Water Resour Manage (2012) 26:1387–1407 DOI 10.1007/s11269-011-9964-1
Scenario-based Impact Assessment of Land Use/Cover and Climate Changes on Water Resources and Demand: A Case Study in the Srepok River Basin, Vietnam— Cambodia Tran Van Ty & Kengo Sunada & Yutaka Ichikawa & Satoru Oishi
Received: 15 October 2010 / Accepted: 22 December 2011 / Published online: 11 January 2012 # Springer Science+Business Media B.V. 2012
Abstract This study investigates an interdisciplinary scenario analysis to assess the potential impacts of climate, land use/cover and population changes on future water availability and demand in the Srepok River basin, a trans-boundary basin. Based on the output from a high-resolution Regional Climate Model (ECHAM 4, Scenarios A2 and B2) developed by the Southeast Asia—System for Analysis, Research and Training (SEA-START) Regional Center, future rainfall was downscaled to the study area and bias correction was carried out to generate the daily rainfall series. Land use/cover change was quantified using a GIS-based logistic regression approach and future population was projected from the historical data. These changes, individually or in combination, were then input into the calibrated hydrological model (HEC-HMS) to project future hydrological variables. The results reveal that surface runoff will be increased with increased future rainfall. Land use/cover change is found to have the largest impact on increased water demand, and thus reduced future water availability. The combined scenario shows an increasing level of water stress at both the basin and sub-basin levels, especially in the dry season. Keywords Climate change . Land use/cover change . Water availability . Water demand . Water stress
T. V. Ty (*) College of Technology, Cantho University, Campus 2, 3/2 Street, Ninh Kieu District, Cantho City, Vietnam e-mail:
[email protected] K. Sunada : Y. Ichikawa Interdisciplinary Graduate School of Medicine and Engineering, University of Yamanashi, 4-3-11 Takeda, Kofu, Yamanashi 400-8511, Japan S. Oishi Research Center for Urban Safety and Security, Kobe University, 1-1 Rokkodai, Nada-ku, Kobe 657-8501, Japan
1388
T.V. Ty et al.
1 Introduction Water resources in many countries are currently under severe pressure from human intervention, such as growing domestic, industrial, and irrigation demands; and the changing of runoff patterns caused by climate and land use/cover changes. Assessing water resources becomes a complex task that must consider many aspects, of which climate and land use/ cover changes are considered the two major factors. Better understanding of the impacts of these changes on hydrological processes is thus of paramount importance for sustainable water resources development. Population growth and human-induced development have accelerated the speed of land use/cover changes, which in turn influence interception, infiltration, and evaporation processes in the hydrological cycle, and thus water availability and demand. These changes become more noticeable in tropical developing countries characterized by agriculture-based economics and rapidly growing populations (Grau et al. 2003). In addition, climate change may affect many aspects of natural ecosystems. For example, the amount of water withdrawal for crop irrigation is expected to increase as rainfall decreases and evapotranspiration increases with higher air temperature (Sun et al. 2008). Hence, comprehending climate change impacts on hydrological conditions is essential to enable more efficient water resources development, especially in the Mekong River Basin (MRB) where water flow is considered an essential part of local livelihoods. The impact of climate change and human development on the future state of water resources has been intensively studied at global scale (Vorosmarty et al. 2000; Oki and Kanae 2006) and at basin scale (Mizyed 2009; Abdulla et al. 2009; Elias and Vernon 2010; Feng et al. 2011); whereas it is rare to find a study that assesses the magnitude of these impacts on local water resources. In the MRB, many studies have investigated the impact of climate change on future water availability (Hoanh et al. 2003; Chinvanno 2004; Kiem et al. 2008; Eastham et al. 2008). For example, Kiem et al. (2008) estimated that annual mean rainfall projected by the Japan Meteorological Agency (JMA) would increase 4.2% during 2080–2099, compared to 1979–1998 for the entire MRB. Eastham et al. (2008) selected 11 GCMs (scenario A1B) to construct scenarios of future (2030) rainfall in the MRB and found that the most likely projected response in annual rainfall averaged across the basin is a 13.5% increase. However, previous studies focus only on the large scale, with the results indicating broadly similar responses to future climates; for example, rainfall is projected to increase across the basin. In addition, on the local scale, Kawasaki et al. (2010) have carried out a study on evaluating the potential impacts of rainfall and land use changes on streamflow in the Srepok River basin. This study assumed an annual rainfall increase of 2.5% and 5% for the year 2025 and 2050, respectively. The increased rainfall was then applied uniformly to the daily observed 2001 rainfall at seven stations in the study basin to assess the impact of climate change on streamflow at the outlet of the basin. Because climate characteristics, the degree of impact, and adaptations vary across spatial scales, investigating and understanding the extent of the potential impact of climate, land use/cover, and population changes on local water resources, particularly in a developing area such as the MRB, is therefore needed. In this study, the impacts of climate, population, and land use/cover changes on water availability and demand, as well as water stress in the Srepok River basin, were quantitatively assessed, individually or in combination. This is achieved by developing future scenarios of climate, population, and land use/cover changes; and by examining future water stress situation using the calibrated hydrological model under different developed scenarios across 13 sub-basins.
A Case Study in the Srepok River Basin, Vietnam—Cambodia
1389
2 Study Area and Data 2.1 Study Area The Mekong River rises in the Tibetan Plateau and empties into the South China Sea after travelling 4,200 km through six countries: China, Myanmar, Thailand, Lao PDR, Cambodia, and Vietnam. The Srepok River, one of the main tributaries of the Mekong River, originates in the Central Highlands of Vietnam and passes through Cambodia before merging into the Mekong River, with a total length of 315 km. The Srepok basin lies between approximately 12° to 14°N and 106° to 109°E. The total basin area is 30,965 km2, of which 18,000 km2 lie in Vietnam, and the total population is around 1.6 million (2000). Most of the population resides in Vietnam, with the Cambodian side sparsely populated. The average altitude of the basin ranges from 150 m in the northwest to 1,000 m (MSL) in the southeast. The topographical characteristics directly affect the rainfall distribution with average annual rainfall ranges from 1,896 mm (1978–2007) at the Duc Xuyen station to 1,489 mm (2000–2003) at the Okrieng station downstream. More than 70% of the total annual rainfall occurs during the wet season (June to November), and about 41% is converted into runoff (Ty 2008). In addition, high immigration rates from other regions (more than 10%) and rapid local population growth (4–5%) upstream have driven an uncontrolled expansion in the area under cultivation (Hook et al. 2003). Forest cover in Dak Lak province—one of the eight provinces in the basin, declined 25% between 1975 and 2000 (Müller 2003). Consequently, water demand for agriculture has increased significantly and dominates total water demand. The Srepok basin, a representative example of a mountainous and transboundary basin in the LMB in terms of the hydrologic characteristics and water-related problems mentioned earlier, was selected for this study. The study area was divided into 13 sub-units, shown in Fig. 1. 2.2 Data The data required for this study was either collected from government organizations or downloaded from public domains. Land use/cover, soil type, and topography data were
Fig. 1 Study area and hydro-meteorological stations
1390
T.V. Ty et al.
