Climate change impact assessment on Zhoshui River water supply in Taiwan

This study evaluates the impact of climate change on water resources. An integrated procedure is proposed for assessing the water resources system response to climate change on the basin scale. The Zhoshui River basin in Central Taiwan was selected for the impact assessment. Five downscaled general circulation models based on the A1B scenario for 2046 2065 were adopted to assess the climate change impact, including (1) the irrigation water requirement downstream of the basin, (2) the river flow upstream of the basin, and (3) the water resources utilization related to supply and demand in the basin. Rising temperatures will cause the irrigation water requirement to increase by 10%. Precipitation in the basin will substantially decrease and likely cause a 20% decrease in river flow. Thus, irrigation water shortages may become more severe in the future. As an adaptation, the Hushan Reservoir, which will begin operation in mid-2016, can assist in offsetting domestic and industrial demand. To maintain the irrigation deficit at the present level (2001 2010) in the future, conveyance losses should reduce to 30% and the farming area used in the second paddy growth season should be decreased by 10%. Article history: Received 17 May 2016 Revised 23 September 2016 Accepted 26 October 2016


INTRODUCTION
The Intergovernmental Panel on Climate Change (IPCC) has stated that the observed increase in global average temperature since the mid-20th century is very likely due to the observed increases in anthropogenic greenhouse gas concentrations (IPCC 2007).Climate change has led to increased temperature, rising sea levels, and increased rainfall intensity.Thus, we will face more extreme weather events in the future.
Taiwan, located in the West Pacific, is a climate change impact hot spot.Some studies indicated that Taiwan's surface temperature will rise in the future (Lin et al. 2009;Hsu et al. 2011).The temporal precipitation distribution may become more uneven, and rainless days might occur more frequently (Huang et al. 2012(Huang et al. , 2014a)).The changes in hydrologic projections indicate that water resources management in Taiwan could become a challenging task.The climate change impact on the Shihmen Reservoir water supply in Northern Taiwan has been assessed by numerous authors (Lee and Huang 2014;Huang et al. 2014b).The results show that reduced inflow and increased demand will result in drought for the Shihmen Reservoir supply zone.
The purpose of this study is to assess climate change effects on water resources in the Zhoshui River basin in the Changhua and Yunlin areas of Central Taiwan (Fig. 1).This area is crucial for rice production in Taiwan.However, the area has suffered from severe land subsidence (Tung and Hu 2012;Hung et al. 2012), limiting the potential for groundwater pumping.Therefore, understanding the impact of climate change on this area's water resources system is essential.
The Zhoshui River originates in the Central Mountain Range. Figure 1 shows the Zhoshui River basin water resources infrastructure.Figure 2 depicts the basin water resources system.The Wushe Reservoir has a catchment area of 219 km 2 and is used for hydropower.Its operating rule curve (Fig. 3a) indicates that the reservoir will generate as much electricity as possible if the water surface level is above the rule curve.Otherwise, it will generate electricity according to the inflow.The Wujie Dam raises the water level and conveys water to the Sun-Moon-Lake Reservoir.Sun-Moon-Lake is the largest off-site reservoir in this basin.Although pumped-storage hydroelectricity is its main objective, it supports the water supply when the water from the main stream cannot satisfy the demand.Figure 3b indicates that if the Sun-Moon-Lake Reservoir water surface  level is over the rule curve, the reservoir release meets the maximum hydropower and downstream use requirements.
Otherwise, it is at the minimum level for hydropower and downstream requirements.There are two tributaries in the midstream section, the Tanchung and Chenyulan Rivers, which encompass 700 and 450 km 2 , respectively.The water resources in the upstream and midstream section eventually converge into the Jiji Division weir (with a catchment area of 2304 km 2 ).This weir is the Zhoshui River basin management center and is responsible for the Changhua and Yunlin area demand.If the Jiji Division weir storage is lower than the total demand, the release for different water users (irrigation, domestic, and industrial) should be in accordance with their water rights (Fig. 3c).Industrial users have no water rights from February through May.
Hushan is a newly built reservoir with a catchment area of 259 km 2 .It is an off-site reservoir and receives Chingshui River flow from the intake at the Tungtou weir.The main task of this reservoir is to satisfy domestic demand and reduce groundwater pumping in Yunlin.
In the impact assessment performed in this study, we (1) calculated the Changhua and Yulin irrigation water requirements (in which agriculture is the main sector), (2) estimated the Zhoshui River basin streamflow, and (3) simulated the water resources system according to hydrological data.We then assessed the impact of climate change through a comparison between the present (2001 -2010) and the future (2046 -2065).

