Trans-KARST 2004, Proceedings of the International Transdisciplinary Conference on Development and Conservation of Karst Regions, Hanoi, Vietnam, 13-18.9.2004 Eds O Batelaan, M Dusar, J Masschelein, Vu Thanh Tam, Tran Tan Van, Nguyen Xuan Khien FLOOD PREDICTION IN THE KARSTIC SUOIMUOI CATCHMENT, VIETNAM Y.B LIU1, O BATELAAN1, N.T HUONG2, V.T TAM2 and F DE SMEDT1 Dept Hydrology and Hydraulic Engineering, Vrije Universiteit Brussel, Belgium yongbliu@vub.ac.be Research Institute of Geology and Mineral Resources, Thanh Xuan-Hanoi-Vietnam Abstract The major obstacles to modelling flood processes in karstic areas are a lack of understanding and model representations of the distinctive features and processes associated with runoff generation in those regions and a lack of field data In this study, a distributed flood modelling approach, WetSpa, was performed by modifying model representations of some of the predominant features and processes of the karstic Suoimuoi catchment with complex-terrain and mixed land use in the northwest Vietnam The model was calibrated based on 15 months of hourly hydrometeorological data, topography, land use and soil types in GIS format, and used to continuously simulate both baseflow and fast-responding overland, conduit and channel flows during stormflow periods Considerable variability in simulation accuracy was found among storm events and within the catchment The simulation results showed that the model represents reasonably well stormflows generated by rainfall events in the study catchment, and the potential of using distributed flood simulation for estimating future flood conditions under changing land use conditions It is argued therefore that the WetSpa approach is suitable for application in karstic areas under human and natural pressure Keywords: flood prediction, WetSpa, GIS, karstic Suoimuoi catchment surface runoff occurs (Tam et al., 2001) The Suoimuoi catchment is characterized by a humid subtropical climate and influenced by the monsoon regime prevailing in Northern Vietnam Two distinct seasons can be observed in the area: the dry winter lasting from November to April and the extensive rainfall summer from May to October The yearly mean temperature is 21.1°C, the mean annual precipitation is 1450 mm of which about 85% Introduction The Suoimuoi River catchment is situated in the mountainous Da River basin in the Northwest Vietnam It covers an area of 273 km2 with the Suoimuoi sinkhole as the catchment outlet The catchment is confined by two regional deep fault systems trending in NW-SE direction, the Son La Fault on the east and the Da River Fault on the west A range of non-limestone and limestone rocks of different formations are exposed within the catchment as shown in Fig There is almost no surface water drainage in the karst area Instead, closed depressions exist here and there with cave systems developed in the bottom or in the rock walls (Tam, 2003) The karst aquifers receive water, mainly by the regional groundwater flow, with additional important in-situ recharge by rainfall, surface water and exotic water from higher-lying non-karstic areas The movement of karst groundwater is closely controlled by these tectonic conditions The groundwater is mainly stored in fractures, crushed zones and caves Along the river course, there exist a number of karst springs/resurgences and sinkholes in which the interaction between karst groundwater and Boundary Limestone formation Hamrong Streams Banpa p and Chiengpa c # Suoi Muoi Holocene sink hole Camthuy and Nam tham Upper Donggiao # N 10 km Fig 1: Distribution of karst limestone in the Suoimuoi catchment 139 Trans-KARST 2004 Liu et al infiltration, evapotranspiration, soil moisture storage, interflow, percolation, groundwater storage and discharge The model uses the spatial information of catchment topography, soil type and land use, and recorded meteorological data to predict river flow hydrographs and spatially distributed hydrological characteristics, such as soil moisture, infiltration rates, groundwater recharge, surface water retention or runoff, etc For the non-limestone areas in the Suoimuoi catchment, the WetSpa model is applied, for which the runoff of each grid cell is calculated by a modified rational method Vs = CP(θ θ s ) (1) where Vs is the amount of surface runoff [L], P the net precipitation [L] (rainfall minus interception), θ the soil moisture content [L³/L³], θs the saturated soil moisture content [L³/L³], and C a potential runoff coefficient [-], which is assumed to depend upon slope, soil type and soil cover and interpolated from values collected from literature Next, the generated runoff is routed to the catchment outlet