obtained from the Mekong River Commission (MRC); the hydrometeorological data was collected from the Hydro-Meteorological Data Center in Vietnam (HMDC); gridded rainfall projection was downloaded from the SEA-START Regional Center website; and the population density data was downloaded from the Center for International Earth Science Information Network (CIESIN), Columbia University website. The collected data and their sources are summarized in Table 1.
3 Methodology First, a Hydrologic Modeling System (HEC-HMS) model was set up, calibrated, and validated to model the rainfall-runoff process in the basin. Second, the climate projections of SEA-START (ECHAM4, scenarios A2, B2) were downscaled to the six stations in the study basin, bias correction was carried out and climate change scenarios were developed; future land use was predicted using a GIS-based logistic regression approach; and the population projected from the historical data. Finally, these developed scenarios were input into the calibrated HEC-HMS model to examine the corresponding impacts. 3.1 Hydrological Model HEC-HMS was developed by the United States Army Corps of Engineers (USACE 2008) to simulate the rainfall-runoff processes of dendritic watershed systems. It was selected for this study due to its versatility, the suitable of the studied basin shape, the available data (which can be utilized to derive required input parameters) and the authors’ prior knowledge and familiarity with this model. HEC-HMS is divided into three models: the basin model, the meteorological model, and the control specifications model. It was first set up and data requirement for simulation were collected, and proceeded to estimate necessary input parameters. The model was calibrated, and validated to model the rainfall-runoff process in the study basin. It was then used to simulate the stream flow and to compute the water availability for each sub-basin. The studied basin was delineated into 13 sub-basins based on the Digital Elevation Model. However, only three sub-basins upstream, namely Krong No, Krong Ana, and Lower Srepok Vietnam, were considered for continuous hydrological simulation because of a lack of observed data on the Cambodian side (Fig. 1). Rainfall data from six stations (Da Table 1 Data and sources Data/Theme
Data type
Year
Data Source
Land use/cover
Polygon, 1:50,000
1993/1997
MRC
Soil map
Polygon; 1:50,000 (Vietnam)
2002
MRC
Digital Elevation Map
1:100,000 (Cambodia) 50 m resolution
2000
MRC
Gridded Population of the World: Future Estimates
2.5 arc-minutes resolution
2008
CIESIN
Gridded Rainfall Projection
20 km resolution
1978–2007
SEA-START RC
Rainfall
6 stations, daily basis
2035–2064 1978–2007
ECHAM4 (A2, B2) HMDC
Stream flow
2 stations, daily basis
1978–2007
HMDC
A Case Study in the Srepok River Basin, Vietnam—Cambodia
1391
Lat, Duc Xuyen, Cau 14, Krong Buk, Buon Ma Thuot (BMT), and Ban Don) were mapped to each sub-basin using the Thiessen polygon method. The water volume for runoff was calculated using the Soil Moisture Accounting (SMA) model recommended for continuous simulations. The Clark model was selected for surface runoff calculation based on available data/information and the baseflow was determined by the exponential recession model. The flow routing was calculated using the Muskingum model since it has the lowest data/ information requirements. In this study, the parameters of each sub-basin were estimated by combining the available data/information. For the SMA model, GIS data of topography, soil, and plant cover characteristics were used to estimate the associated parameters using the relationships proposed by García et al. (2008) (canopy storage, surface storage, maximum soil storage, tension zone storage, maximum infiltration, and soil percolation rate). The other parameters for the SMA model were solely estimated through model calibration. Two parameter values for the Clark model—time of concentration was estimated using Kirpich formula and storage coefficient was recommended by García et al. (2008), respectively. The parameters for the baseflow model were estimated based on the historical streamflow measurement and those for the Muskingum routing model were selected from the range recommended by USACE (2008). Streamflow was simulated on a daily basis and the results were calibrated to observed daily streamflow data for the period 1978–1992 at two stations (Cau 14 and Ban Don). The calibrated parameters were then used to validate the model for the period 1993–2007. The calibration and validation results were evaluated for goodness of fit using three criteria recommended by Krause et al. (2005) and Moriasi et al. (2007): Nash-Sutcliffe Efficiency (NSE) with logarithnic values of observed and simulated flow, percent bias (PBIAS), and ratio of root mean square error (RMSE) to standard deviation of observed data (STDEVobs) (RSR) at daily and monthly time scales. Calculations were made using the following formulae: 2P n 2 3 ln Yiobs ln Yisim 6 i¼1 7 7 NSE ¼ 1 6 n 4P 25 ln Yiobs ln Y mean
ð1Þ
i¼1
2P n
6 i¼1 PBIAS ¼ 6 4
3 Yisim Yiobs 100 7 7 n 5 P obs Yi
ð2Þ
i¼1
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi n 2 P Yiobs Yisim
RSR ¼
RMSE i¼1 ¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi STDEVobs n 2 P Yiobs Y mean
ð3Þ
i¼1
where Yiobs and Yisim are the ith observed and simulated flow, respectively; Ymean is the mean of observed flow data; and n is the total number of observations.
1392
T.V. Ty et al.
