《APPLICATION OF THE SWAT MODEL FOR RIVER FLOW FORECASTING IN SRI LANKA.doc》由会员分享，可在线阅读，更多相关《APPLICATION OF THE SWAT MODEL FOR RIVER FLOW FORECASTING IN SRI LANKA.doc（12页珍藏版）》请在三一文库上搜索。
1、Application of the SWAT Model for River Flow Forecasting in Sri Lanka H. A . Prasantha HAPUARACHCHI1 , LI Zhijia1 2: International Water Management Institute, Battaramulla, Sri Lanka) Abstract In the present study, the SWAT model and the Xinanjiang model have been used for daily flow forecasting of
2、the Kalu River upper catchment in Sri Lanka. Kalu River is the second largest river in Sri Lanka and due to heavy rainfalls over the catchment, steep river slopes with narrow valleys in the upper catchment and mild riverbed slopes with wide and flat plains in the middle and lower catchments, the flo
3、ods in Kalu River basin have become regular. The SWAT model has been used for daily river flow predictions in the Kalu River, and compared with the results obtained using the Xinanjiang model. In this study, the Xinanjiang model has performed slightly better than the SWAT model for forecasting the d
4、aily flow of Kalu River. In fact it might be partly attributable due to the poor quality and inadequate data, since the output of the SWAT (distributed model) strictly depends on the quality of input data. In addition, many people in Sri Lanka use well water for their domestic purposes. When conside
5、ring a catchment as a whole, normally it is a very large area, and therefore it is not possible to record or count all the individual minor scale water utilizations in detail such as small irrigation, animal husbandry in minor scale and industrial water utilizations in minor scale. The cumulative va
6、lue of such water utilizations might be large. The absence of these data may specially affect the distributed models in water balancing. But the conceptual watershed models (e.g. Xinanjiang model) are capable of adjusting their parameters while calibrating, according to the situation since most of t
7、heir parameters have no physical background. As a result conceptual watershed models show better performance than distributed models where the catchment characteristics and model inputs are limited or incomplete. Keywords: Xinanjiang model, SWAT model, conceptual watershed models, distributed waters
8、hed models, river flow forecasting In the present study, the SWAT model 1 and the Xinanjiang model 2-3 have been used for daily river flow forecasting of the Kalu River upper catchment in Sri Lanka. Kalu River is the second largest river in Sri Lanka. Basically Sri Lanka receives rainfall in two mon
9、soon seasons. Supported by the International Water Management Institute (IWMI). Received:2003-08-01;Accepted:2003-12-17. H. A. Prasantha HAPUARACHCHI, male, Ph.D, email: hapuarayamanashi.ac.jp. Due to its geographical location, Kalu river catchment receives rain during both of these monsoon seasons.
10、 Average annual rainfall of the overall catchment is around 4000mm and it ranges from 2750mm in coastal areas to 5000mm in mountainous areas. Since the catchment is entirely situated in the wet zone, it has a high rainfall to runoff response. This high volume of water often discharges as floods. Flo
11、ods result in damage to houses, property and even lives. Severe environmental problems such as deforestation and soil erosion can be seen in this catchment due to Chena cultivation and gem mining. In normal practice of Chena cultivation, farmers destroy a part of forest by burning and cultivate trop
12、ical plants. After some years, when the land becomes non-arable, they move to another place and practice the same. Due to Chena cultivation, tree felling on an extensive scale and the periodic replanting of tea and rubber plantations, the upper slopes of the catchment are not stable and landslides c
13、an be seen often. Also Kalu river upper catchment is popular for gem mining. Normally gem-bearing gravels occur in beds or pockets and are found 2-20m beneath the surface. Gem bearing gravels show horizontal extensions and therefore horizontal tunneling is resorted to when mining. Some times these t
14、unnels are several kilometers long and a causative factor for land subsidence in later time. Besides this, there are many environmental problems that could be attributed to gem mining. Sedimentation of clay minerals in rivers and tanks, lowering the ground water table, slope instability, limiting th
15、e extent of cultivable land and reducing yield due to soluble minerals which are products of gemming such as calcium, magnesium, potassium, and mosquito breading in abandoned pits are some of these problems. Considering these factors, it is important to model the Kalu river upper catchment to identi
16、fy future environmental hazards. In the present study, the SWAT model has been applied to generate daily flows in the Kalu river upper catchment to identify in the catchment hydrological responses. The Xinanjiang model (a conceptual rainfall runoff model) has also been applied to forecast the daily
