Stoch Environ Res Risk Assess (2017) 31:1171–1182 DOI 10.1007/s00477-016-1263-1
ORIGINAL PAPER
Assessment of the climate change impacts on flood frequency (case study: Bazoft Basin, Iran) Parisa Almasi1 • Saeid Soltani1
Published online: 18 May 2016 Ó Springer-Verlag Berlin Heidelberg 2016
Abstract The present study attempts to investigate potential impacts of climate change on floods frequency in Bazoft Basin which is located in central part of Iran. A combination of four general circulation models is used through a weighting approach to assess uncertainty in the climate projections. LARS-WG model is applied to downscale large scale atmospheric data to local stations. The resulting data is in turn used as input of the hydrological model Water and Energy Transfer between Soil, plants and atmosphere (WetSpa) to simulate runoff for present (1971–2000), near future (2020–2049) and far future (2071–2100) conditions. Results demonstrate good performance of both WetSpa and LARS-WG models. In addition in this paper instantaneous peak flow (IPF) is estimated using some empirical equations including Fuller, Sangal and Fill–Steiner methods. Comparison of estimated and observed IPF shows that Fill–Steiner is better than other methods. Then different probability distribution functions are fit to IPF series. Results of flood frequency analysis indicate that Pearson III is the best distribution fitted to IPF data. It is also indicated that flood magnitude will decrease in future for all return periods. Keywords Climate change Flood frequency LARSWG WetSpa IPF Bazoft Basin Iran
1 Introduction One of the main assumption of flood frequency analysis is that the return period of a flood peak of given magnitude is stationary with time (e.g., Natural Environment Research & Saeid Soltani
[email protected] 1
Department of Natural Resources, Isfahan University of Technology, Isfahan 8415683111, Iran
Council (NERC), 1975). This is valid if long-term climatic, hydrological and physical characteristics of a basin do not change with time (Cameron et al. 2000). However, changes in climate and hydrology over time have been well documented in the literature (e.g., Wigley and Raper 1992; Arnell and Reynard 1996; Hulme et al. 1999). Intergovernmental Panel of Climate Changes (IPCC) investigations showed that variables like temperature, rainfall, snow cover and sea level are changing as a result of climate change. According to IPCCs reports, average temperature of the earth’s surface in twentieth century has increased about 0.6 °C (IPCC 2007). Likewise, snow cover which form water resource of many basins, have decreased by 10 %. Climate change not only affects average values, but also influences the extreme events and frequency of drought and floods especially in Africa and Asia (IPCC 2007). IPCC has also reported that under climate change conditions in twenty first century average annual runoff and water resources are expected to decrease significantly. Increasing intensity and frequency of rainfall in many regions results in higher risk of drought and floods which can affect quality and quantity of water resources (Stern 2007). This shows the necessity of study and analysis of flood frequency under the influence of climate change. Most of the studies on flood frequency in the literature are based on peaks-over-threshold or maximum of mean daily flow in each year (Reynard et al. 2004; Raff et al. 2010; Dobler et al. 2012; Ballinger et al. 2011; Demissie and Cunnane 2002). Instead, in this study we have converted maximum mean daily flow to instantaneous peak flow (IPF) and used it to analyze flood frequency. In recent years, many researchers have been interested in potential effects of climate change. For instance, Raff et al. (2010) selected four basins with different geographical conditions for assessing the impact of climate change
123
1172
on flood frequency. In their study, annual maximum (AM) flows were extracted by a weather generator and rainfall– runoff model. Results showed that for all basins, climate change leads to increase in potential AM flows. In assessment of the climate change effects on flood characteristics in Brahmaputra River Basin (Ghosh and Dutta 2012); analysis time series of simulated flow showed significant increase in peak flow and its duration is expected in future. This investigation also predicted more destructive floods under designed climate change scenarios. Climate change prediction with attention to obscurities of physics of the climate change and conditions and patterns of human development in future is faced considerable uncertainties. So it may be impossible to present exact prediction of climatic variables in future. In this situation, probable and plausible changes in climate are explained by scenarios (Khoramian 2012). Using general circulation models (GCMs) is one of the tools to study climate change under different scenarios. GCMs are numerical models which simulate physical processes in atmosphere, ocean, cryosphere and earth’s surface. These models are the most advanced tool to simulate global climatic systems reaction to increasing in greenhouse gases concentration. Apart from the lack of sufficient certainty involved in the use of GCMs in predicting meteorological parameters, there is another problem which undermines the use of such models. This problem is related to the large-scale outputs of these models. In other words, the GCMs predict metrological data within a large-scale network and for high altitudes of the atmosphere (IPCC 2007). The most common way to tackle this problem is the use of downscaling methods. Based on this method, the GCM outputs are converted to the local scale using data received from weather stations (Semenov 2007; Strauss et al. 2013). Researchers have suggested the different methods for converted downscaled data from the local to regional state so that it would be possible to investigate climate change in large scales (e.g., Cline 2007; Kucharik and Serbin 2008; Goldblum 2009; Winkler 2015). In general two main methods including statistical and dynamical climate modelling are introduced in reviews. Comparison of two mentioned methods are discussed in many studies (e.g., Wilby et al. 2002; Boe´ et al. 2007; Hellstrom et al. 2001). LARS-WG model which is applied in this study, is one of the statistical models used in many researches (e.g., Hashmi et al. 2011; Dibike and Coulibaly 2005; Semenov and Stratonovitch 2010; Zareian et al. 2014). In this study, climate change effects on flood frequency are assessed using a stochastic weather generator (LARS-WG), a distributed hydrological model and estimating IPF.