3.2 Land Use/Cover Change Although many land use/cover change models have been developed using the advantages of GIS (Verburg et al. 2002; Müller 2003), a simple approach—namely, a GIS-based logistic regression often used as a methodology in land use/cover change research (Müller 2003; Allen and Lu 2003; Müller and Munroe 2005, 2007; Huang et al. 2009)—was used in this study. Logistic regression is a form of regression used when the dependent variable is dichotomous and the independent variables are continuous or categorical. In this method, a maximum likelihood estimation is needed to transform the dependent variable into a logit variable (the natural log of the odds that the dependent variable occurs or not). The output from a logistic regression is the probability of occurrence of an event, using the independent variables as predictor values (Garson 2000). In land use/cover change study, it means that the occurrence of a certain land use type will be predicted by several independent values, also called the driving factors. First, available land use types were reclassified into five principal land use types: thick forest, thin forest, grassland, agriculture, and urban. The relationship between the spatial distribution of each land use type and its driving factors was then determined using logistic regression analysis, and probability maps were produced, accordingly. Land use types were then predicted considering the produced probability maps, actual land use maps, conversion elasticity, and future demands for different land use types. To predict future land use/cover, new probability maps are calculated with updated values of the driving factor that changes in time (population), with other factors assumed to be constant in time. 3.2.1 Driving Factors The influences of different driving factors on land use/cover were calculated using a logistic regression analysis. These factors were derived from available data including site specific characteristics: socioeconomic (population density) and biophysical (elevation, slope, soil type); and proximity (distance to provincial road, distance to river, and distance to town). All of the driving factors and land use type data were converted into grid cells with a resolution of 1 km. a)
Regression coefficients
The probability is a function of the logit coefficients (regression coefficients), also called maximum likelihood estimates (MLE) that are belonging to the independent variables. The regression coefficients can only be determined by an iterative process of estimation (SPSS software). The MLE tries to maximize the odds that the observed values of the dependent variables may be predicted from the observed values of the independent variables (Garson 2000). For each land use type u, the regression coefficients were determined from a fitted logistic regression function with n independent variables (driving factors) (Verburg et al. 2002), and the results of regression coefficients are presented in Table 2. Pu ¼ b 0;u þ b1;u X1 þ b 2;u X2 þ . . . þ b n;u Xn ln 1 Pu
ð4Þ
where Pu is the probability of a considered land use type u (calculated from observed land use map); Xn is the nth driving factor; β0,u is the intercept of the regression model; and β1,u,… βn,u is the regression coefficient of the 1st, … nth driving factor of land use type u.
A Case Study in the Srepok River Basin, Vietnam—Cambodia
1393
Table 2 Results of logistic regression analysis for land use types Factors
Thick forest
Thin forest
Urban
0.001
−0.001
–
−0.001
–
−0.001
–
0.002
–
0.039
0.018
–
−0.038
–
–
Slope Soil type Acrisol
Agriculture
−0.006
Elevation Population density
Grassland
–
–
–
Andosol
−0.545 –
−0.270
–
–
–
Cambisol
–
−0.347
0.423
–
–
−1.307
Ferralsol
0.280
–
−0.522
– –
Fluvisol
–
–
0.327
−0.680
Gleysol
–
–
−0.186
0.607
−6.321
0.259
−3.673
−6.479
−14.829
0.878
0.652
0.602
0.691
0.905
Constant ROC
–
– Not significant at 95% level
b) Probability of occurrence maps The probability maps for each land use type u, in a grid cell i were calculated as follows (Verburg et al. 2002): ln
Pi;u 1 Pi;u
¼ b0;u þ b 1;u X1;i þ b 2;u X2;i þ . . . þ bn;u Xn;i
ð5Þ
Setting: Zu ¼ b 0;u þ b 1;u X1;i þ b 2;u X2;i þ . . . þ bn;u Xn;i Pi;u ¼
eZu 1 þ eZ u
ð6Þ
where i is the ith grid cell; Pi,u is the probability of a considered land use type u, occurring in a grid cell i; Xn,i is the driving factor; and n is the number of driving factors. Table 2 lists the estimated coefficients (only statistically significant factors are shown) and relative operation characteristics (ROC) values for all land use types that are used to measure the goodness of fit of a logistic regression approach. As can be seen in Table 2, no factor significant at the 95% level is found for urban areas. This is understandable as the study area is predominantly rural. It is generally acknowledged that a relationship between population and land use/cover change does exist, but the causality has not been fully understood and the chicken-egg dilemma persists. To reflect urban development in 2050, the relationship between population density and urban land use type was developed in this study with the assumption that urban area development is highly dependent on population growth. Setting an urban population density of more than 800 per km2 (Kawasaki et al. 2010), approximately 75% of the urban areas in the observed land use/cover in 1997 were found to have this density. The proposed population density was then applied to the new population density grids in 2050 to predict future urban areas.
1394
T.V. Ty et al.
3.2.2 Demand Future land use types (demand) are non-spatial input data, which can be defined with the aid of economic models, future policy projections, or with a simple linear extrapolation of the growth in area under the land use/covers. One widely adopted method is based on population predictions assuming that the land use area per capita remains constant. In this study, it is estimated based on the growth ratio (R) defined by a ratio of land use growth rate (%) to population growth rate (%) and addressed as below (Campbell et al. 2007): R¼
ΔA ΔP
P P0 A ¼ A0 1 þ R P0
ð7Þ
ð8Þ
where R is the growth ratio; ΔA is growth rate of considered land use type between 1993 and 1997 (%); ΔP is population growth rate (%); A and A0 are future and current area of considered land use type (km2), respectively; and P and P0 are future and current population, respectively. 3.2.3 Conversion Elasticity and Protected Area Conversion elasticity is used to indicate how readily a certain land use type is allowed to change, as defined by policies or other decision rules. It varies between 0 and 1, with 0 for the most dynamic land use types and 1 for the most stable ones. Based on previous studies on land use/cover change in Vietnam (Willemen 2002; Müller and Munroe 2005), the following elasticity values were assigned to the five reclassified land types: namely 1.0 to thick forest, 0.6 to thin forest, 0.2 to grass, 0.8 to agriculture and 1.0 to urban. These values were then added to the probability values to obtain a total probability and were also used as a parameter for calibration. No land use changes were allowed for protected natural areas, mostly on the Cambodian side, such as national parks, protected landscapes, and wildlife reserves, as identified by ICEM (2003a, b) and mapped by Hook et al. (2003). The existence of forest reserves following government guidelines that discouraged agricultural production on steep slopes was also taken into account. It was empirically assumed that all primary forest on slopes greater than 15° was reserved (Müller and Munroe 2005). 3.2.4 Procedure (Verburg et al. 2002) 1. For all grid cells i the total probability (Pi,u) is calculated for each of the land use types u: 1 2 Pi;u ¼ Pi;u þ Pi;u
ð9Þ
1 is the probability at where Pi,u is the total probability at grid cell i of land use type u; Pi;u 2 is the grid cell i of land use type u calculated from logistic regression analysis; Pi;u conversion elasticity specified in the decision rules and is only given a value if grid cell i is already under land use type u.
