| Home | E-Submission | Sitemap | Contact Us |  
Environ Eng Res > Volume 21(2); 2016 > Article
Yin and Seo: Analysis of optimum grid determination of water quality model with 3-D hydrodynamic model using environmental fluid dynamics code (EFDC)

Abstract

This study analyzes guidelines to select optimum number of grids to represent behavior of a given water system appropriately. The EFDC model was chosen as a 3-D hydrodynamic and water quality model and salt was chosen as a surrogate variable of pollutant. The model is applied to an artificial canal that receives salt water from coastal area and fresh water from a river from respective gate according to previously developed gate operation rule. Grids are subdivided in vertical and horizontal (longitudinal) directions, respectively until no significant changes are found in salinity concentrations. The optimum grid size was determined by comparing errors in average salt concentrations between a test grid systems against the most complicated grid system. MSE (mean squared error) and MAE (mean absolute error) are used to compare errors. The CFL (Courant-Friedrichs-Lewy) number was used to determine the optimum number of grid systems for the study site though it can be used when explicit numerical method is applied only. This study suggests errors seem acceptable when both MSE and MAE are less than unity approximately.

1. Introduction

While a water quality model with higher complexity may provide more detailed or accurate prediction results, it is prone to provide increased uncertainty due to greater chance of error propagation [1]. Also, inappropriate integration step size in either temporal or spatial scale may result in errors due to numerical dispersion of numerical methods [2] and thus can make model results less reliable. Seo and Kim [3] and Seo et al. [4] reported that it required few days to few weeks for annual water quality modelling of the Nakdong River in Korea using combination of 3-dimensional EFDC hydrodynamics model [5] and WASP water quality model [6, 7] and this could be prohibitive to use for intensive scenario analysis or statistical analysis. While many researchers reported that a smaller grid size is preferred for more accurate prediction for their modelling applications [811], it seems many other researchers arbitrarily select grid resolutions without sufficient consideration on grid size determinations [12, 13]. Seo [14] compared a different number of cells when using a combination of EFDC and WASP model to calibrate the thermal stratification of Lake Yongdam, Korea. Kim [3] confirmed that detailed resolution was necessary by increasing number of vertical layers to accurately model thermal stratification of Lake Yongdam, Korea.
There have been many researches on effect of grid size especially for hydrological study using GIS data [1517]. However there have not been many practical studies on how to select the optimum grid size for water quality modelling especially when used with hydrodynamics model. In this study, methods to determine the optimum number of grid size for accurate and efficient water quality modelling applications was analysed and discussed. Salinity was chosen as a surrogate of water quality contaminant and an artificial canal was chosen as a test study site.

2. Materials and Methods

2.1. Study Area

For this study an artificial canal constructed recently in Korea chosen as a test site. Data shown here may not be accurate and thus can be regarded as data for reference purpose due to limited availability of information of the construction. The Ara Canal was constructed to prevent flood damages in the Goolpocheon River basin and also to promote economic development of the area by reducing logistic costs. The canal is connecting the Han River that passes through Seoul City and the western coastal area of Korea (Fig. 1). There are two gates at both ends of the canal. The canal gate in the Han River side is located near Haengju Bridge at a downstream area of the river. The other gate in the ocean side is located in Incheon City facing western costal area of Korea. The canal is 18 km long overall while the navigation channel length is 15.6 km. The width of the navigation channel is 80 m and the average water depth is 6.3 m. Table 1 shows the general information of the Ara Canal [18].
Mixing of freshwater from the Han River and salt water from the coastal area with frequent gate operation for navigations may create complicated salinity distribution due to density differences between salt water and fresh water.
This hydrodynamic behaviour inside the canal will affect water quality kinetics including the growth of phytoplankton and dissolved oxygen depletion. To consider both water quantity and water quality issues during gate operations, it is suggested to use an advanced hydrodynamic and water quality model.

