximum streamflow data generated by the SPM is ranged by various regional maximum flood equations. Figs. 3~5(a) show the applications of the Creager envelope curve. The two Creager curves show the range of a maximum streamflow in each watershed. We selected optimal envelope curves based on RMSE between propagated values and envelope curves. As an example, the parameters of upper and lower Creager curve equations for CJD are C = 12, a = 0.936, and b = -0.025; and C =1, a = 1.18, and b = −0. 007. As shown in Figs. 3~5(a), there are larger specific variances in smaller watersheds, because the smaller watersheds response sensitively on rainfall amount and terrain conditions. Figs. 3~5(b) show the applications of Kovacs area-specific flood equation. The given maximum specific floods of CJD, SJD, and ADD watersheds are in the Kovacs coefficient between K = 5.4 and K = 2.4. The maximum range of the specific flood can be represented in K=5.4 – 5.2 for three dams and the minimum ranges are 2.4 – 3.7. Figs. 3~5(c) show the applications of the Francou-Rodier envelope curve. The Francou-Rodier curve also show the maximum/minimum formula for specific floods in each watershed. In Figs. 3~5(c), it can also be proved that there is a larger specific variance in smaller watersheds. Figs. 3~5(d) are drawn for the Korean regional maximum flood (KRMF) equation. The maximum of specific floods based on KRMF do not represent the real maximum specific flood values produced by SPM. Therefore, the equation needs to be modified to represent the maximum of specific floods in Korea (Kim et al., 2021).

The RMSE-max (i.e. optimization result of upper envelop curve) and RMSE-min (i.e. optimization result of lower envelop curve) in the three watersheds are shown in Table 1. From Table 1, the optimal RMSEs of the RMSE-max from the Creager curve equations for SJD and ADD were 2.07 and1.87. The RMSE-max for CJD using Creager envelop curve has slightly higher value. The RMSE values of the RMSE-max obtained by the Kovacs curve equations for CJD, SJD, and ADD have 1.65, 2.22, and 2.21, respectively. The envelop curves of Francou-Rodier have higher RMSE vales for all watersheds. The optimization results of the RMSE show that the Creager curve equation in the three watersheds fit the given envelope curves well based on the averaged RMSE-max and -min. Larger RMSE-max were obtained at the Francou-Rodier envelope equations (1.61, 2.37 and 2.33). In the case of the RMSE-max obtained from the Kovacs and Francou-Rodier envelope curves, the SJD and ADD watersheds have relatively larger RMSE compared to the CJD because the SJD and ADD watersheds have more mountainous areas. It also shows that the three envelope curves we selected also represent the RMSE-min well.

Fig. 3. The Specific Flood with the Maximum Flood Equations for CJD Watershed
../../Resources/KSCE/Ksce.2022.42.5.0627/fig3.png
Fig. 4. The Specific Flood with the Maximum Flood Equations for SJD Watershed
../../Resources/KSCE/Ksce.2022.42.5.0627/fig4.png
Fig. 5. The Specific Flood with the Maximum Flood Equations for ADD Watershed
../../Resources/KSCE/Ksce.2022.42.5.0627/fig5.png
Table 1. The Maximum and Minimum RMSE Values Calculated by the Three Methods for the CJD, SJD, and ADD Watersheds

Unit: ㎥/s·㎢

CJD

SJD

ADD

Creager

Kovacs

Francou-Rodier

Creager

Kovacs

Francou-Rodier

Creager

Kovacs

Francou-Rodier

RMSE-max

1.82

1.65

1.61

2.07

2.22

2.37

1.87

2.21

2.33

RMSE-min

0.3

0.84

0.93

0.87

0.83

1.03

0.98

0.97

0.88

Average

1.28

1.29

1.31

1.57

1.65

1.81

1.51

1.74

1.79

4.3 Generalization

The dimensions of the CJD, SJD, and ADD watersheds are different as 6648 km², 786 km², and 1576 km², respectively. Based on the distribution of specific floods shown in Fig. 6, the distribution of specific flood data in the range of 100-1000 km² is relatively concentrated with the largest specific flood data. In the range of 1000-10000 km², the specific flood data shows a decreasing state with the increase of the watershed size. Considering the characteristics of watersheds, which refers to area, shape, slope, longitudinal section and topography have a possible impact on the size of specific flood. The higher and more stepped terrain has the faster water flow which increased flood size (Gitau and Indrajeet, 2010). The mountain percentages in the CJD, SJD, and ADD watersheds were 41.43 %, 38.65 %, and 67.84 %, respectively. The ADD watershed with the highest proportion of mountains (73.4 %) has the higher specific flood in the rage of 400-1000 km². Greater streamflow, which makes specific floods larger, exacerbating flooding. Therefore, the characteristics of topographical conditions produces different scales of specific floods.