A Case Study in the Srepok River Basin, Vietnam—Cambodia
1395
2. Each candidate cell is assigned to the land use based on the total probability values excluding protected and reserved cells, starting from the highest probability and working down, until the total area of a certain land use type is equal to its desired future area. Land use types were calibrated against a land use map in 1993 and validated to a land use map in 1997. The results were evaluated by calculating cross-tabulation. From the generated land use maps, the parameters of the runoff process of each sub-basin associated with land use/cover as mentioned earlier were recalculated and then input into the calibrated hydrological model for future simulations. 3.3 Water Use/Demand Estimation Water diverted to meet domestic and industrial demand was estimated as the product of population and per capita water demand, which reflects different levels of access of the population to clean water. Industry has not developed in this area. Hence, unit domestic demand and industrial demand are considered together and are currently taken as 32 and 67 L/cap/day for the Cambodian and Vietnamese sides, respectively. Projected per capita water demand was taken as 150 L/cap/day for the basin in 2050 (MRC 2004). Water use/demand in the irrigation sector was estimated on the basis of the irrigated area, crop water requirements (CWR), effective rainfall, and water losses. Current irrigation water use was calculated for each sub-basins by Ty et al. (2010). Due to a lack of data for the required climatic inputs, the FAO Penman-Monteith equation of estimating ETo with limited data was used. The method estimates the other climatic variables (solar radiation, wind speed, relative humidity, and sunshine) based on downscaled maximum and minimum temperature and from geographical locations (Allen et al. 1998). Future irrigation water demand was estimated on a monthly basis for each sub-basin based on projected land use maps (agricultural area), future CWR calculated from potential evapo-transpiration (ETo), and effective rainfall, which was estimated using the USDA Soil Conservation Service formula developed by USCS (CROPWAT 8.0). 3.4 Water Availability and Water Stress Index The streamflow (regulated flow) at the outlet of the three gauged sub-basins was used for disaggregation to estimate the regulated flow of the remaining ungauged sub-basins based simply on the area proportion. Although this approach may not be accurate, it was recommended for estimating water availability in water resources planning and management in scarcely gauged basins (Amarasinghe et al. 1999). Because information/data on groundwater was very sparse and unreliable, groundwater availability was not considered in this study. As regulated flow (simulated flow) is in fact post-depletion flows, water availability (unregulated flow) was augmented by consumptive uses. Therefore, withdrawals were added and return flows subtracted. Water availability (unregulated flow) was expressed by the following formula: WA ¼ SF þ WW RF
ð10Þ
where WA is water availability (unregulated flow) (m3); SF is regulated flow (m3); WW is water withdrawals for different uses (m3); and RF is return flows from uses (27% for irrigation use and 35% for domestic and industrial uses (MRC 2004)). The water stress index (WSI) (Sun et al. 2008) was used to examine the water situation in the study basin. The index was initially developed to examine “water stress” on the national
1396
T.V. Ty et al.
or regional scale; in this study, however, it was applied at the basin and sub-basin levels to evaluate the water stress situation. The index quantitatively assesses the relative magnitude of water demand and water availability. If water demand exceeds 40% of available water resources, an area is considered to be experiencing ‘severe’ water stress. An area with withdrawals from 20% to 40% is considered to have ‘medium to severe’ water stress; from 10% to 20% is considered to represent ‘moderate’ water stress; and areas with less than 10% are considered to have ‘little or no’ water stress. WSI ¼
WD WA
ð11Þ
where WSI is water stress index; WD is total water withrawal for the domestic, industrial and agricultural sectors; and WA is water availability. 3.5 Scenario Development 3.5.1 Climate Change Scenario a)
Model Selection
This study utilized the output from a high-resolution Regional Climate Model (RCM) (20 km grid) developed by SEA-START Regional Center for the period of 2010–2099, using the period of 1960–1999 as a baseline. The simulation is based simulation by PRECIS (Providing Regional Climates for Impacts Studies) RCM using the GCM–ECHAM4 dataset as initial data for calculations covering IPCC emission scenarios A2 and B2. These two emission scenarios are commonly used in climate change impact studies because they allow the investigation of the entire range of potential system response to climate change. Details of climatic variables downloaded from SEA-START Regional Center are available at http://www.start.or.th/. b) Bias Correction Since there is much uncertainty because adaptive capacity to climate change might alter emission scenarios, it should be noted that the selected scenarios (A2 and B2) for this study are only two plausible descriptions of how future emissions might develop and are not any more likely than are any other scenarios (Roosmalen et al. 2007). Rainfall output from RCM cannot be used directly as input for hydrological simulation due to biases between the simulated variables for the current (control) climate and observed values. The delta change approach was used to correct these biases by transferring the signals of climate change derived from a climate model simulation to observed database (Hay et al. 2000). Although this method was found to correct both rainfall intensity and frequency distribution inadequately (Ines and Hansen 2006), it was considered a better bias correction of rainfall on monthly and seasonal basis. In this study, the means were calculated on a monthly basis for each 30-year period of climate output. The 12 delta change factors for rainfall were used to perturb the observed database and calculated as follows: ΔP ðjÞ ¼
PΔ ði; jÞ ¼ ΔP ðjÞ Pobs ði; jÞ
Pscen ðjÞ Pcontr ðjÞ ði ¼ 1 31; j ¼ 1 12Þ
ð12Þ
ð13Þ
A Case Study in the Srepok River Basin, Vietnam—Cambodia
1397
where Δp is the delta change factor; Pcontr ðjÞ and Pscen ðjÞare the rainfall in month j averaged for the 30-year control (1978–2007) and scenario (2035–2064) periods centered in 2050 simulated by the RCM, respectively; PΔ(i, j) is rainfall input into the hydrological model for the A2 and B2 scenario runs; Pobs(i, j) is the observed rainfall representing current climate; suffixes i and j stand for the ith day and the jth month. For temperature, absolute change is used for the delta change factors, as follows: ΔT ðjÞ ¼ T scen ðjÞ T contr ðjÞ