2.2. The EFDC (Environmental Fluid Dynamics Code) Model

The Environmental Fluid Dynamics Code (here after EFDC) model and has been used widely for hydrodynamic and water quality modeling in rivers, lakes, estuaries and coastal systems [5]. Grid development of the EFDC model for the study site is as shown in Fig. 2. Wool applied a combination of hydrodynamic module of EFDC and the WASP water quality model for modeling of the Neuse River including its estuarine area [7]. Seo [4, 19] and Kim [4] also used the EFDC-WASP combination to model major rivers in Korea. Jung compared a 2-dimensional SMS model [20] and 3-dimensional EFDC model to predict salt intrusion in the Sumjin River, Korea and reported that the EFDC model showed more accurate results [21]. Noh analyzed the effect of flushing discharge on salt water intrusion in the Sumjin River estuarine area using EFDC model [22]. The detailed theoretical background and explanation of the model can be found in EFDC manuals [23].

2.3. Input Development of the EFDC Model

2.3.1. Initial grid

Morphological information of the area was obtained from their environmental impact assessment report of the study area [18]. Information about in-stream structures and the grid system for HEC-RAS [24] model were used to develop grid systems for EFDC application in this study. Fig. 3 compares water level versus the volume curve of the canal from the HEC-RAS model of the previous study and the EFDC model in this study. Two model results are almost identical and this shows that the EFDC grid in this study reflects the physical characteristics of the canal appropriately.

2.3.2. Boundary conditions

Inflow for the west gate of the canal is influenced greatly by the tidal effect of the ocean area as shown in Fig. 4. Therefore, there must be an optimum gate operation method to maintain the water level inside the canal. As shown in Table 2, several gate operation scenarios are evaluated and Scenario 2 was chosen as the boundary condition for this study. Fig. 5 shows the flow pattern in the West Gate and the Han River Gate of the canal for scenario 2.

2.4. Grid Size Scenarios

The trial and error method was used to determine the optimum grid system. Since there is no field measurement available for this analysis, most complicated grid cases were assumed to be the ideal one. The maximum number of grids was determined until no further changes in salt concentrations for selected locations in the channel in vertical and horizontal direction, respectively. The 3 different sites in the canal were selected for the optimum grid size analysis; 1) West Sea Gate area, 2) center of the canal way and 3) The Han River Gate area. The average salt concentrations in the upper, middle and lower layers at each station were compared when vertical layers are divided into 3 groups. Considering the above results, the grid systems to be tested are developed as shown in Table 3. The number of lateral cells was fixed to 3 cells for all cases for the sake of simplicity. The vertical grid was divided from 3 to 30 layers for 216 horizontal layers. The thickness of each layer varied from 0.2 m to 2.1 m. The horizontal grids for the channel were subdivided between 18 and 504 for 15 vertical layers in the channel. Also, the length of the horizontal direction for each cell varies from 35 m to 1000 m.

2.5. Error Analysis Method

MSE (Mean Squared Error) and MAE (Mean Absolute Error) statistics were used for error analysis can be defined as in Eq. (1) and Eq. (2), respectively.
(1)
MSE=i=1n(Si-Xi)2n
(2)
MAE=i=1n|Si-Xi|n
where,
  • n = 60

  • Si = standard value at ith day (the most complicated grid system, 504 horizontal reaches × 30 vertical layers)

  • Xi= ith day value for other grid system

2.6. The CFL Conditions

Mathematically, the Courant–Friedrichs–Lewy condition (here after CFL condition) [25] can be described as a necessary condition for convergence while solving certain partial differential equations numerically by the method of finite differences. The CFL condition requires that the time step must be less than a certain time in many explicit time dependent computer simulations, in order to avoid incorrect results. Heuristically, this means that, if a wave is propagating across a discrete spatial grid at discrete time steps of equal length, then this length must be less than the time for the wave to travel to adjacent grid points [26]. Therefore, if the grid length decreases, the time step limit needs to be reduced accordingly. For a one-dimensional case, the CFL condition, or Courant number, C can be defined as in Eq. (3).
(3)
C=uxΔtΔxCmax
where,
  • ux = the velocity in x-direction (whose dimension is length/time)

  • Δt = the time step (whose dimension is time)

  • Δx = the length interval (whose dimension is length).

However, the CFL condition is only a necessary condition and may not be sufficient for the convergence of the finite-difference approximation of a given numerical problem. EFDC model uses explicit numerical method for horizontal calculation and implicit method for vertical calculation. Therefore, CFL method can only be good for horizontal case.