Meteorological Projections
General circulation models (GCMs) are the most advanced tools available for simulating the global climate system response to increasing greenhouse gas concentrations.This study focuses on the projected conditions for the years 2046 -2065.Meteorological data projections (precipitation and temperature) based on the A1B scenario (Nakicenovic and Swart 2000) were adopted from the following five GCMs: (1) CGCm3 from the Canadian Center for Climate Modeling and Analysis (CCCma), (2) Cm3 from the Centre National de Recherches Meteorologiques (CNRM), (3) Mk3.0 from Australia's Commonwealth Scientific and Industrial Research Organization (CSIRO), (4) Cm2.0 from the Geophysical Fluid Dynamics Laboratory (GFDL), and (5) FGOALS-g1.0from the State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics (LASG).The original GCM outputs have coarse resolutions that cannot be directly applied to watershed scale calculations.Statistical downscaling from GCM scenario-run grids to local meteorological stations is therefore necessary.We obtained the aforementioned GCM downscaled projections from the Global Change Research Center of National Taiwan University.The downscaling process can be briefly explained as follows: (1) the GCM output near Taiwan was adjusted with respect to the National Centers for Environmental Prediction (NCEP) reanalysis data through the training period by linking the normalized probability distribution function of the mean climate parameters; (2) transfer functions such as multiple-variant linear regression were established to link the NCEP reanalysis variants with local records through the training period; (3) meteorological projections at each station through the verification period were adjusted with respect to local records using the first stage procedure, and the established linkage was then extended to adjust the forward projection.For more downscaling process details, please refer to Kalnay et al. (1996) and Lin et al. (2009Lin et al. ( , 2010)).

Paddy Water Requirement
The crop water requirement relates directly to the crop variety and the field evapotranspiration.The Food and Agriculture Organization (FAO) of the United Nations uses the following equation to estimate the crop water requirement grown in large fields under disease-free and well-fertilized conditions (Allen et al. 2006): where ET c and ET o are the crop and reference evapotranspiration (mm day -1 ), respectively, and K c is a crop coefficient (dimensionless).K c is based on a distribution instead of a constant throughout the growth duration and differs depending on the crop variety.Shih et al. (1996) proposed a K c for paddy rice in Taiwan; the K c values for the first and second cropping seasons are shown in Table 1.
To estimate reference evapotranspiration, ET o , the FAO Penman-Monteith (PM) equation is recommended (Allen et al. 2006): where Δ and c are the saturation vapor pressure curve slope and the psychrometric constant (kPa °C-1 ), respectively.R n and G are the net radiation at the crop surface and the soil heat flux density (MJ m 2 day -1 ), respectively.T is the mean daily air temperature at a height of 2 m (°C).e s and e a are the saturation and actual vapor pressure (kPa), respectively.u 2 is the wind speed at a height of 2 m (m s -1 ).
Because only daily temperature and precipitation are available from the GCM projections, information for the PM equation is insufficient for evapotranspiration estimation for future scenarios.Instead, the Hamon method, a temperature-based equation for estimating evapotranspiration, was used (Shaw and Riha 2011) where PE is the potential evapotranspiration according to the Hamon method (mm day -1 ) and N h is the sunshine duration (hours).The theoretical future N h was defined as in Allen et al. (2006).