along its flow path using the method of linear diffusive wave approximation (Liu et al., 2003) falls during the rainy season in summer According to Tam (2003), the total discharge of the Suoimuoi River tributaries located in the non-limestone area west of the Sonla fault contributes only 7-10% to the total river discharge measured at the Suoimuoi sinkhole However, their discharge contribution can rise up to 25% of the total river discharge during storms due to surface runoff generated from the steep non-limestone rocks During the period 2000-2003, an extensive hydrological and geophysical survey was conducted to study the mechanisms of hydrogeological processes in the Suoimuoi catchment Many sophisticated methods, such as computer modelling, hydrogeological mapping, tracer and pumping test, etc., are performed to analyze complex groundwater systems However, computer modelling is difficult to realize accounting for turbulentflow conduits in the karst areas Even dye tracing, which is usually considered the most convincing tool for delineating groundwater basins in karst, is not physically or economically feasible in some cases, and rarely gives more than an outline of the major conduit flow Geophysical surveys may help to delineate the local geologic framework and major conduits, but the surveys cannot determine detailed flow patterns and divides in karst areas However, all these methods provide significant information and analysis in delineating drainage systems and determining hydrological characteristics of the karst aquifer In this paper, a flood simulation approach for the Suoimuoi catchment using the modified WetSpa hydrological model is presented, for which modelling processes and parameters are adjusted separately for the limestone and non-limestone areas based on 15 months of hourly hydrometeorological data U (t ) = ⎡ (t − t )2 ⎤ exp ⎢− ⎥ (2) ⎢⎣ 2σ t t ⎥⎦ σ 2πt t 03 where U(t) [T-1] is the flow path unit response function at time t, t0 [T] and σ [T] are mean flow time and its standard deviation The parameters t0 and σ are spatially distributed, so that each flow path has different parameters depending on the length of the flow path and the physical characteristics of the flow path elements The total direct flow at the catchment outlet is calculated by the convolution integral of all flow path responses subjected to the spatially distributed runoff computed for each grid cell The infiltrated water into the soil is used for consequent percolation, interflow and evapotranspiration, which are controlled by the moisture content, hydraulic gradient and soil textures The groundwater flow is simulated using a linear reservoir method on a small subcatchment scale forming the baseflow of river discharge The schemes of WetSpa model are not valid in simulating processes of the kastic area in the catchment due to the change of hydrological regime Water may flow overland Methodology The watershed model approach used in this study is a modification of the WetSpa model, which was originally developed by Wang et al (1997) to study the Water and Energy Transfer between Soil, Plant and Atmosphere, and adapted to flood prediction on hourly basis by De Smedt et al (2000) and Liu et al (2002, 2003) The hydrological processes are simulated in a grid-based schematisation of a river basin including precipitation, interception, depression, surface runoff, 140 Trans-KARST 2004 Liu et al have a strong vertical component in the unsaturated zone and a strong tendency to follow the strike in the saturated zone Conduits carry high-velocity turbulent flow, and they include caves that are large enough to explore The statements about preferred flow routes in this study are supported by the mapping of accessible conduits (Hung et al., 2002) The total hydrographs at the catchment outlet are obtained by summation of the direct flow, interflow and groundwater flow from the non-limestone areas and the conduit flow and groundwater flow from limestone areas from ridge tops, and then enter the ground in upland regions through recharge features and resurgent at springs in low areas Diffuse infiltration can also take place through the soil or through epikarst On steep slopes that not readily develop sinkholes, diffuse infiltration can occur through the soil or into bedrock fissures Moreover, it is difficult to identify groundwater flow paths and divides in karst aquifers, which arises from the extreme heterogeneity and anisotropy of the karst aquifer, and from changes in groundwater patterns with different stages of flow For example, groundwater flow paths, divides, and basin boundaries can shift in response to rising groundwater levels during and after major