123
Stoch Environ Res Risk Assess (2017) 31:1171–1182
2 Study area and data set The investigation was performed on Bazoft Basin located in Chaharmahal and Bakhtiari Province, Iran. It extends between 49°340 –50°300 N longitude and 31°370 –32°390 E latitude and covers an area of 2133 km2. Mean annual precipitation is approximately 966 mm and average annual temperature is 10.4 °C. Maximum and minimum of elevation are 4135 and 844 m, respectively (Fig. 1). Average land slope of the basin is 42 %. In present study daily values of precipitation, temperature, evapotranspiration and discharge have been used. Daily precipitation, temperatures and evapotranspiration data was available for three stations inside and nearby the basin. Likewise daily mean discharge and IPF data was available for the water gauge at Marghak (Table 1) which cover the period from 1971 to 2000. Additionally, digital maps including land-use map, soil map (Jalalian and Mohammadi 1997) and digital elevation model (from: http://Gdem.ers dac.jspacesystems.or.jp) were prepared with pixel size of 100 m by 100 m.
3 Materials and methods The investigation was performed at three steps involving (I) weighting GCMs output and downscaling large-scale atmospheric variables, (II) hydrological modelling and (III) flood frequency analysis. Figure 2 gives a schematic representation of the methodology adopted in this study. First, the outputs of four GCMs under two scenarios (A2 and B1) were weighted and then downscaled by applying LARSWG model. Then, Water and Energy Transfer between Soil, plants and atmosphere (WetSpa) model was calibrated on the basis of observed variables and then downscaled data were applied to this model to simulate runoff in future. Finally, future IPF data was estimated and flood frequency analysis was performed based on observed and estimated IPF and results were compared. 3.1 GCM and scenario selection For this study, outputs of four different GCMs from fourth assessment report of IPCC (AR4) are used (Table 2). Moreover; two emission scenarios (A2 and B1) are employed to get the GCMs outputs. GCM outputs were extracted from the data distribution center of IPCC (DDC: http://www.ipcc-data.org/) for three weather stations which are used in this paper. A2 emission scenario stresses the rapid population growth along with trivial technological and economic development until 2100. The B1 storyline and scenario family describes a
Stoch Environ Res Risk Assess (2017) 31:1171–1182
1173
Fig. 1 Bazoft Basin and locations of the selected weather stations
Table 1 Location and coordinates of stations
Stations
Data
Longitude
Latitude
Height (m)
Marghak
Precipitation–temperature evapotranspiration and flow
50°270
31°390
949
32°270
2372
32°170
2048.9
0
Chelgerd
Precipitation
50°7
Shahrekord
Temperature and evapotranspiration
51°500
convergent world with the same global population that peaks in mid-century and declines thereafter, as in the A1 storyline, but with rapid changes in economic structures toward a service and information economy, with reductions in material intensity, and the introduction of clean and resourceefficient technologies. The emphasis is on global solutions to economic, social, and environmental sustainability, including improved equity, but without additional climate initiatives (IPCC 2007).
3.2 Weighting GCMs As the GCM outputs are prepared for both the future and the baseline periods, the duration of 1971–2000 is chosen as baseline for analyzing the GCM outputs (IPCC 2007).
Therefore, the values predicted for temperature and precipitation by each of four GCMs are extracted for the baseline period for all weather stations. Equations (1) and (2) are used to estimate the differences between the values of observed data in weather stations and those of GCM outputs during 1971–2000: B B PEGi ¼ P Pm ; ð1Þ m m Gi
O
PEGi m
show the absolute error of each of GCMs in where estimating precipitation. G represents GCMs and i is the counter of GCMs. P show the average values precipitation within 30-year records. Index B represents the baseline period and index m stands for the month considered. Index O shows the observed data for each of the weather stations during 1971–2000.