Bias Correction
ET o is regarded as a standard to be followed; therefore, the estimated PE value must be corrected.The quadrant transformation method (Huang et al. 2012) was applied in this study for bias correction.Figure 4 shows the quadrant transformation method concept, in which the curve indicates the percentage of time that evapotranspiration is equaled or exceeded during the period of study.The difference between the first and fourth quadrants was estimated from observations or projections, whereas the difference between the third and fourth quadrants was due to climate change.The curve in the second quadrant yields the corrected PE and was produced by conversions in the other three quadrants.
The bias correction process can be described by the following steps: (1) constructing the curves that indicate the percentage of time equaled or exceeded (Searcy 1959) in the first (observations in the present), third (projections in the future), and fourth (projections in the present) quadrants; (2) determining the corresponding percentile in the fourth quadrant using the given projection in the third quadrant; (3) finding a new value in the first quadrant and point in the second quadrant based on the percentile; and (4) repeating steps ( 2) and ( 3) to obtain the curve for the correction in the second quadrant.
Monthly bias correction was adopted in this study  1.K c distribution, accumulation of growing degree days and limit of ponding storage for growth stages of paddy rice.based on seasonal variations.The above statement could be expressed as the equation: where x is the daily potential evapotranspiration (mm); f and f --1 are the function and inverse function of the monthly curves that indicate the percentage of time equaled or exceeded.The superscripts P and F represent the present and future time, respectively; and the subscripts obs., proj., and crt.symbolize the observation data, GCMs projections, and corrected data, respectively.

Paddy Development Mechanism
Calendars are often used to predict crop development for management.However, calendar days can be misleading, particularly during early crop growth stages (Miller et al. 2001).In addition, climate change complicates using constant days to represent each growth stage.
Numerous studies have found that crop growth is directly related to temperature (Russelle et al. 1984;Eskridge and Stevens 1987).Therefore, the phenology crop model can be constructed according to the growing degree days (GDD) equation: where T b is the base temperature for a crop variety.ΔT is set to 0 if the daily temperature is less than T b ; thus, the crop stops developing at that time (Gilmore and Rogers 1958).
The GDD equation is a simple thermal time equation.It accumulates thermal time linearly with increasing temperature above a constant crop-specific base temperature.
Theoretically, if thermal time is accumulated correctly for a given crop variety, crop development should then be directly related to thermal time accumulation (i.e., GDD).The GDD equation indicates that each crop growth stage is related to the accumulation of thermal time, regardless of the planting time or location (Hodges 1991).
For paddy rice, T b is generally set to 10°C (Hardke et al. 2013).Moreover, the GDD accumulation value is set to 1725 °C-days according to recommendations for the common paddy variety in Taiwan (Hualien District Agricultural Research andExtension Station 2006, 2007).The GDD accumulation details for each paddy growth stage are shown in Table 1.

Irrigation Water Requirement
A paddy field water balance model is presented in Fig. 5.We applied it to calculate the irrigation water requirement.Considering different soil types in paddy fields and the irrigation canal water conveyance loss, the continuity equation for a paddy field is expressed as follows: where S t , P t , ET ct , and IR t indicate the daily water storage, rainfall, crop evapotranspiration, and irrigation water requirement in the paddy fields at time t (mm day -1 ), respectively.ƒ j is the percolation rate of the jth soil type (mm day -1 ), A j c is the percentage of the jth soil type area (%), A i * denotes the percentage of total farming area controlled by the ith canal (%), and CL i is the average water conveyance loss in the area controlled by the ith canal (%).
The irrigation water requirement must be externally supplied to offset the deficit for paddy growth when precipitation and storage do not satisfy water demand.By contrast, the irrigation water requirement is 0 when demand is satisfied.Therefore, Eq. ( 6) can be rewritten as follows: The rainfall stored in the paddy field for growth is termed the effective rainfall (P  1).