precipitation events Taking account of the above specific characteristics, the WetSpa model is modified as follows in order to better represent the predominant hydrological features of the karst area in the catchment Surface runoff coefficient is set to zero to reflect the condition that almost no surface flow is apparent in the karst areas Water that contributes to conduit runoff in the unsaturated zone is estimated taking account of the effects of slope, soil type, land use, moisture content, and is assumed to be a linear function of the expected surface runoff in the WetSpa model, i.e Vc = αCP(θ θ s ) (3) where Vc is the amount of water that contributes to conduit flow [L], and α [-] is a global parameter within the range – and realized through model optimization Routing of conduit flow is accomplished by the method of diffusive wave approximation as described in Eq 2, but its concentration time and variance are adjusted based on the analysis of observed hydrographs The parameter of hydraulic conductivity and other soil features (porosity, pore size distribution index, etc.) are readjusted through model optimization Groundwater flow is simulated using a linear reservoir method, for which the flow recession coefficient is obtained from the analysis of observed hydrographs Through above modification, the WetSpa model is used to simulate the flow responses to storm events in the karstic Suoimuoi catchment Specifically, the pattern of individual groundwater flow paths tends to Application The measured hydrometeorological data during October 2000 to March 2001 are used to calibrate model parameters in this study The hourly stream flow into the Suoimuoi sinkhole was captured by an automated water level logger The recorded hourly series of water level was converted to the flow hydrograph by a well calibrated rating curve that was constructed on the basis of many discharge-water level pairs measured at different stages of the stream flow (Tam, 2003) The resulting hydrographs are used in the baseflow separation and the model validation Hourly precipitation was monitored by an automated logger located km upstream of the Suoimuoi sinkhole, and was assumed uniformly distributed over the catchment In addition, the data of potential evapotranspiration and air temperature were collected from a nearby gauging station, which are used as input to the WetSpa model Topographical maps at scale 1:50,000 were available and cover the entire Suoimuoi catchment Based on these maps consisting of a 20m contour level, A DEM with 50m spatial resolution and the surface drainage network with drainage density of 0.66 km/km2 were created as shown in Fig The topography of the catchment is characterized by highlands in the upper part and lowlands in the lower part of the catchment Elevation ranges from 539 to 1815m with an average catchment slope of 33.2% The major soil types of the catchment are Cambisol (43.7%) distributed in the highland areas and bed rock (22%) distributed in the lowland areas in which mature karst landscapes are characterized Other soil types are Fluvisol, Luvisol, Leptosol, Travertin, Acrisol and Nitrosol, which are distributed 141 Trans-KARST 2004 Liu et al Boundary Streams Elevation ( m) 53 - 616 61 - 726 72 - 836 83 - 946 94 - 1055 10 55 - 1165 11 65 - 1275 12 75 - 1385 13 85 - 1495 14 95 - 1815 Boundary Streams N N Land use Paddy field Grass land Open canopy forest Close canopy forest Upland field Shrub Residential area 10 km 10 km Fig 2: Topographic map of the Suoimuoi catchment Fig 3: Land use map of the Suoimuoi catchment over the catchment All these soil types were converted to the USGS soil texture classes for use in the WetSpa model (Huong, 2002) The distinguished land use types are: close canopy forest (1.7%), open canopy forest (4.2%), shrub (40.4%), grass land (5.6%), upland fields (38.3%), paddy fields (5.2%), residential area (4.5%) and open water (0.01%) and are distributed as shown in Fig The above land use categories were further converted to the WetSpa land use types based on vegetation and land use assessment (Huong, 2002) Model parameters are identified firstly using GIS tools and lookup tables, which relate default model parameters to the base maps, or the combination of base maps Starting from the 50 by 50 m pixel resolution digital elevation map, hydrologic features including surface slope, flow direction, flow accumulation, flow length, stream network, drainage area and sub-catchments are delineated The threshold for determining the stream network is set to 50, i.e the cell is considered to be drained by streams or conduits when the total drained area becomes greater than 0.125 km² The threshold for delineating