123
1174
Stoch Environ Res Risk Assess (2017) 31:1171–1182
SRES
GCMs
A2
GFDL-CM
Hydrological Modelling
Downscaling
HadCM3
WetSpa
LARS-WG B1
BCM 2.0 CGCM3-T
Precipitation
Obs. Data
Temperature
(Precipitation, Temperature, Evapotranspiratio
Flood Frequency Present runoff series
IPF Estimation
Future runoff series
Fig. 2 Flowchart of various stages involved in research study
Table 2 Description of the four GCMs of IPCC’s fourth assessment report (AR4)
WPGi m
¼
1 PEGi m
P15
i¼1
1
Models
Developer
Resolution
References
BCM2.0
BCCR (Norway)
1.9° 9 1.9°
De´que´ et al. (1994)
CGCM3-T63
CCCMA (Canada)
1.9° 9 1.9°
McFarlane et al. (1992)
GFDL-CM2.1
NOAA/GFDL (USA)
2.0° 9 2.5°
Delworth et al. (2006)
HadCM3
UKMO (UK)
2.5° 9 3.75°
Gordon et al. (2000)
:
ð2Þ
PEGim
stations are made for A2 and B1 scenarios in both NF (2020–2049) and FF (2070–2099). 3.4 Hydrological model
WPGi m
shows the weight of each GCM In this equation, in the prediction of precipitation in each month. It is obvious that these weights vary with changing GCMs, weather stations, and different months (Zareian et al. 2014). Then, data for near future (NF) and far future (FF) was provided according to weights of GCMs and using LARS-WG model. 3.3 Stochastic downscaling of data LARS-WG1 is one of the most well-known stochastic weather generators that can produce daily time series of meteorological data. This generator gets the observed meteorological data in baseline period and climate change patterns as inputs and predicts the daily time series of meteorological data in the future. This is based on semi empirical distribution functions which can predict dry and humid periods in the future. To ensure the accuracy of the obtained meteorological data, this model makes some comparisons using v2, t, and F tests (Semenov 2007). In this study, daily precipitation time series in the weather 1
Long Ashton Research Station-weather generator.
123
WetSpa is a distributed physically based hydrological model which is used for predicting the WetSpa on regional or basin scale and daily time step developed in the Vrije Universiteit Brussel, Belgium (Wang et al. 1997; Batelaan et al. 1996). The model represents hydrological system of a basin composed of atmosphere, canopy, root zone, transmission zone and saturation zone. The basin is divided into grid cells to deal with the heterogeneity. Additionally, each cell is divided into two parts including bare soil and vegetated part. Then, the water and energy balance are maintained for each part (Liu and De Smedt 2004). Structure of WetSpa model is represented in Fig. 3. 3.5 Flood frequency analysis Flood frequency analysis is one of the most important fields in statistical hydrology. The main problem in water resource engineering, is estimating flood magnitude of a given return period (Durrans 1992). Primary goal of frequency analysis, is relating the magnitude of extreme events to their frequency by using statistical distributions (Bobee and Robitaualie 1977; Hosking and Wallis 1997).
Stoch Environ Res Risk Assess (2017) 31:1171–1182
1175
where Qmax is IPF (m3 s-1), Q is mean daily peak flow (m3 s-1) and A is watershed area (km2). 3.6.2 Sangal method Sangal (1983), proposed Eq. (4), assuming a triangular unit hydrograph for flow. Qmax ¼ ð4Q2 Q1 Q3 Þ=2;
ð4Þ
where Qmax is IPF (m3 s-1), Q2 mean daily flow at the time when peak flow was occurred, Q1 and Q3 are daily mean flow in previous and next day of peak (m3 s-1). 3.6.3 Fill–Steiner method Fig. 3 Structure of the WetSpa at a pixel cell level (Liu and De Smedt 2004)
In this section, flood frequency analysis has performed using observed IPFs of 36 years. In flood frequency analysis for future, it should be concerned that if only daily data be available, rainfall–runoff models only simulate daily discharge and these models are unable to estimate IPFs due to the lack of maximum intensity of precipitation with different return periods. Additionally, downscaled data for future are provided as daily data. Thus, in this case hydrological models only simulate daily discharge too. So in this study, IPFs have estimated using some empirical equations and the best method to estimate IPF has chosen based on some statistical criterion such as mean absolute error (MAE), root-mean-square error (RMSE), relative absolute error (RAE), R-squared (R2) and Nash–Sutcliffe efficiency (CE). To do this, first, IPFs for historical period estimated using Sangal (1983), Fill–Steiner (2003) and Fullers (1914) method and compared with observed data. According to the results, the best method was applied to estimate IPF in future. Finally, IPFs series were analyzed. Maximum likelihood method which is one of the most efficient methods was applied to fit distributions and estimate parameters. Since this method presents minimum sampling variance from estimated parameters and thus its estimated quantiles are comparable to other methods (Rao and Hamed 2000).
Fill and Steiner (2003) used an equation the same of Sangal (1983) for IPF estimation using daily mean flow in southern Brazil [Eq. (5)]. Qmax ¼ 0:8Q2 þ 0:25ðQ1 þ Q3 Þ=K; 3
ð5Þ
-1
where Qmax is IPF (m s ), Q2 is daily mean flow (m3 s-1) at the time when peak flow was occurred, Q1 and Q3 are daily mean flow in previous and next day of peak (m3 s-1). K is correction factor that is calculated by Eq. (6): K ¼ 0:9123X þ 0:360;
ð6Þ
where X value is obtained from Eq. (7): X ¼ ðQ1 þ Q3 Þ=2Q2 :
ð7Þ
4 Results 4.1 Weighting GCMs Figures 4 and 5 represent average weights of different GCMs in stations. The results of the weighting GCMs in the prediction of precipitation show that all of the GCMs have a weight less than 0.7 (Fig. 4). According to Fig. 5 all GCMs have weights less than 0.75. Also it is found that in some months (especially in June, July and August) HadCM3 has the maximum weight.
