Introduction

It is widely accepted that although seasonal snow cover forms only a small fraction of world’s fresh water, its hydrologic contribution is significant. In high and mid-latitudes, runoff from shallow snow cover often provides 80 percent or more of the annual surface runoff (Maidment 1993). In Korea, the snowmelt by winter snowfalls contributes considerably to water resources in high-altitude regions. For example, the frequency of heavy snowfall event (>20 cm in depth) is often greater than 40 % of the total events in mid-eastern mountainous areas, and the snowmelt from these regions supplies a considerable amount of stream discharge during spring period. However, a quantitative analysis of watershed snowmelt is generally difficult to perform because of the insufficient spatial snow information and complexity of the physical processes involved with snowmelt and runoff generations (Shin et al. 2007).

As climate change appearance has affected recent patterns in frequency, strength, spatial range, and duration of snowfall, the snowmelt changes will provoke various problems in hydrology and water quality. For example, as the snow decreases in the Rocky mountains of American west, river stages become lower and policy decisions related to spring water resources were questioned (Stewart et al. 2005; Regonda et al. 2005). Moreover, while snowfall of deep drifts during the winter melt in early spring, the forest fire frequency and strength increase (Westerling et al. 2006). In the context of assessing the impacts of climate change on water resources, it is important to adequately evaluate the impacts of climate change on snowmelt and the subsequent flow changes.

Recently, many studies have evaluated the effects of climate change on the hydrology of snowmelt-driven watersheds. Vicuna et al. (2010) showed climate change impacts on the hydrology of a snowmelt-driven basin in Chile. Tahir et al. (2010) simulated snowmelt runoff under various climate scenarios in a large mountainous watershed in Northern Pakistan by employing the snowmelt runoff model (SRM), which is based on a simple degree–day method for snowmelt simulation. Kobierska et al. (2011) investigated the climate change effects on snowmelt discharge of a glacierized watershed. In the study, two different models including a detailed energy balance model and a conceptual runoff model were used to compare the benefit of each model for snowmelt simulation.

Many models that deal with snowmelt and snow accumulation have been developed (Martinec 1960; Anderson 1968, 1976; Rango and Martinec 1979; Blöschl and Kirnbauer 1991; Jordan 1991; Coughlan and Running 1997; Fontaine et al. 2002; Garen and Marks 2005). Soil and Water Assessment Tool (SWAT) is one of the many models that include snow processes with simulation of water balances. It was initially developed for comprehensive modeling of the impacts of management practices on water yield, sediment yield, crop growth, and agricultural chemical yields in large ungauged basins (Arnold et al. 1998).

This study is to assess the impact of potential climate change on snowmelt and its non-point source pollution loads to stream in a 6,640 km2 high-elevation watershed using SWAT model. The Terra MODIS (moderate resolution imaging spectroradiometer) image is used to obtain the snow cover area (SCA) and determine the snow depletion parameters using 11-year (2000–2010) data. After the SWAT model is set up using the observed dam inflow, sediment, TN, and TP data, the model is applied to evaluate the future climate change impact on the snowmelt runoff and the nutrients load to the dam reservoir under five scenarios (HadCM3 SRES A2, A1B, and B1 and HadGEM3-RA RCP4.5 and 8.5). Figure 1 shows the schematic flow chart of this study.

Fig. 1
figure 1

Schematic flow chart of the study process

Full size image

Data and method

Study area

A 6,640 km2 mountainous watershed was selected. It is located in the north-eastern part of South Korea within the latitude–longitude range of 127.9°E–129.0°E, 36.8°N–37.8°N. The watershed is one of the heavy snowfall areas in South Korea. Figure 2a shows the snowfall area in South Korea with the heavy snowfall frequency over 20 cm during 30 years (1981–2010). The study watershed is marked with slashed lines, and Fig. 2b shows the enlarged study watershed with the subwatershed boundaries for modeling and the locations of weather and gaging stations. The watershed elevation ranges from 112 to 1,562 m, with average slope of 36.9 %. The annual average precipitation is 1,261 mm, and the mean temperature is 9.4 °C over the last 30 years (1981–2010). For model calibration, 10-year (2000–2010) data at six weather stations, three stream water level gaging stations (YW #1, YW #2, and CD), and two stream water quality measuring stations (YW #1 and YW #2) were used.

Fig. 2
figure 2

The study area: a heavy snowfall area of South Korea and b Chungju dam study watershed

Full size image

SWAT snow hydrology description

The SWAT model, developed by Arnold et al. (1998), is a physically based continuous, long-term, distributed parameter model designed to predict the impact of land management practices on hydrology, snowmelt, sediment, and agricultural chemical yields in large watersheds with varying soils, land use, and management conditions over long periods of time (Srinivasan and Arnold 1994; Rosenthal et al. 1995; Arnold et al. 1998, 1999).