subcatchments and main streams is set to 1000 Maps of porosity, field capacity, wilting point, residual moisture, saturated hydraulic conductivity and pore size distribution index are obtained from the soil type map Maps of root depth, Manning’s roughness coefficient and interception storage capacity are derived from the land use map Maps of default runoff coefficient and depression storage capacity are calculated from the slope, soil type and land use class combinations The residential areas are mainly distributed besides the Suoimuoi river channel as villages or small towns Due to the grid size, the residential cell is assumed 10% covered by impervious materials (roof, road, etc.), and the rest covered by farmland The average flow depth is estimated using the power law relationship (Molnar and Ramirez, 1998) with an exceeding probability of a 2-year return period resulting in a minimum overland flow depth of 0.005 m and the channel flow depth of 1.0 m at the catchment outlet By combining the maps of the average flow depth, the Manning’s roughness coefficient and surface slope, average flow velocity in each cell is calculated using Manning’s equation, which results in a minimum value of 0.005 m/s for overland flow, and up to 2.5 m/s for some parts of the main river Next, the celerity and dispersion coefficient at each cell are produced, and the values of concentration time and its standard deviation for each contributing cell are generated as described by Liu et al (2003) With the above information, the unit flow path response functions are calculated from each cell to the sub-basin outlet and from the sub-basin outlet to the basin outlet In dealing with the specific problems of karst areas in the Suoimuoi catchment, the WetSpa model is modified, the surface runoff coefficient is set to zero, and the conduit flow and groundwater flow are estimated separately by a conceptual method and a linear reservoir method The volume of water contributed to the conduit flow in the unsaturated zone is assumed to be a linear function of the surface runoff in the non-limestone areas under the 142 Trans-KARST 2004 Liu et al factor for estimating the volume of conduit flow is found around 0.15 The concentration time of conduit flow is about 1.5 times the surface runoff, while the hydraulic conductivity is about 2.5 times the default value and the soil pore size distribution index is around 1.0, which leads to a very high percolation to the saturated zone in the karst areas A graphical comparison between observed and predicted hydrographs during the simulation period is presented in Fig It can be seen from the figure that the hydrograph at the Suoimuoi sinkhole is well reproduced by the model Four statistical evaluation criteria were applied to the 15 months simulation results to assess the model performance It is found that the WetSpa model reproduces the observed water volume with -3.4% under estimation The model Nash efficiency for reproducing the river discharges is 69% (Nash and Sutcliffe, 1970) The adapted Nash efficiency for reproducing low flows is 85%, and for high flows 70%, which indicate that the model is suitable for water balance simulation and flood prediction in the karstic Suoimuoi catchment, but the accuracy of peak discharge prediction needs to be improved The model is also able to simulate the spatial variation of other hydrological characteristics at each time step, including surface runoff, infiltration, actual evapotranspiration, groundwater recharge, etc This gives the advantage of computer automation and analyzing the effects of same condition of slope, soil type and land use Likewise, the concentration time of conduit flow is estimated by the calculated surface flow time multiplied by a correction factor Using GIS tools and the hydrological modelling extension, the average calculated flow time of the karst areas is computed by integration of the flow time of each grid cell weighed by its percentage of area and flow coefficient A correction coefficient is then obtained by comparing the value with observed hydrographs at the catchment outlet, and applied to each karst cell in the catchment Additionally, model parameters in the karst areas, such as the hydraulic conductivity, pore size distribution index, etc., are adjusted during model calibration by multiplying a correction factor for each of them The groundwater recession constant at the catchment outlet is found from the baseflow separated from the observed hydrographs This value is around 0.018 day-1 according to Tam (2003), and is adjusted for each subcatchment based on its slope, drainage area and geological features Model calibration is implemented by comparing the