3.6 Methods for estimating IPF
4.2 Downscaling model performance
3.6.1 Fuller method
In present study, LARS-WG model is used to downscale daily precipitation and temperature. In this model, different probability distributions are fitted to the data and the best one is chosen. The accuracy of the fitted distributions is checked based on the goodness of fit tests such as v2 test. The low values of v2 show the more conformity of
Fuller (1914), using data from 24 areas in the USA, proposed Eq. (3) for estimating IPF: Qmax ¼ Qð1 þ 2:66A0:3 ; ð3Þ
123
1176
Stoch Environ Res Risk Assess (2017) 31:1171–1182 0.7 BCM2.0
CGCM3-T63
GFDL-CM2.1
HadCM3
weight of model
0.6
Table 3 The statistical details of LARS-WG verification in the Marghak Weather Station Months
0.5
2
v
0.4 0.3 0.2 0.1 0
Temperature
JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
Fig. 4 Weights of the different GCMs to predict the precipitation
1 BCM2.0
CGCM3-T63
GFDL-CM2.1
Precipitation p value
v2
p value
January
0.15
0.91
0.12
0.99
February
0.11
0.99
0.06
1.00
March April
0.11 0.05
0.99 1.00
0.09 0.07
1.00 1.00
May
0.11
0.99
0.11
0.99
June
0.11
0.99
0.12
0.88
July
0.16
0.91
0.12
0.92
August
0.11
0.99
0.10
0.89
September
0.08
1.00
0.11
0.94
October
0.05
1.00
0.14
0.84
November
0.08
1.00
0.09
0.99
December
0.11
0.99
0.12
0.98
HadCM3
0.9
weight of model
0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
Fig. 5 Weights of the different GCMs to predict the temperature
distribution to the observed data. Each distribution generates the statistical parameter (p-value) that should be measured at the significance level. The significance level for this study is considered 0.05. As all p-values obtained for daily time series of temperature and precipitation are greater than 0.05 (as seen in Table 3), good coincidence is achieved for distribution function estimated by LARS-WG and observed values for monthly precipitation and temperature. It means that LARS-WG is able to downscale data with good accuracy. As an example, values of v2 test and p-value for Marghak Station is presented in Table 3. 4.3 Hydrological model performance WetSpa is calibrated by adjusting 11 global parameters of model (Table 4). These parameters are described in WetSpa manual (Liu and De Smedt 2004). Model performance during calibration and validation is evaluated based on some statistical criterion like Nash–Sutcliffe coefficient,
123
Nash–Sutcliffe coefficient for low and high flows and bias (Table 5). Among these criterions, the Nash–Sutcliffe coefficient (1970) is of particular importance. Figures 6 and 7 show results of calibration and validation. It is worth noting that data from the beginning of 1981 to the end of 1984 is used in calibration process and data from 1985 to 1987 is used for validation. Comparing observed and simulated hydrographs and values of statistical criterion, shows the model ability in simulating hydrological processes of the basin. Therefore the model is applied for simulating future runoff using downscaled climatic data as input. 4.4 Statistical distribution selection Flood frequency analysis was performed using 36 years of observed IPF data by maximum likelihood method and different distributions were fitted to the series of data. For the v2 and Kolmogorov–Smirnov (KS) goodness of fit procedure, the acceptability of the distribution functions are selected on the basis of a p value with a confidence level of 95 % (a \ 0.05). The null hypothesis will therefore be rejected if the p-value is smaller than the 5 % significance level (Hosking and Wallis 1997). According to the p-values (based on v2 test = 0.875 and KS test = 0.886), graphs, and standard errors, Pearson III was chosen as a best distribution (Fig. 8) (Almasi 2014). Afterward, frequency analysis was performed for future series. Results showed that Pearson III is better than other distributions for future and under both A2 and B1 scenarios (Fig. 9). It means that distribution does not change during different time periods.