TΔ ði; jÞ ¼ Tobs ði; jÞ þ ΔP ðjÞ
ði ¼ 1 31; j ¼ 1 12Þ
ð14Þ
ð15Þ
where ΔT is the delta change factor; T contr ðjÞand T scen ðjÞare the temperature in month j averaged for the 30-year control (1978–2007) and scenario (2035–2064) periods centered in 2050 simulated by the RCM, respectively; TΔ(i, j) is temperature input into the hydrological model for the A2 and B2 scenario runs; Tobs(i, j) is the observed rainfall representing current climate; suffixes i and j stand for the ith day and the jth month. For reference evapo-transpiration (ETref), the delta change factors were calculated in the same way as for the rainfall factors. ETref was calculated using the FAO Penman-Monteith equation (Allen et al. 1998) from RCM output. The described reference evapo-transpiration is the potential evapo-transpiration for a hypothetical grass reference crop with an assumed crop height of 0.12 m, a fixed surface resistance of 70 s/m, and an albedo of 0.23 (Allen et al. 1998). The reference evapo-transpiration can be converted to potential evapo-transpiration by multiplying it with a surface coefficient (crop coefficient). However, in this study, it is not necessary to include the crop coefficient because the relative change in evapo-transpiration between the current and future climate was used. 3.5.2 Land Use Change Prediction After the calibration and validation, a land use change approach was executed for the year 2050, differentiated by the various land use demands and their spatial distribution compared to the year 1997 (baseline). Future demand for land use was estimated assuming that a moderate level of development—current levels of agricultural production—was maintained. It was projected using the growth ratio of the historic trend of land use (1993–1997) to the population trend (1990–2000), as mentioned earlier. 3.5.3 Population Projection Many studies have predicted the future population of the MRB. Kristensen (2001) predicted that the population would increase to 120 million by 2025; while Hoanh et al. (2003) forecast that the population would reach 132 million by 2030. Pech and Sunada (2008) noted the uncertainty in population growth and expected that the population would at least double by 2050. The uncertainties in the population estimates for the MRB are unavoidable, given that the basin covers six countries and given the consistency in census data collection and frequency in different administrative units. In this study, the population in the Srepok basin was quantified for the year 2000 using a gridded population of the world obtained from CIESIN (2008) and validated by Eastham et al. (2008), with the population estimates proved to be close to the published population projections for the basin (MRC 2003). The future population was estimated using the past population growth rate (1990–2010) as the basis,
1398
T.V. Ty et al.
taking into account the different growth rates in each population category (country and rural or urban) for each sub-basin. 3.5.4 Simulation Scenarios Three scenarios were developed to examine historical and future water resources under average historical or projected changes in climate, and land use/cover and population by individual factor or combined factors, as presented in Table 3. Baseline scenario represented the average historical climate (1978–2007), population (2000), and land use/cover conditions (1997) across the 13 sub-basins of the Srepok basin. Water availability and demand calculated in this scenario served as the baseline for comparisons across the changes.
4 Results and Discussion 4.1 Calibration and Validation 4.1.1 Hydrological Model Looking at the qualitative performance of streamflow, the shape and character of the simulated flow fits quite well with observations, as can be seen in Fig. 2, which shows streamflow at Ban Don station. Although the peak flows are under- or overestimated, this is not considered to be a serious problem, since the objective of this study is to assess the mean flows and low flows, which are well fitted to observed low flows at all stations. Table 4 summarizes the results of the model calibration and validation performance on daily and monthly time scales. It reveals fairly good agreement between simulated and observed flow according to the three criteria (NSN, PBIAS, and RSR). As can be seen in Table 4, the model performance improves as the catchment area is increased. This may be explained by compensation for bias among individual sub-basins. It is clear that the calibrated HECHMS model simulates the runoff with reasonable quality, so it can be used to evaluate the effects of different scenarios on streamflow in the study area. 4.1.2 Land Use Change Prediction The results of calibration and validation were evaluated as a percentage of correction measured by locations predicted correctly, as presented in Table 5. The overall percentage of correction is of 85.7% and 82.8% for calibration and validation, respectively. The percentage of correction of land use types ranging from 69.6% to 93.5% suggests that this approach has the capacity to explain the spatial variation of land use patterns (Pontius and Schneider 2001).
Table 3 Combined simulation scenarios Scenario
Climate
Population
Land use
Baseline
1978–2007
2000
1997
Sc1: Climate change
ECHAM4(A2, B2)
2050
1997
Sc2: Land use/cover change and population
1978–2007
2050
2050
Sc3: Climate, land use/cover and population changes
ECHAM4 (A2, B2)
2050
2050
A Case Study in the Srepok River Basin, Vietnam—Cambodia
1399
Fig. 2 Monthly observed and simulated flow at Ban Don station. a Calibration. b Validation
Existing land use/cover in 1997 and predicted land use/cover in 2050 are shown in Fig. 3. As the figure shows, most of the hotspots (areas with change) for future land use types occur in the upstream area of the basin in Vietnam. This principally reflects the rapid expansion of agriculture to keep pace with population growth. For the entire basin, the area under cultivation is expected to rise from 15.3% in 1997 to 28.1% in 2050. This is understandable because the population in the basin relies heavily on agriculture-based subsistence. The increased agricultural area from this study is less than that from Kawasaki et al. (2010), in which agricultural area is projected to increase from 22% in 2000 to 39% in 2050. This difference is understandable because the previous study predicted future land use change based entirely on the relationship between land use types and population density. The urban area is also predicted to increase, from less than 0.03% in 1997 to 2.2% in 2050; while thin forest area is likely to decline from 49.0% to 34.2% between 1997 and 2050. A crosstabulation of the current and predicted land use/cover maps reveals that significant percentages
1400
T.V. Ty et al.