Rainfall-Runoff Model
The tank model (Sugawara 1961) was adopted to assess stream flow in the climate change context in this study.As the tank model is a conceptual and deterministic rainfallrunoff model composed of several tanks that accounts for a watershed storage effect.In other words, the tank model effective simulation performance is due to continuous damping and lagging effects (Lee 1993).The tank model advantages include a simple structure, easy calculations and reasonable functions (Yokoo and Kazama 2012).
Figure 6 shows the runoff mechanism for a 2-layer tank model.When precipitation reaches the height of ha 1 in the top tank, the watershed is saturated.If the water surface depth in the top tank is over ha 1 , additional precipitation discharges from the tank through orifice a 1 .This process is the major component of surface runoff.In a high-intensity storm, the top tank water surface depth is higher than ha 2 .The outflow through orifice a 2 represents the runoff in this scenario.Orifice a 0 simulates the percolation process in a watershed, in which the water is transported into the ground.The bottom of the 2-layer tank model represents the interflow and groundwater and is opposite the top tank.Parameter hb 1 is a threshold that is similar to ha 1 .The discharge through orifice b 1 represents the outflow of interflow and groundwater.
Six parameters in the 2-layer tank model require calibration.We used the multi-start Powell method to calibrate the tank model parameters.The Powell method is an efficient and robust conjugate direction method used to find the minimum of a function (Powell 1964).An inappropriate initial point converges toward a local minimum instead of a global minimum when the problem structure is more complex.As an improvement, the multi-start Powell method can overcome the Powell method disadvantages through providing several initial points.Tanakamaru (1995) suggested that the multi-start Powell method is an effective and powerful tool for tank model parameter calibration.Chen et al. (2005) recommended that the number of initial points should be equal to or more than 50 to avoid searching failure.
The parameter search boundaries for tank model calibration are shown in Table 2.If the orifice coefficient summation of each tank is more than 1, the tank outflow might be greater than the inflow and storage summation.To avoid this illogical relationship, the orifice coefficient of the top tank must be subject to the following equation during calibration: a a a 1 Additionally, we selected the average percentage error as the objective function during tank model parameter calibration: where Q t obs and Q t sim are the observed and tank model-simulated discharges, respectively.N d denotes the number of simulation days.

Simulation Assessment Indices
We used the degree of satisfaction, reliability and shortage index (SI) to represent the water resources simulation performance.The definition of these indices is provided in the following equations: where SPL and DM are the supplied water and water demand, respectively (m 3 ).N s is the total satisfied days, and N y indicates the simulation time in years.

Validation of Bias Correction
Figure 7 shows the difference before and after bias correction for evapotranspiration.The curve distributions indicating the percentage of time equaled or exceeded between observations and projections in the present (first and fourth quadrant) are quite different.However, after the quadrant transformation, the distribution of adjusted data (second quadrant) is more similar to the observations (first quadrant).

Water Balance Model Validation for Irrigation
Figure 8 shows the observed water consumption and estimated water requirements for paddy rice in the Chan-ghua and Yunlin areas.The bias range is approximately 2.3 -34.6% in Changhua and 0.2 -87.8% in Yunlin.On average, the biases through the first and second paddy growth seasons (approximately February -June and July -November) are 10.8 and 16.1% in Changhua and 17.0 and 30.0% in Yunlin, respectively.
This difference might arise owing to variations in hydrological conditions.The irrigation consumption records are based on actual water received from canals, whereas the model estimations assume that irrigation requirements can be fulfilled with abundant water.For example, (1) there is considerable underestimation in 2002 in Yunlin in the second paddy growth season (Fig. 8d).Because 2002 was a dry year, managers decided to transfer most of the water to Changhua rather than Yunlin (Fig. 8b); (2) there is overestimation in Yunlin throughout 2005 and 2006 because these two years were wet, and more water could be supplied to Yunlin (Figs. 8c and d).
The report of Water Resources Angecy proposed that crop water requirement minus effective rainfall throughout the whole paddy growth season should range between 1136 -1317 mm in Changhua and Yunlin area (Kan 2002).If considering conveyance loss, the proposed irrigation consumption will be approximately 1550 -1800 mm.Following the standard, the irrigation water requirement estimation (Fig. 8) is reasonable when using the water balance model.Excluding 2002Excluding , 2005Excluding , and 2006, the differences in Changhua between the observations and estimations in the first and second paddy growth seasons are 11.4 and 14.8%, respectively.The differences in Yunlin are 13.9 and 17.0%, respectively.
Currently, the planned farming areas in Changhua in the first and second paddy growth seasons are 370.3 and 390.8 km 2 , respectively.The areas in Yunlin are 106.7 and 369.9 km 2 , respectively.By multiplying the estimated values from 2001 -2010, the annual irrigation volume requirement is nearly 1.9 billion m 3 , which is close to the annual irrigation supply from the Jiji division weir at 1.8 billion m 3 .