In the SWAT model, snowmelt hydrology is realized on an hydrologic response unit (HRU) basis. The model has the function of snow mass routing and areal depletion estimation. Depending on data availability and modeling accuracy, one sub-basin may have one or several HRUs. When the daily mean air temperature is below a defined value, the precipitation within an HRU is classified as snow and the liquid water equivalent of snow precipitation is added to snowpack. The snowpack increases with additional snowfall but decreases by snowmelt or sublimation. The mass balance for the snow pack on a daily basis is

$$SNO = SNO + R_{\text{day}} - E_{\text{sub}} - SNO_{\text{mlt}} ,$$
(1)

where SNO is the water content of snow pack (mm), R day is the amount of precipitation (mm), E sub is the amount of sublimation (mm), and SNO mlt is the amount of snowmelt (mm).

The amount of snow is expressed as depth over the total HRU area. The factors that contribute to variable snow coverage are usually similar across years, making it possible to correlate the areal coverage of snow with the amount of snow present in the sub-basin at a given time. This correlation is expressed as an areal depletion curve, which is used to describe the seasonal growth and recession of the snow pack as a function of the amount of snow present in the sub-basin (Anderson 1976).

The areal depletion curve requires a threshold depth of snow, SNO 100, to be defined above which there will always be 100 % cover. The threshold depth is unique to the watershed depending on factors such as vegetation distribution, wind loading of snow, wind scouring of snow, interception, and aspect. The equation for the curve based on a natural logarithm can be expressed as

$$sno_{\text{cov}} = \frac{SNO}{{SNO_{100} }} \cdot \left( {\frac{SNO}{{SNO_{100} }} + \exp \left( {{\text{cov}}_{1} - {\text{cov}}_{2} \cdot \frac{SNO}{{SNO_{100} }}} \right)} \right)^{ - 1} ,$$
(2)

where sno cov is the fraction of the HRU area covered by snow, SNO is the water content of the snow pack on a given day (mm), SNO 100 is the threshold depth of snow at 100 % coverage (mm), and cov1 and cov2 are coefficients that define the shape of the curve. The values used for cov1 and cov2 are determined by solving the equation and using two known points: 95 % coverage at 95 % SNO 100 and 50 % coverage (SNO50COV) at a user-specified fraction of SNO 100. As the value for SNO 100 increases, the influence of the areal depletion curve will assume more importance in snowmelt processes.

Snow coverage from Terra MODIS NDSI (Normalized Difference Snow Index)

To obtain the SCA by HRU base in study watershed, Terra MODIS image was used. MODIS is an imaging spectroradiometer to provide imagery of the Earth’s surface and clouds in 36 spectral bands from 0.4 to 14.0 μm (Barnes et al. 1998). The MODIS/Terra Snow Cover 1-Day L3 Global 500m Grid (MOD10A1) for 10 years from 2000 to 2010 during the snow season (November–April) was used in this study. The data were developed using MODIS bands 4 (0.545–0.565 μm) and 6 (1.628–1.652 μm) by calculating the normalized difference snow index (NDSI) (Hall et al. 1995):

$${\text{NDSI}} = ({\text{band 4}} - {\text{band 6}})/({\text{band 4}} + {\text{band 6}}).$$
(3)

Meanwhile, the distribution maps by snow depth during the snow period are necessary to determine the snow depletion parameters of Eq. (2) for the watershed. Thus, a set of snow depth distributions (SDD) in time series was made from the inverse distance weight (IDW) interpolated map using the six ground snowfall data by masking it with the MODIS SCA map. As in Eq. (2), the areal snow depletion curve requires the threshold depth of snow, SNO 100 and the coefficients cov1 and cov2. By analyzing the SNO 100 and the corresponding SCA, and the fractions from the SDD set of each year, the snow depletion curve was determined as shown in Fig. 3. Table 1 shows the derived parameters for snow depletion curve of each year (2000–2010). The key parameter SNO50COV for determining the depletion shape, the 50 % coverage at a fraction of SNO 100 ranged from 0.42 to 0.70. This means that the watershed main snowfall area usually covers 42–70 % at high elevation of the watershed as seen in Fig. 2a and the snow is depleted as time passes. The average SNO50COV for nine datasets was 0.56. The average value of the parameters was applied for future snow cover depletion.