Fig. 6. Applications of Envelop Curves for Integrated Streamflow Data from CJD, SJD, and ADD Watersheds
../../Resources/KSCE/Ksce.2022.42.5.0627/fig6.png

To generalize the ranges of maximum of specific flood, we sum all data for the three watersheds and try to obtain an envelope curve. Before adding all data, ANOVA test was performed the one-way balanced to evaluate different specific flood data (SF) across the three watersheds as variables. The differences in specific floods between different watersheds may result from errors. With a total of 3 subjects in our study, the obtained F ratio was 8.9031 with a p-value of 0.0003. According to the ANOVA test result, the combined specific flood generated by SPM was not significantly different. Although the scales of the SJD and ADD watersheds were small, the specific flood data in their watersheds did not fluctuate significantly.

Figs. 6(a), (b) and (c) are the aggregation of specific flood with application of various envelop curves for the CJD, SJD and ADD watersheds. Fig. 6(d) is to analyze the variance specific floods for divided sizes of regions. The upper and lower Creager curve equations are optimized at a = 141.78, b =−0.403, C = 70 and a = 5.019, b =−0.397, C = 2. The larger watersheds have little impacts on rainfall sizes, compared to smaller watersheds because within a smaller watershed the response of maximum streamflow is lager with a faster peak time. The Kovacs coefficients were ranged between 2.4 and 5.4 which is same to the previously analyzed as separated regions at the section of 4.2. The Francou-Rodier envelop curve has a larger averaged RMSE (2.22) of RMSE-max and -min compared to other methods to represent the degree of agreement.

The generalized equations for the maximum specific flood extent are as follows:

(7)

Maximum:y = 2.89 × 10$^{-5}$$x^{2}$ - 2.399 × 10$^{-5}x$

RMSE-max:1.04

(8)

Minimum:y = 1.34 × 10$^{-5}$$x^{2}$ - 3.854 × 10$^{-5}x$

RMSE-min:1.01

where x is the size of the watershed.

The Creager and Kovacs envelop curves have averaged RMSEs at 1.45 and 1.51, respectively. Beside of these formulas, the comprehensive generalized formula has lower RMSE (1.03 <1.45), which is smaller than the value calculated by the Creager and Kovacs. This shows that the proposed comprehensive generalized formula can accurately generate specific flood ranges for various scales of ungauged watersheds. Table 2 and Fig. 7 show the results of the generalized formula for range representation.

Fig. 7. Generalized Equations for the Maximum Specific Flood Extent
../../Resources/KSCE/Ksce.2022.42.5.0627/fig7.png
Table 2. The Maximum and Minimum RMSE Values Calculated by the Three Methods for the Specific Flood for Three Watersheds

Unit: ㎥/s·㎢

Creager

Kovacs

Francou-Rodier

Generalization

RMSE-max

1.92

1.96

2.58

1.04

RMSE-min

0.75

0.87

1.81

1.01

Average

1.45

1.51

2.22

1.03

5. Conclusion

Streamflow information is an important part of managing water resources and flood control. Different scales of watersheds have different effects on the streamflow data simulations. However, due to the lack of streamflow data, it is difficult to provide sufficient information on spatial distribution of streamflow. The streamflow propagation method (SPM) can provide spatially distributed streamflow data for ungauged watersheds.

In the process of analyzing the relationship between specific floods and watershed sizes, the topography of watersheds potentially has an influence on the distribution pattern of specific floods. If the terrain in a watershed has higher slopes, the streamflow data in the middle and lower reaches of a watershed is relatively large. For this study, watersheds with mountain area over 58 % (i.e., ADD) has more fluctuated specific flood distributions.

The three integrated dam watersheds (i.e., CJD, SJD and ADD) also use the Creager, Kovacs, and the Francou-Rodier envelope curve formula to get the representative equation for maximum specific flood for various sizes of watersheds. Based on the RMSE comparison, the generalized formula is recommended to represent the range of specific floods for various sizes of watersheds.