simulated hydrograph with the observed hydrograph Each of the correction factors and functions involved the use of coefficients is determined using an independent model optimization process (Doherty and Johnston, 2003) The objective function is the sums of squares of the difference between observed and predicted flows at the Suoimuoi sinkhole The correction Fig 4: Observed and calculated flow hydrographs at Logger station during 2000-2001 143 Trans-KARST 2004 Liu et al using GIS and remote sensed land use information, ed., Brebbia, C.A., 295-304, Risk Analyses II, WIT press, Southampton, Boston topography, land use and soil type on the hydrologic behaviour in a river basin Conclusions Doherty, J and Johnston, J.M., 2003 Methodologies for calibration and predictive analysis of a watershed model, Journal of the American Water Resources Association, 39(2), 251-265 A test of a GIS-based modelling approach for flood prediction in the karstic Suoimuoi catchment was described The model uses a modified rational method to calculate surface runoff in non-limestone areas and conduit flow in limestone areas based on the spatial characteristics of topography, soil type, land use and moisture condition Flow into the outlet sinkhole was routed with the linear diffusive wave approximation method, while the concentration time of conduit flow is multiplied by a correction coefficient Total discharge at the basin outlet was calculated by summing predicted flow from both nonlimestone and limestone areas in the catchment The model was calibrated on a 15month flow data series collected at the Suoimuoi sinkhole The results of the calibration show that in general flow hydrographs are well predicted, especially the baseflow at the catchment outlet However, the predictions of peak discharge for some of the storms are not satisfied indicating the need for improved methods of runoff volume calculation flow routing in karstic catchments As described in the paper, the karstic aquifers in the Suoimuoi catchment possess large underground reservoirs of water, but these reservoirs are difficult to exploit because little is known about their hydraulic behaviour A simple hydrological model, like the WetSpa model used in this study, can provide useful information about the behaviour of such complex flow system The model explicitly acknowledges the lack of detailed knowledge about the location and size of conduits and other flow paths with fewer data requirements and calibration parameters In addition, the effects of topography, soil type and land use on potential runoff, recharge and outflow can also be evaluated Work is continuing on incorporating a more physical-based approach in estimation of runoff volume and flow transport into the model to study the complex hydrological behaviour of the river catchment Hung, L.Q., Dinh, N.Q., Batelaan, O., Tam, V.T and Lagrou, D., 2002 Remote sensing and GISbased analysis of cave development in the Suoimuoi Catchment (Son La - NW Vietnam), Journal of Cave and Karst Studies, 64(1), 23-33 Huong, N.T., Application of the WetSpa model to the Suoimuoi catchment, Vietnam, MSc Thesis, Inter-University Programme in Water Resources Engineering, Katholieke Universiteit Leuven and Vrije Universiteit Brussel, Belgium, 2002 Liu, Y.B., Gebremeskel, S, De Smedt, F, Hoffmann, L and Pfister, L., 2003 A diffusive transport approach for flow routing in GISbased flood modelling, Journal of Hydrology, 283, 91-106 Liu, Y.B., Gebremeskel, S., De Smedt, F and Pfister, L., 2002 Flood prediction with the WetSpa model on catchment scale, eds., Wu et al., 499-507, Flood Defence ‘2002, Science Press, New York Ltd Molnar, P and Ramirez, J.A., 1998 Energy dissipation theories and optimal channel characteristics of river networks, Water Resources Research, 34(7), 1809-1818 Nash, J.E and Sutcliffe, J.V., 1970 River flow forecasting through conceptual models, Part 1: A discussion of principles, Journal of Hydrology, 10, 282-290 Tam, V.T., 2003 Characterization of a Kastic system by an integrative and multi-approach study, a case study of Suoi Muoi and Nam La catchments in the Northwest Vietnam, Doctoral Thesis, Vrije Universiteit Brussel, Belgium Tam, V.T., Vu, T.M.N and Batelaan, O., 2001 Hydrological characteristics of a karst mountainous catchment in the Northwest of Vietnam, Acta Geologica Sinica, Journal of the Geological Society of China, 75(3), 260-268 Wang, Z.M., Batelaan, O and De Smedt, F., 1997 A distributed model for water and energy transfer between soil, plants and atmosphere (WetSpa), Physics and Chemistry of the Earth, 21(3), 189-193 References De Smedt, F., Liu, Y.B and Gebremeskel, S., 2000 Hydrological modelling on a catchment scale 144