3. Results and Discussion

Fig. 6 and Fig. 7 show the 60 days simulation results for three selected points in the canal for different vertical and horizontal grid systems compared in this study. As shown in the figure, it seems that average salt concentrations in each layer tend to converge to a fixed value as the number of cells increases.
Fig. 8 shows that differences in salt concentration are negligible when 30 vertical layers and 504 horizontal divisions were used in this study, respectively. By visual inspection, it seems there must be at least 15 vertical cells and 360 horizontal cells to declare that errors seem negligible.
Fig. 9 shows the calculated, 60 day average vertical salt concentration at a location near the West Gate of the Ara Canal. As number of vertical layers increases, the concentration difference between the top and bottom layers increases. This indicates that the optimal number of vertical layers is essential for the accurate prediction of salinity in the study area. Fig. 10a shows changes in average salinity when vertical layers are grouped to three parts as shown in Fig. 9. As shown in this figure, the average salinity in each layer approaches to a certain point. For the bottom layer, the average salinity change seems to be stabilized when the number of vertical layers is greater than 6. However, the average concentrations in middle and upper layers decrease as the number of layers increases. This condition is caused by differences in vertical transport due to segmentation while in each cell can be regarded as completely mixed. This also indicates the importance of optimum number of grid. Fig. 10b and Fig. 10c show changes in average salinity for three vertical layers at the Canal Way and at the Han Gate locations, respectively. These figures also show similar patterns as seen in Fig. 10a.
Fig. 11a, Fig. 11b and Fig. 11c shows the average salinity concentration for different number of horizontal cells at the West Gate, in the middle of the canal way and at the Han Gate, respectively. Average concentrations in three different vertical layer groups tend to be stabilized as number of horizontal grid increases.

3.1. Error Analysis

Table 4 and Table 5 show MAE and MSE values from the most complicated grid system (30 vertical × 540 horizontal) in three different locations for different vertical and horizontal grid systems, respectively. These results show when vertical layers are 15 or more, average values of both MAE and MSE become less than 1.0. For the horizontal case, when there are more than 216 cells MAE and MSE become less than 1.0. In this simple analysis, it is suggested that 15 vertical layers and 216 horizontal cells can be regarded as optimum numbers in each direction.
Fig. 12a and Fig. 12b show distributions of average salt concentration difference groups for different vertical layer and horizontal cells cases, respectively. These figures show that there must be at least 15 vertical layers and 72 horizontal cells to have more than 90% cases have average vertical concentration differences less than 2 ppt.

3.2. The CFL Condition Analysis

The CFL (Courant–Friedrichs–Lewy) conditions can be important in determining the optimum horizontal grid system for the study site. Though, theoretical maximum for the CFL number is 1.0, it is desirable to be less than unity. Fig. 13 shows CFL numbers estimated from three different locations in the study site. If 0.5 can be arbitrarily chosen as a limiting value for a CFL number, 144 horizontal cells can be chosen as an optimum horizontal number of cells for the study site. However, CFL number only can be applied for horizontal scale for EFDC model where explicit numerical method was used. Since EFDC model use implicit scheme for solving vertical direction, CFL number cannot be used for vertical analysis.

3.3. Determination of the Optimum Grid System for the Ara Canal

Fig. 8 shows that differences in salt concentration are negligible when 30 vertical layers and 504 horizontal divisions were used. There must be at least 15 vertical cells and 360 horizontal cells to declare that errors seem negligible. In Table 4 and Table 5, 15 vertical and 216 horizontal cells were chosen as the optimum to obtain MAE and MSE values less than 1.0. From CFL condition analysis, 144 horizontal cells were suggested as the optimum number of the horizontal grid. This study suggests that the Ara Canal has to be divided into at least 15 vertical and 144 horizontal cells to avoid significant salinity estimation error as shown in Fig. 14.

4. Conclusions

The method to determine optimum number of grid for application of the 3-D hydrodynamic model was discussed and suggested by using the Ara Canal as a test site. Salinity was chosen as a surrogate of water quality contaminant and an artificial canal was chosen as a test study site Salt concentration estimations were compared in different grid systems against the most complicated grid system. Since CFL index can only be valid for explicit numerical method, it can only be used as necessary conditions where it can be applied. MSE and MAE values seem to be appropriate to use as indices for selecting optimum grid size for the tested system. A value of 1.0 is suggested as the optimum value in the selection. This study suggests it is appropriate to subdivide at least 15 layers for 6.3 meter deep water and 144 segments in the horizontal direction for an 18 km long channel for the accurate prediction of salinity distribution of the Ara Canal, the case site, when both MSE and MAE are near to 1.0. These results will also be helpful evaluating effect of instream water quality management alternatives including aeration.