Tank Model Validation
Wushe, Tanchung, Chenyulan, and Chingshui are the four sub-basins selected for constructing the tank model for the Zhoshui River basin (Fig. 1).The river flow gauges at these four sub-basins are unaffected by water-related infrastructures such as reservoirs.These sub-basins cover about 60% of the basin area.In the rest of the area, the area ratio is applied to estimate river flow.For example, the Chingshui River flow below the Tungtou weir (154 km 2 ) is equal to 0.59 times the Tungtou weir flow (259 km 2 ).
We used 8-yr data for parameter calibration and the remaining 2 years (2009 -2010) for verification.Table 3 shows the tank model calibrated parameters.The annual differences in volume in the four main sub-basins are verified between 3.7 -13.1%.In addition, the bias throughout the wet season (May -October) is higher than in the dry season (November -April), except in the Tanchung sub-basin (Table 4).Figure 9 shows the daily discharge validation in the sub-basins and demonstrates that the tank model is able to simulate the river flow characteristics in these sub-basins.
In addition, when applying the tank model we assumed that the watershed characteristics will remain unchanged in the future-in other words, that the parameters identified for the current climate will remain valid.

Projection of Meteorological Data
Figure 10 illustrates the temporal precipitation and temperature distributions.Note that the meteorological data in the Zhoshui River basin are a weighted average of the sub-basins.
According to precipitation projections, the annual precipitation in the Zhoshui basin in the future (2046 -2065) will range from 2104.0 -2489.9mm, with an average of 2287.5 mm.Based on the current annual precipitation (2469.7 mm for 2001 -2010), most GCMs project a decrease in the future (-14.8-0.8%).On average, the precipitation projections are 7.4% lower than at present.However, the temporal precipitation distribution in the Zhoshui basin indicates that more fluctuations would occur in each month, especially in July and August (see top left graph of Fig. 10).
The precipitation projections for the upstream and downstream areas are similar.However, the monthly precipitation distribution in the future has obvious fluctuations.As the top center and top right graphs of Fig. 10 indicate, the temporal precipitation distribution will be more uneven in both areas in the future.The difference between the average annual precipitation in the present and future is insignificant.The present annual precipitation levels in Changhua and Yunlin are 1435.9and 1878.7 mm, respectively.For Changhua, the variations in projections of annual precipitation are between -241.9 and 260.1 mm (-16.8 -18.1%), with an average of 33.4 mm (2.3%).For Yunlin, the variations in projections of annual precipitation are between -270.1 and 578.8 mm (-14.4 -30.8%), with an average of 49.6 mm (2.6%).
Temperature projections for the Zhoshui basin show

Sub-Basins
Orifice coefficient (10 Note: "+" means overestimation; "-" means underestimation.that the difference in the annual mean between the present and future is insignificant.The weighted annual temperature average in the Zhoshui basin is currently 18.8°C.The variations in projections are between -1.1 and 0.1°C (-5.9 -0.6%), and the average decrease in the future is 0.5°C (2.8%).By contrast, the variations in temperature projections for Changhua and Yunlin are significant.For Changhua, the increment in annual mean temperature projections range from 0.7 -2.1°C (3.1 -9.2%), with an average of 1.4°C (6.0%).

Impact of Climate Change on Irrigation Water Requirements
Table 5 shows that, compared with the present irrigation requirement (2001 -2010), the projected irrigation water requirement for 2046 -2065 increases slightly, within the range of 19.4 -155.6 mm (1.2 -9.5%).GCMs projections indicate that the monthly mean temperature in the future will increase 0.5 -3.0°C on average, as shown in Fig. 10.Rising temperature will certainly cause crop water requirements (ET c ) to increase during 2046 -2065.The in-crements of ET c are within 85.0 -132.4 mm (16.1 -22.0%) throughout each paddy growth season in the Changhua and Yunlin areas.
However, rising temperatures associated with GDDs will reduce the growth duration.Compared to the present growth duration, the future growth duration of paddy crops will decrease by 5.2 -10.7 days in both areas.Hence, there is a marked increment in ET c , but the irrigation requirement increment will be smaller than 10% in the future.