Stoch Environ Res Risk Assess (2017) 31:1171–1182 Table 4 Global parameters of WetSpa and their final values
1177
Parameters
Calibrated values
Correction factor for potential evapotranspiration
0.1
Scaling factor for interflow computation
0.1
Groundwater recession coefficient
0.016
Initial soil moisture
1.2
Initial groundwater storage
100
Base temperature for snowmelt
1.1
Temperature degree-day coefficient
1.1
Rainfall degree-day coefficient
1.1
Surface runoff exponent for a near zero rainfall intensity
1.1
Rainfall intensity corresponding to a surface runoff exponent of 1
7
Maximum rainfall intensity
160
Table 5 Values of statistical evaluation of WetSpa during calibration and validation Criterion
Calibration
Validation
Nash–Sutcliffe
0.63
0.65
Nash–Sutcliffe for low flow
0.71
0.75
Nash–Sutcliffe for high flow
0.65
0.65
Bias
0.095
0.071
Fig. 7 Comparison of observed and simulated hydrographs during validation
Fig. 6 Comparison of observed and simulated hydrographs during calibration
4.5 IPF estimation for future As it mentioned in previous section, IPF was estimated using Fuller, Sangal and Fill–Steiner methods. These methods have compared based on statistical criterion which named in Sect. 3.5. Comparison the results of these methods in historical period showed that Fill–Steiner method has more precision in estimating IPF in Bazoft Basin (Table 6). Therefore this method was applied to estimate IPF in future. Studies about evaluation of
Fig. 8 Fitting Pearson III distribution to observed IPF
experimental methods in different regions of Iran, indicates that some of these methods have satisfactory performance in IPF estimations. For example, Dastorani et al. (2011)
123
1178
Stoch Environ Res Risk Assess (2017) 31:1171–1182 Table 6 Statistical comparison of Fuller, Sangal and Fill–Steiner formulas Method Criterion
Fill–Steiner
Sangal
Fuller
MAE (m3 s-1)
146
145
199.491
RMSE (m3 s-1)
205
215
248.319
RAE (–) R2 (–)
0.277 0.927
0.275 0.925
0.436 0.912
CE (–)
0.899
0.889
0.8027
in both NF and FF and maximum and minimum reduction is occurred under B1FF and B1NF, respectively.
5 Discussion
Fig. 9 Fitting Pearson III distribution to estimated IPFs in near future under A2 (a) and B1 scenarios (b)
compared Fuller, Sangal and Fill–Steiner methods in 12 stations with different climate and topography in Iran. Two out of 12 stations (Ghale-Shahrokh and Sarabe Hendeh) locate near Bazoft Basin and have similar conditions with stations applied in our study. Results of this study confirm the ability of Fill–Steiner method in IPF estimation in mentioned stations. So that correlation coefficient is about 0.807 and 0.980 in Ghale-Shahrokh and Sarabe Hendeh, respectively. Then Pearson III distribution was fitted to estimate IPF resulted from Fill–Steiner equation. Figure 10 shows IPFs and AM values in different return periods (T) for base line and future periods (A2NF, A2FF, B1NF and B1FF). Likewise, IPFs relative changes from base line are estimated in Table 7. According to Table 7, IPFs will decrease
123
Assessment of the climate change impacts on hydrological processes have widely performed in recent decades. GCMs are the most advanced tool to simulate global climatic systems reaction to increasing in greenhouse gases concentration. GCMs are limited in their utility for local impact studies by their coarse spatial resolution. To cope with this limitation, downscaling techniques are applied to convert large scale atmospheric variables to the local scale. In general downscaling techniques have grouped into two types including statistical and dynamical methods that each one has advantages and disadvantages. For instance, statistical downscaling models are cheap, computationally undemanding, readily transferable and applicable to exotic predictands such as air quality and wave heights. Also, ensembles of climate scenarios permit uncertainty analyses. On the other hand, statistical models have some weakness such as dependence on the realism of GCM boundary forcing and high quality data requirement for model calibration. Moreover, in these models choice of domain size and location affects results, predictor–predictand relationships are often non-stationary, choice of predictor variables and empirical transfer scheme affects results. Dynamical downscaling has advantages and disadvantages too. For example, dynamical methods respond in physically consistent ways to different external forcings, resolve atmospheric processes such as orographic precipitation and they are consistent with GCMs. But in addition to dependence on the realism of GCM boundary forcing and effects of choice of domain size and location on results (that are considered as weakness of statistical models), dynamical methods have different weakness too. They require significant computing resources and Initial boundary conditions affect results in dynamical models (Wilby et al. 2002).
Stoch Environ Res Risk Assess (2017) 31:1171–1182 Fig. 10 Estimated IPF in different return periods by Pearson III distribution
1179
4500 4000 obs-AM 3500
A2FF-AM B1FF-AM
QT(M3/S)
3000
A2NF-AM 2500 B1NFAM 2000
obs-IPF A2FF-IPF
1500
B1FF-IPF 1000
A2NF-IPF
500
B1NF-IPF
0 2
5
10
20
50
100
200
1000
Return Period (year)
Table 7 Estimated IPF (m3 s-1) using Pearson III distribution and relative changes from base line Return periods
Base line
A2 Near future
B1 Far future
Near future
Far future
Percent change of A2
Percent change of B1
Near future
Near future
Far future
Far future
2
1542
402
579
1093
345
-73
-62
-29
-78
5
2151
503
875
1633
374
-77
-59
-24
-83
10
2505
581
1083
1997
389
-77
-57
-20
-84
20
2817
661
1285
2345
401
-76
-54
-17
-86
50
3190
769
1550
2791
416
-76
-51
-12
-87
100
3451
853
1749
3123
425
-75
-49
-9
-87
200
3699
938
1949
3451
434
-74
-47
-7
-88
1000
4239
1142
2414
4208
453
-73
-34
-0.7
-89
With due attention to uncertainties related to GCMs, in present study four GCMs are applied through a weighting approach. Results demonstrated that GFDL-CM2.1 has the highest annual weight for predicting precipitation. It meant data of this model are more similar to observed data rather than other GCMs data. Besides, HadCM3 is determined as a model with highest weights in simulating of temperature. Comparison of weighting GCMs with single GCM (like HadCM3 that has used in Bazoft Basin) shows that using different GCMs can improve the precision and accuracy of prediction and there is significant difference between results of two methods. In this paper a stochastic weather generator (LARSWG) was applied to downscale large scale climatic predictors. Results showed that this model is efficient in downscaling temperature and precipitation. The main advantage of this technique is that it can exactly reproduce many observed climate statistics and has been widely used, particularly for agricultural impact assessment.