Acknowledgement

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (NRF-2022R1A2C1092215). Specially thanks to the K-water for the model support (COSFIM).

References

1 
Biondić D., Barbalić D., Petraš J. (2007). "Creager and Francou-Rodier envelope curves for extreme floods in the Danube River basin in Croatia.", In Predictions in Ungauged Basins: PUB Kick-off (Proceedings of the PUB Kick-off meeting held in Brasilia, 20–22 November 2002). IAHS Publ, Brasilia, Brazil, Vol. 309, pp. 221-228URL
2 
Creager W. P., Hinds J., Justin J. D. (1945). "Engineering for Dams", By William P. Creager, Joel D. Justin and Julian Hinds, Etc. John Wiley & Sons, New Jersey, USA.Google Search
3 
Furey P. R., Gupta V. K. (2005). "Effects of excess rainfall on the temporal variability of observed peak-discharge power laws.", Advances in Water Resources, Vol. 28, No. 11, pp. 1240-1253DOI
4 
Furey P. R., Gupta V. K. (2007). "Diagnosing peak-discharge power laws observed in rainfall-runoff events in Goodwin Creek experimental watershed.", Advances in Water Resources, Vol. 30, No. 11, pp. 2387-2399DOI
5 
Gitau M. W., Indrajeet C. (2010). "Regionalization of SWAT model parameters for use in ungauged watersheds.", Water, Vol. 2, No. 4, pp. 849-871DOI
6 
Heřmanovský M., Havlíček V., Hanel M., Pech P. (2017). "Regionalization of runoff models derived by genetic programming.", Journal of Hydrology, Vol. 547, pp. 544-556DOI
7 
Kim N. W., Jung Y., Lee J. E. (2019). "Spatial propagation of streamflow data in ungauged watersheds using a lumped conceptual model.", Journal of Water and Climate Change, Vol. 10, No. 1, pp. 89-101DOI
8 
Kim N. W., Kim K. H., Jung Y. (2021). "Spatial recognition of regional maximum floods in ungauged watersheds and investigations of the influence of rainfall.", Atmosphere, Vol. 12, No. 7, pp. 800DOI
9 
Kimura T. (1961). "The flood runoff analysis method by the storage function model", The Public Works Research Institute, Ministry of Construction, Japan.Google Search
10 
Kovacs Z. (1988). "Regional maximum flood peaks in Southern Africa", Department of Water Affairs, Pretoria, Technical Report TR137.Google Search
11 
Maréchal J. C., Ladouche B., Dörfliger N. (1997). "Karst flash flood in a Mediterranean karst, the case of Fontaine de Nîmes", BRGM, Water Department, 1039 rue de Pinville, 34 000 Montpellier, France.Google Search
12 
McDonnell J. J., Beven K. (2014). "Debates-The future of hydrological sciences: A (common) path forward? A call to action aimed at understanding velocities, celerities and residence time distributions of the headwater hydrograph.", Water Resources Research, Vol. 50, No. 6, pp. 5342-5350DOI
13 
Merz R., Blöschl G. (2004). "Regionalisation of catchment model parameters.", Journal of Hydrology, Vol. 287, No. 1-4, pp. 95-123DOI
14 
Nash J. E., Sutcliffe J. V. (1970). "River flow forecasting through conceptual models part I-A discussion of principles.", Journal of Hydrology, Vol. 10, No. 3, pp. 282-290DOI
15 
Oudin L., Andréassian V., Perrin C., Michel C., Le Moine N. (2008). "Spatial proximity, physical similarity, regression and ungaged catchments: A comparison of regionalization approaches based on 913 French catchments.", Water Resources Research, Vol. 44, No. 3DOI
16 
Tran Q. Q., De Niel J., Willems P. (2018). "Spatially distributed conceptual hydrological model building: A generic top‐down approach starting from lumped models.", Water Resources Research, Vol. 54, No. 10, pp. 8064-8085DOI
17 
Wagener T., Wheater H., Gupta H. V. (2004). "Rainfall-streamflow modelling in gauged and ungauged catchments.", World Scientific, Vol. 26, pp. 145-168Google Search
18 
"Water Environment Information System", Available at: https://water.nier.go.kr/ (Accessed: June 16, 2022).URL
19 
Zhao R. J., Liu X. R. (1995). "The Xinanjiang model, Computer Models of Watershed Hydrology", Water Resource Publications, LLC, Colorado, USA., pp. 215-232Google Search