Impact of Climate Change on River Flow
In 2046 -2065, the annual precipitation in these subbasins will decrease, except in Wushe.On average, the projected annual precipitations in the Tanchung, Chenyulan, and Chingshui sub-basins decrease by 404.4, 439.1, and 22.4 mm (20.9, 17.9, and 0.7%), respectively.Wushe is projected to increase by 754.2 mm (24.9%) in the future.Consequently, the variations in projected annual discharge in the sub-basins are approximately -272 -3 million m 3 (-33.3-0.5%).Figure 11 shows a comparison of temporal distributions of river flow between the present and future.Overall, an obvious decrease will occur in the discharge of the Zhoshui River from July to December.

Impact of Climate Change on the Water Supply and Benefits of Hushan Reservoir
In the following section, we assume that the industrial and domestic demands will remain the same in the present (2001 -2010) and future (2046 -2065).The daily industrial and domestic demands are 3.5 × 10 5 and 2.0 × 10 5 m 3 , respectively.
Although the Hushan Reservoir operational regulations are not yet stipulated, we assume that the Hushan Reservoir will be responsible for supplying all domestic demand because the Zhoshui River sediment concentration is always high.In addition, this reservoir must draw water from Chingshui Creek without affecting irrigation demand downstream.
Table 6 shows the water supply situation before and after building the Hushan Reservoir.The Hushan Reservoir clearly aids domestic supply.Future water shortages for irrigation and industry are exacerbated because of river flow reductions throughout the Zhoshui River basin (the decrement is 20.4% according to the weighted average; see section 4.3).On average, the annual irrigation and industry deficit increases by 159.2 and 4.5 million m 3 , maximum consecutive water-shortage days for irrigation and industry increase by 2 -4 times, and the shortage index for industry is raised to 3.7 from 2.5.Moreover, the satisfaction and reliability temporal distributions indicate that water shortages from September to February in subsequent years will be more severe than those at present (Figs. 12 and 13).
Domestic demand could be fully satisfied when    the Hushan Reservoir begins operation (Table 6; Figs. 12 and 13).This result demonstrates that the new reservoir achieves its original objective to stabilize the domestic water supply.However, we found that Hushan Reservoir would release only 25.4 million m 3 per year in the future.This annual release is approximately 50% of the reservoir's capacity (52 million m 3 ).To enhance the Hushan Reservoir utilization efficiency, alternatives should be evaluated that assist other water users.

Adaptations for Industrial and Irrigation Demands
Because different GCMs produce different patterns of change in temperature and precipitation, multi-ensemble runs instead of single GCMs were used to reduce uncertainties.Therefore, the results show an average of all GCM simulations.
Compared with domestic demand, future irrigation and industrial demand will encounter higher pressures than at present.The first adaptation indicates that the Hushan Reservoir could significantly support industrial demand in the future without affecting domestic use.Comparing the industrial demand indices with the future average in Table 6 and case (a) in Table 7, in which the annual shortage drops from 22.8 -0.3 million m 3 , indicates that maximum consecutive water-shortage days would decrease to 7.2 from 83.6 days, and the shortage index would decline to 0.02.The temporal satisfaction degree distributions [part (b) in Fig. 12 vs.part (a) in Fig. 14] and reliability degree [part (b) in Fig. 13 vs. part (a) in Fig. 15] indicate that these industrial supply indices will be higher than 90% in the future.Consequently, the Hushan Reservoir utilization efficiency will increase after supporting industrial demand.The future annual Hushan Reservoir release increases to 47.7 million m 3 , compared to 25.4 million m 3 per year at present.
Currently, the conveyance losses due to irrigation in the Changhua and Yunlin areas are 40.7 and 36.0%,respectively.These values are higher than the average loss in Taiwan.The second adaptation further reduces the irrigation conveyance loss in both areas to an average of 30%.Case (b) in deficit occurs after improving irrigation conveyance loss.We considered applying area reduction as a possible alternative to offset future water deficits.Using trial and error, we found that reducing farming areas by 10% in the second paddy growth season would allow the annual irrigation deficit to return to the present level.According to case (c) in Table 7, the annual irrigation deficit would drop to 457.0 million m 3 if the adaptations were applied.However, indices such as irrigation satisfaction and reliability [case (c), Figs. 14 and 15] would not return to the present level [case (a) in Figs. 12 and 13].These indices are affected mainly by climate change.