Fig. 3
figure 3

Snow depletion curves (2000–2010) derived from Terra MODIS images

Full size image
Table 1 Snow depletion curve shape factor (2000–2010) derived from Terra MODIS images
Full size table

Climate change scenarios

The Intergovernmental Panel on Climate Change (IPCC) published future scenarios in its Special Report on Emissions Scenarios (SRES) document (Nakicenovic 2000). Among all SRES scenarios, four marker scenarios (A1, A2, B1, and B2) are commonly used (Van Vuuren and O’Neill 2006). The A1 and B1 scenarios emphasize ongoing globalization and project future worlds with fewer differences between regions, whereas the A2 and B2 scenarios emphasize regional and local social, economic, and environmental development and project more differences (IPCC 2007). The three A1 groups are distinguished by their technological emphasis: fossil intensive (A1FI), non-fossil energy sources (A1T), or a balance across all sources (A1B). In this study, the SRES A1B, A2, and B1 scenarios of HadCM3 by the Hadley Centre at UK Meteorological Office (HC-UKMO) were used since Korea Meteorological Administration (KMA) has cooperated with HC-UKMO. The future data (2011–2100) of 96 km × 73 km grid size were spatially corrected to ensure that the observed data (1981–2010, baseline) and the scenario of the same period have similar statistical properties by the methods of Alcamo et al. (1997) among various statistical transformations. The monthly data to daily generation were downscaled using the LARS-WG downscaling method. LARS-WG is a stochastic weather generator which can be used for the simulation of weather data at a single site (Racsko et al. 1991), under both current and future climate conditions. Details for the downscaling process are found in Park et al. (2011).

Recently, the IPCC (2013) has published new climate change scenarios based on Representative Concentration Pathways (RCPs) with a possible range of radiative forcing values in 2100. The RCP 2.6 scenario is characterized by very low greenhouse gas concentrations, producing forcings around 3.1 W/m2 mid-century, and dropping to 2.6 W/m2 by 2100. In order to achieve such radiative forcing levels, greenhouse gas concentrations are reduced substantially over time. The RCP4.5 (medium low) and RCP 6.0 (medium high) are stabilization scenarios on the assumption that technologies and strategies for reducing greenhouse gas emissions are stabilized, where the total radiative forcing levelis stabilized before 2100 and after 2100 by employinga range of technologies and strategies for reducing greenhouse gas emissions. The RCP8.5 is characterized by high greenhouse gas concentration levels, stabilizing emissions post-2100 and atmospheric concentrations post-2200 (Moss et al. 2010). In this study, RCP4.5 and RCP8.5 scenarios of HadGEM3-RA were used. HadGEM3-RA is a regional atmospheric model based on the atmospheric component of HadGEM3, the latest model developed by HC-UKMO. The KMA regenerated the scenarios with 12.5 km × 12.5 km resolution using the dynamic downscaling method.

For the five scenarios (HadCM3 SRES A1B, HadCM3 SRES A2, HadCM3 SRES B1, HadGEM3-RA RCP4.5, and HadGEM3-RA RCP8.5), the future changes of climate variables were arranged for the 2040s (2020–2059) and 2080s (2060–2099) using the baseline (1981–2010).

Table 2 summarizes the results of statistical analysis for the trends of average precipitation and temperature in annual and snowmelt period (November–April) for the five scenarios. The biggest changes in temperature and precipitation were +4.64 °C at minimum temperature and +45.7 % in 2080s snowmelt period under HadGEM3 RCP8.5 scenario. The future precipitation and temperature of RCPs (AR5) showed bigger changes than those of SRES (AR4). This result may come from the temporal and spatial resolutions between AR4 (monthly, 2.5° × 3.75°) and AR5 (daily, 0.125° × 0.125°). The averaged characteristics at coarse resolution may be resolved to represent the spatial variation at finer resolution.

Table 2 Changes in annual and snowmelt periods precipitation and temperature with the five climate change scenarios
Full size table

Results and discussion

SWAT model calibration and validation

The SWAT model was calibrated for 10 years (2000–2010) of daily runoff data at three locations (YW #1, YW #2, and CD) and monthly stream water quality (SS, TN, and TP) data at two locations (YW #1 and YW #2). Table 3 shows the calibrated parameters for hydrology, snowmelt, and stream water quality. The snow parameters, SFTMP and SMTMP were referenced by analyzing the temperatures of snowfall and rainfall days.