Furthermore, stochastic weather generators enable the efficient production of large ensembles of scenarios for risk analysis (Wilby et al. 2002). Moreover, in the present study; WetSpa model was applied as a hydrological model to simulate surface runoff. With due attention to the value of Nash–Sutcliffe efficiency (Table 3) and according to the model performance categories provided by Porretta et al. (2010) to indicate the goodness level, the performance of this model was very good in Bazoft Basin. Since the output of hydrological models are daily runoff, in most studies about climate change impacts on flood frequency (Raff et al. 2010; Dobler et al. 2012; Ballinger et al. 2011), AM flows have been analyzed; whereas, IPFs should be considered for flood frequency analysis. In fact, the main objective of this study is finding the right solution to determine the impact of climate change on flood frequency on the basis of IPF estimation instead of AM. It is known that hydraulic structures and flood control projects are designed based on IPF. So using daily or AM flow may
123
1180
lead to underestimate the design flood and so financial losses are expected. This point is clarified in Fig. 10 that in base line and future periods always QT in IPF series is higher than AM series. Thus for future planning with attention to climate change, that’s an important point to estimate flood magnitude correctly. For this purpose, in this study empirical equations such as Sangal, Fuller and Fill– Steiner methods were applied to estimate IPF in future. Comparisons of these methods show that Fill–Steiner method is more efficient rather than Sangal and Fullers methods. Fuller’s method has less efficiency in estimating IPF (Table 6). Because in Eq. (3), only drainage area is considered by Fuller and this factor cannot explain the hydrological and physical characteristics of a catchment completely. In other words, many factors are involved in determining the flow rate of the river. For instance, parameters such as slope, land use, soil texture are important factors which control the amount of surface runoff and flow of the river. As these important factors are variable in different parts of the watershed, especially in watershed with considerable areas, factor A in Fuller’s equation can’t reflect hydrological behavior of the basin lonely. Thus this factor cannot express converting daily flow to IPF. In addition with regard to Table 6, there is no significant difference between the results of Sangal and Fill–Steiner methods. Because in large basins with considerable drainage area, results of these methods are close to each other. However, as Fill–Steiner method represented slightly better results in this investigation (Table 6), it was selected to estimate IPF. It is worth noting that for IPF estimation using mentioned equations, physical characteristics of catchment is assumed to be fixed in future and only the land use parameter will change that is considered in definition of climate change scenarios. Hence according to the better performance of Fill–Steiner method in observed period, it is assumed that this method has more efficiency in estimating future IPF too. Results of flood frequency analysis indicate that selected distribution is the best one for future analysis too. Unchanging distribution over time may be due to the basin topography. Since the basin is located in mountainous area and in these regions IPF does not have severe change compared to the base period and it leads to stable distribution during time, but values of different return periods have changed. In other words, due to the constant physical characteristics of the basin, flood frequency curve has the same trend in both base line and future, but it occurs in lower magnitude. This can be related to climate change effects that reduce precipitation in Bazoft Basin. Despite the fact that A2 is more pessimistic scenario than B1, there are larger changes in IPF in FF under B1
123
Stoch Environ Res Risk Assess (2017) 31:1171–1182
scenario compared with those under A2 scenario in Bazoft Basin (Table 7). This is due to the temporal distribution and amounts of rainfall. Assessing daily rainfall values in stations represent high coefficient of variation for this variable. For example, it seen that in Chelgerd Station, maximum and minimum values of precipitation are 303 mm (is recorded in 1976, March) and 0 (in many days), respectively. Also, it is known that LARS-WG uses fitted distribution function to observed daily precipitation based on climate change scenarios to construct daily rainfall in NF and FF. So for FF under A2 scenario, maximum of daily rainfall is estimated 175.4 and 70 mm in Chelgerd and Marghak Stations, respectively. But for NF under this scenario, 84 and 21 mm is estimated for two mentioned stations. Results of rainfall projection are different for B1. So as for B1NF maximum of daily rainfall is 293 and 90 mm and for B1FF is 42 and 31 mm in Chelgerd and Marghak Stations, respectively. Since these values are the base for simulating daily discharge in WetSpa model, and also maximum daily runoff gets from these values, in converting to IPF mentioned values are maximized and show their effects in different return periods in a same way. Thus, difference between A2FF and B1FF is related to values of daily rainfall. However, in many studies which have been done in a place near the Bazoft Basin, e.g., Western Zayandeh-Rud River Basin and Hamedan–Bahar Plain (Taleb Mored 2012; Zareian 2015; Saeidi 2014) similar results have achieved.