Table 4 Model calibration and validation performances Cau 14
Ban Don
NSE
PBIAS
RSR
NSE
PBIAS
RSR
Daily
0.76
9.0
0.57
0.79
9.2
0.55
Monthly
0.78
0.55
0.81
Calibration period 0.52
Validation period Daily
0.71
Monthly
0.73
Whole period Daily Monthly
0.74
−14.2
−3.4
0.75
0.68
0.72
0.66
0.75
0.65
0.77
0.62
0.78
−12.1
0.62 0.60
−1.9
0.59 0.57
NSE, PBIAS and RSR are Nash-Sutcliffe Efficiency with lograrithnic values of observed and simulated flow, percent bias and ratio of root mean square error to standard deviation of measured data, respectively
of future agriculture will be on land converted from thin forest (42%) and grassland (16%). Spatial analysis of the patterns of land use/cover in combination with the probability of change maps makes it possible to determine where those changes are most likely to occur in the future. In this study, spatial distributions of future land use types are predicted based mostly on the bio-physical driving factors (Table 2). In reality, the impacts of socio-economic factors such as land use planning at province or district level on the changes of land use/cover are also very important. Future land use types (demand) was estimated assuming the constant land use area per capita although uncontrolled immigration was found to be strongest during the 1990s from Central and North Vietnam to the Central Highlands (Müller and Zeller 2002). Therefore, the uncertainty in the projection of population is inevitable. In addition, Table 5 Results of calibration and validation matrix Observed land use
Predicted land use
% correct
Thick forest
Thin forest
Grassland
Agriculture
Total
Thick forest
557
0
0
0
557
84.9
Thin forest
6
14017
1184
0
15207
88.4
Grassland
80
1638
8057
153
9928
78.3
Agriculture
0
0
131
4260
4391
93.5
Total
643
15655
9372
4413
30083
85.7
Thick forest Thin forest
524 8
0 13660
0 1206
0 0
524 14874
85.5 88.2
Grassland
73
2672
7264
15
10024
69.6
Agriculture
0
0
142
4519
4661
93.4
Total
605
16332
8612
4534
30083
82.8
Calibration, 1993
Validation, 1997
Land use types with highest produced probability as predicted land use types; bold values along the diagonals indicate correct predictions; and values correspond to the number of pixels (1 km×1 km)
A Case Study in the Srepok River Basin, Vietnam—Cambodia
(a) Land use/cover 1997
1401
(b) Land use/cover 2050
Fig. 3 Existing and predicted land use/cover maps. a Land use/cover 1997. b Land use/cover 2050
the National Five Million Hectare Reforestation Program (5MHRP) which have been implemented in Vietnam since 1999 was not considered in this study (GOV 1998). Taking all mentioned affecting factors into account, the reliability of land use/cover change prediction results would be improved. 4.2 Climate Change Delta changes in the monthly rainfall extracted directly from the grid cells of the ECHAM4, emission scenarios A2 and B2 nearest the stations in the study area were used to generate the daily rainfall series, which were then input into the calibrated hydrological model (HEC-HMS) for simulation. The results of downscaled rainfall revealed that future rainfall in the studied basin would increase 4.1% and 5.3% in the wet season and decrease 13.4% and 15.8% in the dry season under scenarios A2 and B2, respectively. The maximum temperature was projected to increase 1.6°C and 1.4°C in the wet season and increase 1.4°C and 1.3°C in the dry season under scenarios A2 and B2, respectively. Consequently, potential evapo-transpiration (ETo) will increase 3.6% and 3.4% in the wet season, and increase 4.0% in the dry season under scenarios A2 and B2, respectively. Details on the monthly delta change factors of rainfall (%) and ETo (%) and Tmax (°C) of the scenario period (2035–2064) compared with the control period (1978–2007) under scenarios A2 and B2 are shown in Figs. 4 and 5, respectively. 4.3 Water Stress Situation Under Different Developed Scenarios 4.3.1 Baseline Scenario There is broad variation in rainfall across the basin, with higher rainfall found at higher elevations. Agriculture has been developed in upstream Vietnam, where most of the population lives, and irrigation is the dominant use of water (accounting for more than 80% of water used). Consequently, this area has been identified as a ‘hotspot’ in terms of a high WSI level in the dry season, although at the basin scale, water stress levels are considered to be low. ‘Medium to severe’ level of water stress, for example, can be found in the Krong Ana and Lower Srepok Vietnam sub-basins in the dry season. This indicates that assessing water stress at higher spatial and temporal scales may conceal water stress at lower scales because of large variations in water availability and demand.
1402
T.V. Ty et al.
Fig. 4 Monthly changes in rainfall (%) in emission scenarios A2 and B2 (2035–2064) compared with control period (1978–2007)
The overall impact of climate and land use/cover changes on future basin water availability (WA) (%) and WSI (%) are shown in Figs. 6 and 7, respectively; and sub-basin WSI in the dry season under baseline, scenario 1 (A2), scenario 2 (LUCC) and scenario 3 (A2) are presented in Fig. 8a, b, c and d, respectively. 4.3.2 Scenario 1 The variation in projected rainfall in 2050 compared to historical data (1978–2007) shows increasing stream-flow, and thus water availability, during the wet season (9.0% and 24.7%) and decreasing water availability during the dry season (20.6% and 11.7%) under scenarios A2 and B2, respectively as can be seen in Fig. 6. Consequently, across the studied basin, basin water stress is projected to decrease by 1.6% and 19.8% in the wet season and increase by 67.4% and 60.6% in the dry season under scenarios A2 and B2, respectively (Fig. 7). Annual water resources are found to increase by 1.2% and 15.1% in 2050 under scenarios A2 and B2. According to the results from Kawasaki et al. (2010), the annual discharge at the outlet will incease by 3% and 6% in 2025 and 2050, respectively. The difference is understandable because the previous study assumed annual increase of rainfall by 2.5% and 5% for the year 2025 and 2050, respectively; and these increases were uniformly applied to the daily observed 2001 rainfall data. 4.3.3 Scenario 2 Water demand for domestic and industrial sectors is directly related to population growth and an increase in water demand will result in slightly greater water stress. Change in land Fig. 5 Monthly changes in ETo (%) and Tmax (°C) in emission scenarios A2 and B2 (2035–2064) compared with control period (1978–2007)
A Case Study in the Srepok River Basin, Vietnam—Cambodia
1403
Fig. 6 Basin water availability (WA) changes (%) under different scenarios
use/cover directly affects water availability by changing the hydrological processes, and thus water stress. Land use/cover change also affects water demand. For instance, expansion of the agricultural area increases irrigation water demand. Although land use/cover change causes increasing surface runoff (1.3%) due to deforestation and urbanization, this is outweighed by the increase in irrigation water demand. In addition, intensive agricultural development upstream, an area that has recently experienced water stress, will create an even more severe situation if the area under cultivation almost doubles by 2050. Across the basin, water stress is projected to increase more than 100% by 2050 (Fig. 7). Most subbasins in the upstream will face increasing water stress in the future, particularly during the dry season (Fig. 8). 4.3.4 Scenario 3 The individual impacts of climate, population, and land use/cover changes on water availability and demand should be examined at both the basin and sub-basin scales. In reality, however, these changes occur simultaneously. The results from the combination of these three changes reflect the summation of positive and negative impacts on water availability and demand. Climate change increases future water availability while population growth increases water demand. As noted earlier, land use/cover change produces a minor increase in surface runoff, but generates a much greater increase in water demand (irrigation crops). The combination of these three factors causes an overall increase in water stress at the basin and sub-basin scales, especially during the dry season (Fig. 7), with the higher water stress level mostly attributable to land use/cover change. Increased future rainfall is not able to Fig. 7 Basin WSI changes (%) under different scenarios
1404
T.V. Ty et al.