Table 3 The calibrated SWAT model parameters at three stations
Full size table

With the help of SNO50COV value fixed for each year, the other snow parameters were calibrated. Figure 4 shows the observed versus simulated runoff using the SNO50COV value (Sim A) of each year and fixing SNO50COV with 0.56 (Sim B) at CD station (2000–2010). As seen in Fig. 4, the runoff from melting snow is detected to increase in March and April. Table 4 summarizes the statistical results for the SWAT-simulated snowmelt (SM) and runoff (Q) during the snowmelt period (November–April), respectively. The 10-year average Nash–Sutcliffe model efficiency (NSE) for the snowmelt period was 0.53 (Sim A) and 0.50 (Sim B), respectively. The runoff considering snowmelt parameter SNO50COV in each year was simulated well compared to the results of fixing the parameter and the NSE was enhanced by 0.03. Although the snowmelt is a function of many variables, the relatively low NSE value during the snowmelt period in SWAT modeling may come from the uncertainty of SDD generated by six ground snowfall data, snow water equivalent (SWE) by wet or dry snowpack, and the air temperature (cool or warm) for the snowmelt.

Fig. 4
figure 4

Hydrograph comparison between observed and simulated discharges by varying snowmelt parameters of each year (Sim A) and fixing parameters (Sim B) at CD (2000–2010) location

Full size image
Table 4 Summary of the SWAT model calibration and validation statistics for daily runoff with and without snowmelt parameters during the snowmelt period at CD location
Full size table

Figure 5 shows the calibration and validation results for water quality at YW #1, YW #2 with snowmelt parameters. The NSE for SS, TN, and TP during the calibration and validation periods were 0.72, 0.54, and 0.70 at YW #1, and 0.75, 0.85, and 0.70 at YW #2, respectively. The coefficients of determination (R 2) for SS, TN, and TP were 0.87, 0.74, and 0.85 at YW #1, and 0.61, 0.88, and 0.62 at YW #2, respectively. Other detailed discussion for parameter sensitivity and model results of the study watershed can be found in Park et al. (2011).

Fig. 5
figure 5

Comparison between observed and simulated stream water quality at YW #1 (left) and YW #2 (right) stations

Full size image

The future climate change impact on snow hydrology and stream water quality

Using the SWAT calibrated parameters in Table 3 and the average values of SNO50COV, cov1, and cov2 in Table 1, the five future climate change scenarios were applied to the study watershed. Figure 6 shows the future 2040s and 2080s monthly snowmelt for the five climate change scenarios. The future snowmelt amount decreased for all scenarios as it goes to 2080s compared to baseline. Meanwhile, the future snowmelt during the melt period decreased largely for the two RCP scenarios than the three SRES scenarios. This may come from the big increase of future minimum temperature in RCP scenarios as seen in Table 3. The future temperature condition turned the snowfall to rainfall and increased the 3-month overall water yields not by melt flow but by rain runoff from November to January.

Fig. 6
figure 6

The effects of climate change on monthly snowmelt from two SRES (top) and RCPs (bottom) climate change scenarios

Full size image

Table 5 summarizes the future snowmelt and the runoff during the snowmelt period for the five climate change scenarios. Looking at the results of HadCM3 scenarios, the future runoff by snowmelt (SMR) showed similar values ranging from 37.5 to 42.5 % compared to the value of 43.6 % of the baseline period. Meanwhile, the results of HadGEM3-RA showed different patterns for the snowmelt-induced runoff. The future snowmelt decreased up to 45.9 mm (39.9 %) in 2080s HadGEM RCP8.5 scenario from 115.1 mm (100 %) in baseline period (1981–2010). For this scenario, the runoff discharge from November to April decreased by 15.1 from 41.9 % in baseline to 26.8 % in 2080s, while the 1-year whole discharge decreased by 22.9 from 58.8 % in baseline period to 35.9 % in 2080s. As mentioned above, the snowmelt ratio in 2080s HadGEM RCP8.5 scenario decreased up to 11.9 % on account of snowfall reduction by temperature rise during snowmelt period. Thus, the future runoff by rainfall from November to April becomes more important even when the amount is relatively low comparing with the remaining period runoff.

Table 5 Summary of the future predicted annual and snowmelt period runoff ratio and snowmelt ratio for the five climate change scenarios
Full size table

After the evaluation of future snow hydrology, the impact of future snowmelt change on stream water quality was evaluated in terms of SS, TN, and TP at the watershed outlet. Figure 7 shows the changes of future monthly SS, TN, and TP loads, and Table 6 summarizes the percent changes in future annual and snowmelt period for the five climate change scenarios. Looking at the results of the three SRES scenarios in Fig. 7, the future monthly SS, TN, and TP loads during the snowmelt period showed similar trends of big increase in April with the baseline loads. On the other hand, the two RCP scenarios showed different trends of big increases in March for SS and from February for TN and TP.