17、flows and to compare its results with the SWAT model. 1 SWAT model SWAT model 1was developed by the Agricultural Research Service (ARS) of the US Department of Agriculture (USDA). It was developed to predict the impact of land management practices on water, sediment and agricultural chemical yields
18、in large complex watersheds with varying soils, land use and management conditions over long periods of time. Application of the model requires detailed information about weather, soil properties, topography, vegetation, and land management practices in watershed. SWAT is capable of handling water m
19、ovement, sediment movement, crop growth, and nutrient cycling with the above information. In the present study, only the hydrologic processes of SWAT are considered. It is not possible to give a complete explanation about the SWAT model in this document. A detailed description of the SWAT model and
20、user manual can be found in the web site http:/www.brc.tamus.edu/swat/ or Neitsch et a l 1. 2 Xinanjiang model Xinanjiang rainfall-runoff model 2-3 is a conceptual watershed model developed at Hohai University, China in 1970s.It provides an integral structure, statistically describing the non- unifo
21、rm distribution of runoff producing areas, which features it as one of the conceptual, semi- distributed hydrological models developed. By comparing with the Pitman model of South Africa, Sacramento model of USA, NAM model of Europe and the SMAR model of Ireland, Gan 4 concluded that the Xinanjiang
22、model did consistently better, even in dry catchments. The model consists of 15 parameters (see Tab.1) and performs best for the humid and semi humid catchments. All the parameters of Xinanjiang model should be calibrated before application. The hydrological data inputs of the model are areal mean r
23、ainfall, and measured pan evaporation. Besides these, sub catchment areas, and initial state of the catchment are necessary for the calculations. A complete description of Xinanjiang model can be found in 2-3, Tab. 1 Parameters of the Xinanjiang model KRatio of potential evapotranspiration to pan ev
24、aporation Evapotranspiration WUM Tension water capacity of upper layer ParametersWLM Tension water capacity of lower layer CEvapotranspiration coefficient of deeper layer Runoff GenerationWM Areal mean tension water capacity ParametersBExponential of the distribution of tension water capacity IM Rat
25、io of impervious area to the total area of the basin SMFree water storage capacity Runoff SeparationEXExponential of distribution water capacity ParametersKGOutflow coefficient of free water storage to the ground water flow KIOutflow coefficient of free water storage to the inter flow Runoff Concent
26、rationCGThe recession constant of ground water storage ParametersCIThe recession constant of lower interflow storage CSThe recession constant of channel network storage Flow Routing ParameterXEMuskingum coefficient 3 Catchment characteristics The Kalu river upper catchment area is about 603km2 and l
27、ocated in southwest Sri Lanka to the south of the central highlands and lies between 80.40-80.60N latitude and 6.53-6.80E longitude. Elevations vary from about 100m to 2225m above MSL. Mountain ranges, high peaks, dissected plateaus, and escarpments cover a large part of the area. On average, the sl
28、opes vary (depending on their lithology and structure) from about 10 to 35 in the upland ridges 5. Due to the geographical location, Kalu river catchment receives rain during both monsoon seasons, from May to June and from September to October. The mean daily values of precipitation, evaporation and
29、 discharge at the outlet are 9.58mm, 3.14mm and 7.04mm respectively. The average annual temperature in the catchment ranges from 26.9 to 27.8. Distribution of soils in the catchment area has close affinity with topography, geology, and climate. Dominant soil types visible in this area are Red Brown
30、Earths and Low Humic Gray soils, Reddish Brown Earths and Immature Brown Loams, Red Yellow Podzolic soils, Bog and Half-bog soils, and alluvial soils. However, nearly 86% of the area is covered with Red Yellow Podzolic soils 5. The two main vegetation types are tropical rain forests and mountain for
31、ests. There are 13 types of land use classes in the catchment as shown in Table 2. About 30.2 percent of the land is used for Chena cultivation 6. Tea and rubber plantations are the other major land uses (see Tab.2). Tab. 2 Land use classes Land Use Paddy Tea Rubber Chena Forest Other Plantations Gr
32、ass Lands Scrub Rock Garden Build- up area Area (%)4.90 14.1 17.6030.30 18.001.800.400.200.212.300.2 5 Statistical indices Two statistical indices have been used to compare the application results. They are the Nash Sutcliffe coefficient Dy 7 and the percentage of total error (%Err) in each year. (1
33、) (2) Where Qobs is the daily mean observed discharge, Qobs is the observed discharge, Qcal is the calculated discharge, m is the number of time steps in each year and wi is a weighting factor (usually equal weights are used for each year). The %Err is obtained as a percentage and depending on the s