6 Conclusion Applying GCMs through a weighting method reduce uncertainties and thus represent more accurate results in estimating effects of climate change on temperature, precipitation and also runoff and floods in basin. To deal with limitation of GCMs coarse resolution, downscaling is one of the essential parts of climate change impact studies. In present study LARS-WG model was applied to downscale temperature and precipitation. Since LARS-WG model estimate future temperature and precipitation by fitting best distribution to mentioned variables in base line, can represent accurate results. However, using other statistical downscaling models or even dynamical methods in Bazoft Basin can be evaluated in future studies. For runoff simulation WetSpa model was applied to the basin. WetSpa is a distributed model that divides basin to grid cells. Therefore, precision of its results depends on cell size of input raster maps. Therefore, if input raster cell size be very fine or coarse, some hydrological processes may not be considered. Using AM instead of IPF in future flood frequency analysis based on climate change scenarios may make
Stoch Environ Res Risk Assess (2017) 31:1171–1182
errors in estimating design flood. Therefore, it is recommended to use some methods such as Fill–Steiner or artificial neural network to convert daily flow to IPF. Because it leads to better and more accurate results and can be used in designing hydraulic structures. On the whole, results of the present study show that in NF and FF and under both A2 and B1 scenarios, Bazoft Basin will face higher temperature and lower precipitation. It means that the climatic condition of the basin is changing through more arid situation.
References Almasi P (2014) Assessment of the climate change impacts on flood frequency in Bazoft Basin. MSc Thesis, Isfahan University of Technology, Iran Arnell NW, Reynard NS (1996) The effects of climate change due to global warming on river flows in Great Britain. J Hydrol 183:397–424 Ballinger J, Jackson B, Reisinger A, Stokes K (2011) The potential effects of climate change on flood frequency in the Hutt River. NZCCRI-2011-03 Batelaan O, Wang ZM, de Smedt F (1996) An adaptive GIS toolbox for hydrological modelling. In: Kovar K, Nachtnebel HP (eds) Application of geographic information systems in hydrology and water resources management. IAHS Publ. no. 235, p 3–9 Bobee B, Robitaualie R (1977) The use of the Pearson type 3 and log Pearson type 3 distributions revisited. Water Resour Res 13(2):427–443 Boe´ J, Terray L, Habets F, Martin E (2007) Statistical and dynamical downscaling of the Seine Basin climate for hydro-meteorological studies. Int J Climatol 27(12):1643–1656 Cameron D, Beven K, Tawn J, Naden P (2000) Flood frequency estimation by continuous simulation under climate change (with uncertainty). Hydrol Earth Syst Sci 4(3):393–405 Cline WR (2007) Global warming and agriculture: impact estimates by country. Peterson Institute, Washington, DC Dastorani M, Salimi Koochi J, Talebi A, Abghari H (2011) Evaluation of the applicability of some methods to estimate Instantaneous Peak Flow using daily flow data. J Range Watershed 84(1):25–37 (in Persian) Delworth TL, Broccoli AJ, Rosati A, Stouffer RJ, Balaji V, Beesley JA (2006) GFDL’s CM2 global coupled climate models, part I: formulation and simulation characteristics. J Clim 19:643–674 Demissie SS, Cunnane C (2002) Representation of climate change in flood frequency estimation, in: Celtic water in European framework—pointing the way to quality. In: Proceedings of the 3rd inter-Celtic colloquium on hydrology and management of water resources. National University of Ireland, Galway De´que´ M, Dreveton C, Braun A, Cariolle D (1994) The ARPEGE/IFS atmosphere model: a contribution to the French community climate modelling. Clim Dyn 10(4–5):249–266 Dibike YB, Coulibaly P (2005) Hydrologic impact of climate change in the Saguenay watershed: comparison of downscaling methods and hydrologic models. J Hydrol 307(1):145–163 Dobler C, Burger G, Stotter J (2012) Assessment of the climate change impacts on flood hazard potential in the Alpine Lech watershed. J Hydrol 460:29–39 Durrans SR (1992) Parameter estimation for the Pearson type 3 distribution using order statistics. J Hydrol Eng 133:215–232