Acknowledgments

This Research was supported by a Grant (12-TI-C02) from Advanced Water Management Research Program funded by Ministry of Land, Infrastructure and Transport of Korean Government.

References

1. Canale RP, Seo DPerformance, reliability and uncertainty of total phosphorus models for lakes 2. Stochastic analyses. Water Res. 1996;30:95–102.
crossref

2. ChapraSCSurface water-quality modeling. New York: McGraw-Hill; 1997.


3. Kim SJ, Seo D, Ahn KHEstimation of Proper EFDC Parameters to Improve the Reproductability of Thermal Stratification in Korea Reservoir (in Korean). J Korean Water Resour Assoc. 2011;44:741–752.
crossref pdf

4. Seo D, Kim MA, Ahn JHPrediction of Chlorophyll-a changes due to weir constructions in the Nakdong River using EFDC-WASP modelling. Environ Eng Res. 2012;17:95–102.
crossref pdf

5. Hamrick JMA Three-Dimensional Environmental Fluid Dynamics Computer Code; Theoretical and Computational Aspects The College of William and Mary, Virginia Institute of Marine Science Special Report. Rep Va Inst Mar Sci. 1992;


6. Wool TA, Ambrose RB, Martin JL, Comer EAThe Water Quality Analysis Simulation Program, WASP6. Part A: Model Documentation. Athens: Center for Exposure Assessment Modelling; US EPA. 2001.


7. Wool TA, Davie SR, Rodriguez HNDevelopment of three-dimensional hydrodynamic and water quality models to support total maximum daily load decision process for the Neuse River Estuary, North Carolina. J Water Resour Plann Manage. 2003;129:295–306.
crossref

8. Jeon JH, Chung SWA Sensitivity Analysis on Numerical Grid Size of a Three-Dimensional Hydrodynamic and Water Quality Model (EFDC) for the Saemangeum Reservoir (in Korean). J Korean Soc Water Environ. 2012;28:26–37.


9. Kim SJ, Seo D, Ahn KHEstimation of Proper EFDC Parameters to Improve the Reproductability of Thermal Stratification in Korea Reservoir (in Korean). J Korea Water Resour Assoc. 2011;44:741–752.
crossref pdf

10. Miyakoda K, Strickler RF, Nappo CJ, Baker PL, Hembree GDThe Effect of horizontal Grid Resolution in an Atmospheric Circulation Model. J Atmos Sci. 1971;28:481–499.
crossref

11. Molnar DK, Julien PYGrid-Size Effects on Surface Runoff Modeling. J Hydrol Eng. 2000;5:8–16.
crossref

12. Pathapati SS, Sansalone JJCFD Modeling of a Storm-Water Hydrodynamic Separator. Environ Eng Res. 2009;135:191–202.
crossref

13. Rengstorfa AM, Mohnb C, Brownc C, Wiszd MS, Grehan AJPredicting the distribution of deep-sea vulnerable marine ecosystems using high-resolution data: Considerations and novel approaches. Deep Sea Research Part I: Oceanographic Research Papers. 2014;93:72–82.
crossref

14. Seo D, Sigdel R, Kwon KH, Lee YS3-D Hydrodynamic Modelling of Yongdam Lake, Korea using EFDC. Desalin Water Treat. 2010;19:42–48.
crossref

15. Chang CL, Chao YCEffects of spatial data resolution on runoff predictions by the BASINS model. Int J Environ Sci Technol. 2014;11:1563–1570.
crossref

16. Eguen M, Aguilar C, Herrero J, Millares A, Polo MJOn the influence of cell size in physically-based distributed hydro-logical modelling to assess extreme values in water resource planning. Nat Hazard Earth Sys. 2012;12:1573–1582.
crossref

17. Vieux BE, Needham SNonpoint-Pollution Model Sensitivity to Grid-Cell Size. J Water Resour Plann Manage. 1993;119:141–157.
crossref

18. K-Water. Gyung-In Canal Project (Navigation Part), Environmental Impact Assessment Report II (in Korean). K-Water, Korea Water Resour. Corporation; 2009.