(a) Baseline
(c) Scenario 2 (LUCC)
(b) Scenario 1 (A2)
(d) Scenario 3 (A2)
Fig. 8 Water Stress Index (WSI) in the dry season under different scenarios. a Baseline. b Scenario 1 (A2). c Scenario 2 (LUCC). d Scenario 3 (A2)
offset the greater water stress caused by the growing population and agriculture development. In addition, surplus water found in most sub-basins during the wet season can be stored and utilized for different purposes during the dry season. Although the basin does have extensive water infrastructure, the uneven spatial distribution among districts makes the current percentage of water withdrawals low compared to available resources. The construction of water infrastructure such as large dams (for both hydropower and water storage purposes) and small reservoirs should be considered as a means of avoiding future water shortages. Further water resources development, however, would require additional consideration given to the availability of a suitable site and an assessment of the social and environmental impacts. Because land use/cover changes are expected to be the single biggest driver of greater water demand in the next 50 years, land use development policies will need to be adjusted. That leads to changes in crop varieties, planting dates, and crop patterns, and thus irrigation schedules. However, the economic incentives of agricultural development, especially coffee and other crops with high economic value and market demand, make policy changes difficult to implement. This suggests the importance of encouraging better demand management strategies, such as water conservation, and introducing new and efficient technologies such as drip irrigation and reuse.
A Case Study in the Srepok River Basin, Vietnam—Cambodia
1405
5 Conclusions Based on the results, the major findings of this study can be summarized as follows: 1. The area under cultivation is predicted to increase from 15.3% in 1997 to 28.1% in 2050, while the urban area is forecast to increase to 2.2% in 2050. In contrast, thin forest area is likely to decrease from 49.0% to 34.2% between 1997 and 2050. A large percentage of future agriculture will be on land converted from thin forest (42%) and grassland (16%). 2. ‘Medium to severe’ water stress level has occurred in the upstream sub-basins during the dry season. Under the climate change scenario, the water stress level is projected to increase during the dry season. Land use/cover change produces a minor increase in surface runoff, but a much greater increase in water demand. 3. Of the two factors, land use/cover change is found to have the greatest impact on water stress while climate change causes a slight increase in future water stress. The combined scenario shows a rising level of water stress in the future at both the basin and sub-basin scales, especially during the dry season. Higher rainfall in the future will not compensate for the increased water stress associated with a growing population and expanding agriculture. 4. Assessing water resources at the basin scale may hide water stress at lower scales due to variations in water availability and demand. This suggests that assessments of water resources should be done on the lowest possible spatial and temporal scales. In addition, surplus water from the wet season can be utilized during the dry season. Further water resources development, however, would require additional considerations such as the availability of suitable sites and an assessment of the social and environmental impacts. In addition, improving demand management, such as water conservation, and introducing new and efficient technologies such as drip irrigation and reuse should be encouraged. 5. The results from a simple GIS-based logistic regression approach suggest that this approach is capable of predicting future land use/cover change. However, the reliability of the data used to derive the driving factors should be taken in account, because these data are the basis for the statistical relation between driving factors and land use types, and for the probability maps for future land use/cover predictions. With detailed land use/cover data in the future, especially data on agricultural land use types such as rice, coffee and others, the reliability of the results would be improved. The impacts of land use/cover change on water quality limit the availability of water resources. For this reason, water quality, environmental flow, and an adjustment of land development policies should be considered in future research. Most sub-basins assessed receive water from upstream sub-basins, and this should also be considered in the water availability term. Acknowledgements The authors express their sincere thanks to the Global COE (Center of Excellence) Program of the University of Yamanashi for supporting this study.
References Abdulla F, Eshtawi T, Assaf H (2009) Assessment of the impact of potential climate change on the water balance of a semi-arid watershed. Water Resour Manag 23:2051–2068 Allen J, Lu K (2003) Modeling and prediction of future urban growth in the Charleston region of South Carolina: a GIS-based integrated approach. Conserv Ecol 8(2):2. Available from http://www.consecol. org/vol8/iss2/art2/ [Assessed November 2011]
1406
T.V. Ty et al.