Fig. 7
figure 7

The changes of 2040s and 2080s monthly sediment load, TN, and TP by SRES and RCP climate change scenarios

Full size image
Table 6 Percentage changes of SS, TN, and TP loads in annual and snowmelt periods with the five climate change scenarios
Full size table

As seen in Table 6, the future SS, TN, and TP loads during snowmelt period under RCP scenarios showed big increases in annual total, while the SRES scenarios showed overall decreases for the three loads. This was primarily caused by the big differences of precipitation during the snowmelt period between the RCP and SRES scenarios as listed in Table 2. In addition, due to the big differences of snowmelt between SRES and RCP scenarios as shown in Fig. 6, much more runoff discharges turned the snowmelt discharges into the rainfall-runoff contribution in RCP scenarios. The three SRES precipitation changes of −13.5 % in 2080s B1 scenario to +11.8 % in 2040s B1 scenario during the snowmelt period resulted in the changes of −46.4 % in 2080s A1B scenario to −12.5 % in 2040s A2 scenario for SS, −16.6 % in 2080s A2 scenario to +4.1 % in 2040s B1 scenario for TN, and −20.8 % in 2040s A2 scenario to +31.1 % in 2040s B1 scenario for TP, respectively, while the two RCP precipitation changes of +0.57 % in 2040s RCP8.5 scenario to +45.7 % in 2080s RCP8.5 scenario during the snowmelt period resulted in the changes of −16.9 % in 2040s RCP4.5 scenario to +53.0 % in 2080s RCP8.5 scenario for SS, +98.1 % in 2040s RCP4.5 scenario to +145.2 % in 2040s RCP8.5 scenario for TN, and +137.5 % in 2080s RCP8.5 scenario to +225.7 % in 2040s RCP8.5 scenario for TP, respectively. The future increases of sediment, TN, and TP loads flowed in the dam reservoir during the snowmelt period may contribute to the algal bloom in spring period with the nutrients rise by turnover of the water and the eutrophication problem for the long time progress under the condition of future temperature increase.

Summary and conclusions

This study evaluated the potential climate change impact on snowmelt and stream water quantity for a 6,640 km2 high-elevation watershed in South Korea using SWAT model. To determine the SWAT snowmelt parameters, Terra MODIS images were used to obtain the SCA, and the SDD was developed using snowfall data of ground meteorological stations to determine the shape of snow cover depletion curve (SCDC) of the study watershed. For 10 datasets (2000–2010) during the snowmelt period (November–April), the SNO50COV parameter, that is, the 50 % coverage at a fraction of SCDC that determines the shape of snow depletion process, showed values of 0.42–0.70 and an average value of 0.56. Using the snow parameters of each year, the SWAT model was calibrated using daily runoff discharge data at three locations and monthly stream water quality (SS, TN, and TP) data at two locations. The 10-year NSE for runoff was 0.54 during the snowmelt period. The snow on the ground is affected by the wind especially at high elevation, and the compaction and melting processes are certainly different by the terrain aspect with steep slopes. These two factors can improve the snow hydrology when incorporated in the snowmelt modeling process.

Using the SWAT calibrated parameters and the average value of SNO50COV, the five future climate change scenarios of were applied to the study watershed and the future projection results were arranged for the 2040s (2020–2059) and 2080s (2060–2099) using the baseline (1981–2010). Among the five tested scenarios, the future snowmelt decreased to 45.9 mm (39.9 %) in 2080s HadGEM RCP8.5 scenario from 115.1 mm (100 %) in baseline period. For the RCP scenarios, more discharges turned the snowmelt discharges into the rainfall-runoff contribution by the future temperature and precipitation increases during the snowmelt period. The 2 RCP precipitation changes of +45.8 to +54.5 % during the snowmelt period resulted in the changes of −16.9 to +53.0 % for SS, +98.1 to +145.2 % for TN, and +137.5 to +225.7 % for TP, respectively. The future increases of sediment, TN, and TP loads flowed in the dam reservoir during the snowmelt period may contribute to the algal bloom in spring period with the nutrients rise by turnover of the water and the eutrophication problem for the long time progress under the condition of future temperature increase. The examination of various climate change scenarios to high-elevation watershed hydrology is to be considered for the changes of winter snowfalls and the impacts on the spring snowmelt. Even though the results of this study give plausible clue for the future snowmelt impact on hydrologic water cycle and the stream water quality, there may be other factors that need to be considered to explain the impact by the future land-use change and the eco-hydrological changes like the changes of soil environment, leaf canopy inception of snowfall, and snowmelt mechanism considering the litter fall and humus soil layer.