CONCLUSIONS
The impact of climate change on water resources is of high concern.However, the limitations imposed by hydrological spatial and temporal variation conditions and the complexity of human intervention effects in the environment hinder estimation (Rees et al. 1997).This study investigated the impact of climate change on water resources within the period 2046 -2065.The Zhoshui River basin in Central Taiwan was selected for analysis.This paper demonstrates an integrated procedure for assessing water resources system response to climate change on the basin scale, and provides a reference for people concerned with local scale planning practices related to the impact of climate change.Bias correction was attained using a quadrant transformation method.A two-layer tank model was used for rainfall-runoff estimation.A water balance model for paddy fields was proposed for simulating irrigation water requirements.The applicability of these models was demonstrated in this study.
Comparison of the projections from 2001 -2010 and 2046 -2065 suggests that future irrigation water requirements will increase 10% less in each cropping season in the Changhua and Yunlin areas.However, future river flow in the Zhoshui basin will decrease by approximately 20%.Consequently, water shortage is inevitable, particularly from September to February, owing to irrigation.Thus, climate change will have a considerable impact on water supply, particularly in the agricultural sector.The annual deficit associated with irrigation and industrial demand may increase by 30% compared to the present levels, and the maximum number of consecutive water-shortage days will nearly double.Adaptations were introduced to alleviate the impacts of climate change on water resources.Domestic demand will be fully met after the Hushan Reservoir comes on-line in mid-2016.Furthermore, the industrial water shortage can be substantially reduced without affecting domestic demand if the Hushan Reservoir can support it.Alternatively, reducing irrigation conveyance losses and farming areas are effective ways to mitigate water shortages for irrigation in the future.We recommend reducing conveyance loss by 30% and cutting farming areas by 10% in the second paddy growth season to mitigate future annual irrigation deficits and retain deficits at the present level.

Fig. 1 .
Fig. 1.Location and related water resource infrastructures of the study area.

Fig. 5 .
Fig. 5. Conceptual model for water balance in paddy fields.

Fig. 8 .
Fig. 8.Comparison of paddy water requirements between recorded consumption and model estimation in 2001 -2010: (a) Changhua area in February -June; (b) Changhua area in July -November; (c) Yunlin area in February -June; (d) Yunlin area in July -November.

Fig. 9 .
Fig. 9. Comparison of daily discharge between observation and tank model projection for four main sub-basins of the Zhoshui River in 2009 -2010.
Fig. 12. Monthly Degree of Satisfaction: (a) without Hushan reservoir in the Present; (b) with Hushan reservoir in the future.
of water supply simulation of adaptations in 2046 -2065.Note: (a) Hushan supports industrial demand; (b) Hushan supports industrial demand and conveyance loss of irrigation reduces to 30%; (c) further reduces 10% of irrigation area in second cropping season.

Fig. 14 .Fig. 15 .
Fig. 14.Degree of satisfaction for adaptations in 2046 -2065: (a) if Hushan supports industrial demand; (b) if further reduce the conveyance loss of irrigation to 30%; (c) if further reduce 10% of irrigation area in second cropping season.

Table 3 .
Result of calibration for four main sub-basins of the Zhoshui River.

Table 4 .
Tank model validation in volume error for four main sub-basins of the Zhoshui River in 2009 -2010 (unit in %).

Table 6 .
Results of water supply simulation.

Table 7
According to the simulation results shown in Table 6 (in the case of without Hushan in 2001 -2010) and Table 7 case (b), an extra 76.2 million m 3