Allen RG, Pereira LS, Raes D, Smith M (1998) Crop evapotranspiration: guidelines for computing crop water requirements. Irrigation and Drainage Paper 56. UN-FAO, Rome, Italy Amarasinghe UA, Mutuwatta L, Sakthivadivel R (1999) Water scarcity variations within a country: a case study of Sri Lanka. International Water Management Institute, Colombo, Research report 32 Campbell CE, Allen J, Lu KS (2007) Modeling growth and predicting future developed land in the Upstate of South Carolia. Strom Thurmond Institute, Clemson University Chinvanno S (2004) Information for sustainable development in light of climate change in Mekong River Basin. Southeast Asia START Regional Center, Bangkok CIESIN (2008) Gridded population of the world. Center for International Earth Science Information Network (CIESIN), Columbia University. Available from: http://sedac.ciesin.columbia.edu/gpw/global.jsp [Assessed October 2009] Eastham J, Mpelasoka F, Mainuddin M, Ticehurst C, Dyce P, Hodgson G, Ali R, Kirby M (2008) Mekong River basin water resources assessment: impacts of climate change. Water for a Healthy Country National Research Flagship Elias GB, Vernon HK (2010) Watershed modeling to assessing impacts of potential climate change on water supply availability. Water Resour Manag 24:3299–3320 Feng X, Zhang G, Yin X (2011) Hydrological responses to climate change in Nenjiang River Basin, Northeastern China. Water Resour Manag 25:677–689 García A, Sainz A, Revilla JA, Álvarez C, Juanes JA, Puente A (2008) Surface water resources assessment in scarcely gauged basins in the north of Spain. J Hydrol 365:312–326 Garson GB (2000) PA 765 Statnotes: an online textbook. http://faculty.chass.ncsu.edu/garson/PA765/logistic. htm [Assessed May 2011] GOV (1998) Decision No. 661/QD-TTg of the Prime Minister: National Five Million Hectare Reforestation Programme (5MHRP). The Government of Vietnam (GOV). Hanoi, Vietnam Grau HR, Aide TM, Zimmerman JK, Thomlinson JR, Helmer E, Zou X (2003) The ecological consequences of socioeconomic and land-use changes in postagriculture Puerto Rico. BioScience 53:1159–1168 Hay LE, Wilby RL, Leavesley GH (2000) A comparison of delta change and downscaled GCM scenarios for three mountainous basins in the United States. J Am Water Resour Assoc 36(2):387–397 Hoanh CT, Guttman H, Droogers P, Aerts J (2003) Water, climate, food and environment in the Mekong Basin in Southeast Asia. International Water Management Institute (IWMI), Mekong River Commission Secretariat (MRCS) and Institute of Environmental Studies (IVM) Hook J, Novak S, Johnston R (2003) Social atlas of the Lower Mekong basin. Mekong River Commission, Phnom Penh Huang B, Zhang L, Wu B (2009) Spatio-temporal analysis of rural-urban land conversion. Int J Geogr Inf Sci 23:379–398 ICEM (2003a) Cambodia national report on protected areas and development: review of protected areas and development in the Lower Mekong River region. Queensland, Australia: International center for Environmental Management (ICEM) ICEM (2003b) Vietnam national report on protected areas and development: review of protected areas and development in the Lower Mekong River region. Queensland, Australia: International center for Environmental Management (ICEM) Ines AVM, Hansen JW (2006) Bias correction of daily GCM rainfall for crop simulation studies. Agric For Meteorol 138:44–53 Kawasaki A, Takamatsu M, He J, Rogers P, Herath S (2010) An integrated approach to evaluate potential impact of precipitation and land-use change on streamflow in Srepok River Basin. Theory Appl GIS 18 (2):9–20 Kiem AS, Ishidaira H, Hapuarachchi HP, Zho MC, Hirabayashi Y, Takeuchi K (2008) Future hydroclimatology of the Mekong River basin simulated using the high resolution Japan Meteorological Agency (JMA) AGCM. Hydrol Processes 22:1382–1394 Krause P, Boyle DP, Base F (2005) Comparison of different efficiency criteria for hydrological model assessment. Adv Geosci 5:89–97 Kristensen J (2001) Food security and development in the Lower Mekong River basin: a challenge for the Mekong River Commission. Proceedings of Asia and Pacific Forum on Poverty: Reforming Policies and Institutions for Poverty Reduction. The Asian Development Bank, Manila Mizyed N (2009) Impacts of climate change on water resources availability and agricultural water demand in the West Bank. Water Resour Manag 23:2015–2029 Moriasi DN, Arnold JG, Van Liew MW, Bingner RL, Harmel RD, Veith TL (2007) Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Trans ASABE 50 (3):885–900 MRC (2003) State of the basin report 2003. Mekong River Commission, Phnom Penh
A Case Study in the Srepok River Basin, Vietnam—Cambodia
1407
MRC (2004) Modeled observations on development scenarios in the Lower Mekong Basin. Mekong River Commission, Phnom Penh Müller D (2003) Land-use change in the Central Highlands of Vietnam. Dissertation, Georg-August University Gottingen Müller D, Munroe DK (2005) Tradeoffs between rural development policies and forest protection: spatiallyexplicit modeling in the Central Highlands of Vietnam. Land Econ 81(3):412–425 Müller D, Munroe DK (2007) Issues in spatially explicit statistical land-use/cover change (LUCC) models: examples from western Honduras and the Central Highlands of Vietnam. Land Use Policy 24:521–530 Müller D, Zeller M (2002) Land use dynamics in the central highlands of Vietnam: a spatial model combining village survey data with satellite imagery interpretation. Agric Econ 27:333–354 Oki T, Kanae S (2006) Global hydrologic cycle and world water resources. Science 313(5790):1068–1072 Pech S, Sunada K (2008) Population growth and natural resources pressures in the Mekong River basin. J Hum Environ 37(3):219–224 Pontius RG, Schneider LC (2001) Land-cover change model validation by an ROC method for the Ipswhich watershed, Massachusetts, USA. Agric Ecosyst Environ 85:239–248 Roosmalen LV, Christensen BSB, Sonnenborg TO (2007) Regional differences in climate change inpacts on groundwater and stream discharge in Denmark. Vadosezone J 6(3):554–571 Sun G, McNulty SG, Moore Myers JA, Cohen EC (2008) Impacts of multiple stresses on water demand and supply across the southeastern United States. J Am Water Resour Assoc 44(6):1441–1457 Ty TV (2008) GIS application on water infrastructure inventory and accessibility of water resources in the Upper Srepok Basin, Central Highlands, Vietnam. Master Thesis, School of Engineering and Technology, Asian Institute of Technology, Thailand Ty TV, Sunada K, Ichikawa Y, Oishi S (2010) Evaluation of the state of water resources using Modified Water Poverty Index: a case study in the Srepok River basin, Vietnam—Cambodia. Int J River Basin Manag 8 (3–4):305–317 USACE (2008) Hydrologic Modeling System—HEC-HMS: Technical Reference Manual. US Army Corps of Engineers, Washinton DC Verburg PH, Soepboer W, Veldkamp A, Limpiada R, Espaldon V, Mastura SSA (2002) Modeling the spatial dynamics of regional land use: the CLUE-S model. Environ Manag 30(3):391–405 Vorosmarty CJ, Green P, Salisbury J, Lammers RB (2000) Global water resources: vulnerability from climate change and population growth. Science 289(5477):284–288 Willemen L (2002) Modelling of land cover changes with CLUE-S in Bac Kan province, Vietnam. Wageningen University, The Netherlands