1181 Fill HD, Steiner AA (2003) Estimating IPF from mean daily flow data. J Hydrol Eng ASCE 8(6):365–369 Fuller WE (1914) Flood flows. Trans Am Soc Civ Eng 77:564–617 Ghosh Sh, Dutta S (2012) Impact of climate change on flood characteristics in Brahmaputra Basin using a macro-scale distributed hydrological model. Department of Civil Engineering, Indian Institute of Technology, Guwahati Goldblum D (2009) Sensitivity of corn and soybean yield in Illinois to air temperature and precipitation: the potential impact of future climate change. Phys Geogr 30(1):27–42 Gordon C, Cooper C, Senior CA, Banks H, Gregory JM, Johns TC, Wood RA (2000) The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Clim Dyn 16(2–3):147–168 Hashmi MZ, Shamseldin AY, Melville BW (2011) Comparison of SDSM and LARS-WG for simulation and downscaling of extreme precipitation events in a watershed. Stoch Environ Res Risk Assess 25(4):475–484 Hellstrom C, Chen D, Achberger C, Raisanen J (2001) Comparison of climate change scenarios for Sweden based on statistical and dynamical downscaling of monthly precipitation. Clim Res 19(1):45–55 Hosking JRM, Wallis JR (1997) Regional frequency analysis an approach based on L-moment. Cambridge University, Cambridge Hulme M, Barrow EM, Arnell NW, Harrison PA, Johns TC, Downing TE (1999) Relative impacts of human-induced climate change and natural climate variability. Nature 397:689–691 IPCC (2007) Climate change 2007: the physical science basis. Cambridge University Press, Cambridge Jalalian A, Mohammadi M (1997) Land suitability of North Karoon Basin. Iran Ministry of Agriculture Khoramian A (2012) Assessment of the impacts of climate change on snow melting runoff process in upper region of Zayanderud Basin. MSc Thesis, Isfahan University of Technology, Iran Kucharik CJ, Serbin SP (2008) Impacts of recent climate change on Wisconsin corn and soybean yield trends. Environ Res Lett 3(3):034003 Liu YB, De Smedt F (2004) WetSpa extension, a GIS-based hydrologic model for flood prediction and watershed management, documentation and user manual. Department of Hydrology and Hydraulic Engineering, Vrije universiteity Brussel, Brussels McFarlane NA, Boer GJ, Blanchet JP, Lazare M (1992) The Canadian Climate Centre second-generation general circulation model and its equilibrium climate. J Clim 5(10):1013–1044 Nash JE, Sutcliffe JV (1970) River flow forecasting through conceptual model. J Hydrol 10:282–290 NERC (1975) Flood studies report. London Porretta L, Chormanski J, Ignar S, Okruszko T, Brandyk A, Szymczak T, Krezalek K (2010) Evaluation and verification of the WetSpa model based on selected rural catchments in Poland. J Water Land Dev 14:115–133 Raff DA, Sutley D, Pruitt T, Brekke LD (2010) Flood frequency in a changing climate, projections approach and diagnostics. In: 2nd Joint Federal Interagency conference, Las Vegas Rao R, Hamed KH (2000) Flood frequency analysis. CRC Press, Boca Raton Reynard N, Crooks S, Wiley R, Kay A (2004) Impact of climate change on flood flows in river catchments. Final report for Defra/ EA project W5B-01-050 Saeidi H (2014) Climate change impact on low flows in ZayandehRud River Basin. MSc Thesis, Isfahan University of Technology, Iran Sangal BP (1983) Practical method of estimating peak flow. J Hydraul Eng 109(4):549–563
123
1182 Semenov MA (2007) Development of high-resolution UKCIP02based climate change scenarios in the UK. Agric For Meteorol 144(1):127–138 Semenov MA, Stratonovitch P (2010) Use of multi-model ensembles from global climate models for assessment of climate change impacts. Clim Res 41(1):1 (open access for articles 4 years old and older) Stern N (2007) The economics of climate change: the Stern review. Cambridge University Press, Cambridge Strauss F, Formayer H, Schmid E (2013) High resolution climate data for Austria in the period 2008–2040 from a statistical climate change model. Int J Climatol 33(2):430–443 Taleb Mored H (2012) Climate change effects on surface and ground water resources using hydrological model HydroGeoSphere in Hamedan–Bahar Plain. MSc Thesis, Isfahan University of Technology, Iran Wang Z, Batelaan O, De Smedt F (1997) A distributed model for water and energy transfer between soil, plants and atmosphere. J Phys Chem Earth 21:189–193
123
Stoch Environ Res Risk Assess (2017) 31:1171–1182 Wigley TML, Raper SCB (1992) Implications for climate and sea level of revised IPCC emissions scenarios. Nature 357:293–300 Wilby RL, Dawson CW, Barrow EM (2002) SDSM—a decision support tool for the assessment of regional climate change impacts. Environ Model Softw 17(2):145–157 Winkler JA (2015) Selection of climate information for regional climate change assessments using regionalization techniques: an example for the Upper Great Lakes Region, USA. Int J Climatol 35(6):1027–1040 Zareian MJ (2015) Investigation of water allocation in Zayandehrud Dam under climate change effects with a view to optimizing water resources and use. PhD Thesis, Isfahan University of Technology, Iran Zareian MJ, Eslamian S, Safavi HR (2014) A modified regionalization weighting approach for climate change impact assessment at watershed scale. Theor Appl Climatol 122:497–516