19. Seo D, Kim MAApplication of EFDC and WASP7 in Series for Water Quality Modelling of the Yongdam Lake, Korea. J Korea Water Resour Assoc. 2011;44:439–447.
crossref pdf

20. Aquaveo. SMS 12.1 - The Complete Surface-water Solution [Internet]. Aquaveo; Available from: http://www.aquaveo.com/software/sms-surface-water-modeling-system-introduction


21. Jung ST, Noh JW, Huh YTAnalysis of Saltwater Intrusion Using 2-D & 3-D Numerical Model in the Seomjin River Estuary. In : The Annual Conference of Korea Water Resources Association; 21 May 2009; p. 785–790.


22. Noh JW, Lee JY, Shin JKAnalysis of Saltwater Intrusion by Flushing Discharge in the Seomjin River Estuary (in Korean). J Environ Impact Assess. 2011;20:325–335.


23. Hamrick JMUser’s manual for the environmental fluid dynamics computer code, Special report in applied marine science and ocean engineering. College of William and Mary, Virginia Institute of Marine Science; 1996. p. 336


24. US Army Corps of Engineers. HEC-RAS River Analysis System, Hydraulic Reference Manual, Version 4.1. Hydraulic Engineering Center; 2011.


25. Courant R, Friedrichs K, Lewy HOn the partial difference equations of mathematical physics. IBM J Res Dev. 1967;11:215–234.
crossref

26. Wikipedia. Courant–Friedrichs–Lewy condition [Internet]. Wikipedia; Available from: https://en.wikipedia.org/wi-ki/Courant%E2%80%93Friedrichs-%E2%80%93Lewy_con-dition


27. Hamrick JMLinking Hydrodynamic and Biogeochemical Transport Models for Estuarine and Coastal Waters. Estuarine and Coastal modelling. In : The 3rd International Conference of American Society of Civil Engineers; 1994; New York. p. 591–608.


28. Johnson BH, Kim KW, Health RE, Hsieh BB, Butler HLValidation of Three Dimensional Hydrodynamic Model of Chesapeake Bay. J Hydraulic Engineering. 1993;119:2–20.
crossref

29. Kim JK, Kwak GI, Jeong JHThree-Dimensional Mixing Characteristics in Seomjin River Estuary. J Korean Soc Mar Environ Eng. 2008;11:164–174.


30. Seo D, Yu HError Analysis of a Steady State Water Quality Model, QUAL2E using a Combination of EFDC-hydro and WASP7.2. In : IWA World Water Congress and Exhibition; 2008; Vienna.


31. Seo D, Kwon KH, Park BJ, Sigdel RWater Quality Modeling of Yongdam Lake Using EFDC 3-D Hydrodynamics Model and WASP7.3 Water Quality Model. In : WCWF; 2009; Incheon.


32. Seo D, Seo MJ, Goo MS, Yoo JKSerial Use of Hydrodynamic and Water Quality Model of the Geum River using EFDC-Hydro and WASP7.2. J Korean Soc Water Waste Water. 2009;23:15–22.