34、ign (positive or negative), the calculated discharge can be lower or higher than the observed discharge. 6 Results 6.1 Application of the SWAT model Firstly, the watershed was delineated into 15 sub catchments using a DEM (100m100 m cell size) and a digitized river network. The number of sub catchme
35、nts generated depends on the threshold limit of flow accumulation. SWAT was applied for the period 1987 to 1995 for daily stream flow generation. However we only can compare the results of two years, 1994 and 1995, that are common to the validation period of the Xinanjiang model. The application res
36、ults are shown in Tab. 3. It is clear that the %Err (%Err9) is quite high for all years indicating that the water balance of the catchment is not modeled correctly. Furthermore, the highest Nash coefficient (Dy) obtained was only 78% (for 1992). 100% 1 obs, 11 cal,obs, m i i m i m i ii Q QQ Err m i
37、ii m i iii y QQw QQw D 1 2 obsobs, 2 1 2 cal,obs, 2 2 1 Tab. 3 Application results for the SWAT model FROMTOQobs (mm)Qcal (mm)%ErrDy 1987-01-011987-12-311820.681483.1418.540.720 1988-01-011988-12-313230.342552.1720.990.703 1989-01-011989-12-313050.902748.819.900.762 1990-01-011990-12-312260.351882.3
38、616.720.774 1991-01-011991-12-312095.511774.9415.300.759 1992-01-011992-12-312451.882197.6610.370.782 1993-01-011993-12-313404.042409.8729.210.763 1994-01-011994-12-312105.631470.1030.180.715 1995-01-011995-12-313039.802374.7321.880.705 Tab. 4 Calibration Results for the Xinanjiang Model FROMTOQobs
39、(mm)Qcal (mm)%ErrDy 1987-01-011987-12-311820.691865.74-2.480.853 1988-01-011988-12-313230.353130.723.080.848 1989-01-011989-12-313050.912951.333.260.946 1990-01-011990-12-312260.362215.431.990.901 1991-01-011991-12-312095.512082.390.630.862 1992-01-011992-12-312451.892389.632.540.919 1993-01-011993-
40、12-313404.043340.351.870.908 6.2 Application of the Xinanjiang model Seven years of historical data (1987-1993) were used for calibration and two years of data (1994-1995) were used for verification of the model. All the parameters of the Xinanjiang model have been calibrated using the SCE-UA (Shuff
41、le Complex Evolution) method 6,8-9. SCE parameters were set as follows 10 for the calibration process: p=40, m=31, q=16, = 1, = 31, where p is the number of complexes, m is the number of points in a complex, q is the number of points in a sub complex, is the number of consecutive offspring generated
42、 by each sub complex, and is the number of evolution steps taken by each complex. The calibration results for the Xinanjiang model are shown in Table 4 and the verification results are shown in Tab. 5. According to these results, the %Err is much smaller (-3%Err4) and the Nash coefficient (Dy) is gr
43、eater than 84% for all calibration and verification years. Tab.5 Verification Results for the Xinanjiang Model FROMTOQobs (mm)Qcal (mm)%ErrDy 1994-01-011994-12-312105.632070.131.690.894 1995-01-011995-12-313039.802980.711.940.885 7 Discussion and conclusions Based on the results shown in Tab. 3 to T
44、ab. 5, the performance of the Xinanjiang model is better than the SWAT model for daily river flow predictions of the Kalu river upper catchment in Sri Lanka. Fig. 1 shows that the Xinanjiang model gives a good fit between the observed and calculated discharge hydrographs for the model verification y
45、ears. %ERR (in tab. 5) is small (0.88). Thereby it seems that the Xinanjiang watershed model can be successfully applied in humid or semi humid catchments in Sri Lanka for computing river flows. The input data requirement for the Xinanjiang model is much smaller (i.e. precipitation, pan evaporation,
46、 and observed discharge). Therefore it is suitable for modeling many catchments where comprehensive data are not readily available. Fig. 1 Discharge hydrographs (verification years) for the Xinanjiang Model Fig.2 Discharge hydrographs (verification years) for the SWAT Model 0 50 100 150 200 250 300
47、350 400 450 500 01-01-94 02-01-94 03-01-94 04-01-94 05-01-94 06-01-94 07-01-94 08-01-94 09-01-94 10-01-94 11-01-94 12-01-94 01-01-95 02-01-95 03-01-95 04-01-95 05-01-95 06-01-95 07-01-95 08-01-95 09-01-95 10-01-95 11-01-95 12-01-95 Date Discharge (m3/s) Qcal Qobs 0 50 100 150 200 250 300 350 400 450
48、 500 01-01-94 02-01-94 03-01-94 04-01-94 05-01-94 06-01-94 07-01-94 08-01-94 09-01-94 10-01-94 11-01-94 12-01-94 01-01-95 02-01-95 03-01-95 04-01-95 05-01-95 06-01-95 07-01-95 08-01-95 09-01-95 10-01-95 11-01-95 12-01-95 Date Discharge (m3/s) Qcal Qobs In this study, the performance of the SWAT mode
49、l was poor compared to the Xinanjiang model. Although the Nash coefficient (Dy) is acceptable (Tab.3), the %Err is high, indicating that the model is unable to predict accurately the water balance in the catchment. A careful inspection of Tab.3 and Fig. 2 reveal that the calculated discharge is often less than the observed discharge and consequently, %ERR is positive for each year. It may be partly attributable due to th