Fig. 1
Map of the Ara Canal and locations of important points.
/upload/thumbnails/eer-21-2-171f1.gif
Fig. 2
Initial grid development for the Ara Canal.
/upload/thumbnails/eer-21-2-171f2.gif
Fig. 3
Water Level vs Volume Curves of HEC-RAS model and EFDC model for the study.
/upload/thumbnails/eer-21-2-171f3.gif
Fig. 4
Tidal level of West Sea and bottom level of West Gate of the Ara Canal.
/upload/thumbnails/eer-21-2-171f4.gif
Fig. 5
Flow pattern through water gates of the Ara Canal for scenario 2 gate operation.
/upload/thumbnails/eer-21-2-171f5.gif
Fig. 6
Salt concentration predictions for different vertical grid systems.
/upload/thumbnails/eer-21-2-171f6.gif
Fig. 7
Salt concentration predictions for different horizontal grid systems.
/upload/thumbnails/eer-21-2-171f7.gif
Fig. 8
Salt concentration differences for different grid systems.
/upload/thumbnails/eer-21-2-171f8.gif
Fig. 9
Salinity distributions for a different number of vertical layers.
/upload/thumbnails/eer-21-2-171f9.gif
Fig. 10
Average salinity of three vertical layer groups for different vertical layers.
/upload/thumbnails/eer-21-2-171f10.gif
Fig. 11
Average salinity of three vertical layer groups for different horizontal grids.
/upload/thumbnails/eer-21-2-171f11.gif
Fig. 12
Distributions of average salt concentration difference groups for different number of horizontal cells.
/upload/thumbnails/eer-21-2-171f12.gif
Fig. 13
CFL number for different horizontal cells.
/upload/thumbnails/eer-21-2-171f13.gif
Fig. 14
Matrix to select the optimum grid system for salinity estimation in the Ara Canal.
/upload/thumbnails/eer-21-2-171f14.gif
Table 1
Physical Specification of the Ara Canal
Site Categories Values and Units
Han Gate Gate Dimension 22 × 150 m
Length 1.2 km
Width 300 m
Bottom Elevation EL −3.6 m
Canal Way Length 15.6 km
Width 80 m at Bottom
Bottom Elevation EL −3.6 m
West Gate Gate Dimension 28.5 × 210 m
Length 1.2 km
Width 400 m
Bottom Elevation EL −5.5 m
Table 2
Inflow Scenarios through Both Water Gates of the Ara Canal
Scenario 1 Scenario 2 Scenario 3 Scenario 4
Average Inflow [CMS] West Gate 20 20 20 27
Han River 10 10 3 3

Maximum Flow [CMS] 20 251 205 282
Inflow Interval daily 15 min 3 min 3 min
Table 3
Grid Sizes Scenario for the Analysis Salt Concentration Dynamics in the Ara Canal
15 Vertical Layers 216 Horizontal Cells
Longitudinal Cells 18 36 48 72 144 216 360 504 Vertical Cells 3 6 9 15 21 30
Length [m] 1000 500 370 250 125 80 50 35 Depth [m] 2.1 1.0 0.7 0.4 0.3 0.2
Total Grids 1,140 2,310 2,760 4,605 9,240 13,905 23,100 32,340 Total Grids 2,781 5,562 8,343 13,905 19,467 27,810
Table 4
AME and RMSE Values for Different Number of Vertical Grids when Compared with the Most Complicated Grid System (30 vertical × 540 horizontal) in This Study
No. of Vertical Layers AME RMSE

West Gate Canal Way Han Gate AVE. West Gate Canal Way Han Gate AVE.
3 6.40 4.43 2.47 4.43 6.88 4.81 2.87 4.85
6 2.90 3.99 1.81 2.90 3.33 3.56 2.14 3.01
9 1.22 2.51 0.96 1.56 1.46 3.13 1.21 1.93
15 0.71 0.90 0.47 0.69 0.87 1.12 0.60 0.86
21 0.49 0.35 0.21 0.35 0.58 0.51 0.32 0.47
Table 5
AME and RMSE Values for Different Number of Horizontal Grids
No. of Horizontal Cells AME RMSE

West Gate Canal Way Han Gate AVE. West Gate Canal Way Han Gate AVE.
18 1.01 4.07 2.18 2.42 1.19 4.69 2.60 2.83
36 1.27 2.83 3.25 2.45 1.51 3.18 3.50 2.73
48 0.82 2.46 1.70 1.66 0.97 2.87 2.02 1.95
72 0.63 1.18 0.74 0.85 0.83 1.50 0.92 1.08
144 0.56 1.46 0.80 0.94 0.67 1.42 1.04 1.04
216 0.44 0.87 0.59 0.63 0.54 0.98 0.80 0.77
360 0.32 0.46 0.40 0.39 0.40 0.58 0.55 0.51
TOOLS
PDF Links  PDF Links
PubReader  PubReader
Full text via DOI  Full text via DOI
Download Citation  Download Citation
  E-Mail
  Print
Share:      
METRICS
2
Crossref
2
Scopus
2,743
View
38
Download
Editorial Office
464 Cheongpa-ro, #726, Jung-gu, Seoul 04510, Republic of Korea
TEL : +82-2-383-9697   FAX : +82-2-383-9654   E-mail : eer@kosenv.or.kr

Copyright© Korean Society of Environmental Engineers. All rights reserved.        Developed in M2community
About |  Browse Articles |  Current Issue |  For Authors and Reviewers