Climate change and water resources in arid regions: uncertainty of the baseline time period

Recent climate change studies have given a lot of attention to the uncertainty that stems from general circulation models (GCM), greenhouse gas emission scenarios, hydrological models and downscaling approaches. Yet, the uncertainty that stems from the selection of the baseline period has not been studied. Accordingly, the main research question is as follows: What would be the differences and/or the similarities in the evaluation of climate change impacts between the GCM and the delta perturbation scenarios using different baseline periods? This article addresses this issue through comparison of the results of two different baseline periods, investigating the uncertainties in evaluating climate change impact on the hydrological characteristics of arid regions. The Lower Zab River Basin (Northern Iraq) has been selected as a representative case study. The research outcomes show that the considered baseline periods suggest increases and decreases in the temperature and precipitation (P), respectively, over the 2020, 2050 and 2080 periods. The two climatic scenarios are likely to lead to similar reductions in the reservoir mean monthly flows, and subsequently, their maximum discharge is approximately identical. The predicted reduction in the inflow for the 2080–2099 time period fluctuates between 31 and 49% based on SRA1B and SRA2 scenarios, respectively. The delta perturbation scenario permits the sensitivity of the climatic models to be clearly determined compared to the GCM. The former allows for a wide variety of likely climate change scenarios at the regional level and are easier to generate and apply so that they could complement the latter.


Background
The Breference climate^or Bbaseline^time period can be defined as a period of time from which the potential climate change projections are estimated. The establishment of a baseline period is commonly required for climate change and anthropogenic intervention impact studies. The baseline time period is a key factor in such studies and should be selected and gathered at the beginning of a project so that the potential future impacts can be evaluated against the previous situation; therefore, it serves as a benchmark for project success. The selection of the baseline has often been governed by the availability of the required climate data in addition to the study purpose.
The earlier baseline periods are preferred over later ones (IPCC 2007) for many reasons; for example, later periods such as 1961 to 1990 are likely to have larger trends incorporated with climate data, in particular, the effects of sulphate aerosols over Europe and eastern USA (IPCC 2007). Accordingly, the best baseline period would be in the nineteenth century when human-induced impacts on global climate were insignificant. However, most impact studies aim to determine the influence of climate change with respect to the presence, and therefore, recent baseline periods are preferred (Semenov and Stratonovitch 2010). Furthermore, different approaches have different baseline requirements; the adequacy of the observed baseline can only be evaluated based on the construction methods of a specific climatic scenario. To apply the delta perturbation (DP) scenarios, many researchers (Tigkas et al. 2012;Al-Faraj and Scholz 2014;Mohammed and Scholz 2017c) used, as a baseline time period, certain periods that represent normal climatic conditions rather than using the entire available datasets. For example, in order to develop preparedness strategies to address the consequences of climate change such as drought phenomena, Mohammed and Scholz (2017c) utilised the DP climate scenario. For the DP scenario application, they defined the normal climatic time period through calculating the RDI values based on the entire datasets. Then, they selected the time period during which the RDI values are close to zero, which then can be considered for climate change and simulation studies.
Moreover, for the assessment of the impact of anthropogenic interventions on surface runoff, the concept and the identification of the baseline or the Bpre-anomalyp eriod is different. If the studied hydrological regime has witnessed a sudden change such as a dam construction, then the Indicator of the Hydrological Alteration (IHA) can be applied to test how the flow system has changed through calculating the parameters for two time periods (before and after the alteration). However, if the flow system has experienced long-term human modifications, IHA can be used to evaluate changing trends. Additionally, in order to determine the artificial critical change points, many statistical (e.g. non-parametric analysis) and graphical (double mass curves, single mass curves and flow duration curves) methods are often used (Cheng et al. 2016;Mao et al. 2015;. In order to quantify both climate change and anthropogenic intervention impacts on the streamflow,  applied multi-regression, hydrologic model and hydrologic sensitivity simulation assessments. They used Pettitt, Prunoff double cumulative curve (PR-DCC) and Mann-Kendall techniques for the change points and significant trend assessments. The long-term runoff series from 1979 to 2013 was divided into baseline (1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997) and anthropogenic intervention (1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013) periods.
In general, for climate change impact quantification, the uncertainty that stems from GCM is considered to be the largest. Still, the uncertainty related to the baseline must be taken into consideration for better evaluation of the climate change impact. There is a need to calculate the effects of climate change on the hydrological characteristics of representative example case studies such as the Lower Zab River Basin, while exploring the uncertainties linked to the baseline time period using two time periods such as 1988-2000 and 1980-2010 in this case.

Aim, objectives and novelty
This study evaluated the climate change impacts on the flow within an example case study (Lower Zab River, Iraq), considering the uncertainty that stems from the baseline time period in addition to many other sources such as GCM, greenhouse gas emission scenarios, and DP procedures. This has been achieved by comparing the results of an ensemble of seven GCM methods under the Special Reports on Emission Scenarios SRA2 and SRA1B (Semenov and Stratonovitch 2010) using two different baseline time periods, with the perturbations of P (dP of 0-40%; 2% step) and potential evapotranspiration (PET) (dPET of 0-30%; 2% step). The comparison aims to explore how close these scenarios are and how different they are from each other in terms of their ability to simulate streamflow.
This research can be considered as an initial effort to answer the following open research question: What would be the differences and/or the similarities in the evaluation of climate change impact between the DP and the GCM scenarios using different baseline time periods? This is a remarkable question, as practically, all earlier climate change evaluation studies are generally based on only one baseline time period without taking into consideration the uncertainty that stems from using different baseline periods. Figure 1 shows the suggested procedure for the modelling approach of the evaluation of the potential influences of climate change on the water resource system. The main objectives are to (1) investigate the relevance of LARS-WG5.5 (Semenov and Stratonovitch 2010) for downscaling weather data at the LZRB and (2) to estimate the future variation and the uncertainties in the prediction of surface inflow to the basin storage system that could stem from: (a) climate change scenarios, which are GCM and DP; (b) emission scenarios SRA2 and SRA1B; and (c) the baseline time periods 1988-2000 and 1980-2010. The SRA2 and SRA1B scenarios define a locally oriented economy leading to a heterogeneous world and a balanced technology and fossil fueloriented rapid growth, respectively. The outcomes of this study will be beneficial to investigate the effect of climate change on the inflow to the basin storage system in arid climatic condition.

Data collection
The following data have been collected: (1) Daily weather data such as P as well as maximum and minimum air temperature from six gauging stations were obtained for the water year range between 1979/80 and 2012/13. The weather gauging stations are distributed over the upper part sub-basin of the study area with elevations varying from 651 to 1536 m (Table 1 Table 3 Carbon dioxide (CO 2 ) concentrations for selected climate scenarios specified in the Special Report on Emissions Scenarios (adopted from Semenov and Stratonovitch (2010)) Scenario Key assumptions CO 2 concentration (parts per million) T 1 T 2 T 3 B 1 : the sustainable world Quick alteration in economic organisations; dematerialisation comprising developed equity and ecological concern; global concern regarding ecological and social sustainability; more effort in introducing clean technologies; world population extends to 7 billion by 2100 Facility (GLCF, Global and Land Cover Facility, 2015) databases, respectively.

Data analysis
The following analysis have been performed: (1) For the hydro-climatic gauging station location projections, Theissen network, basin boundary and streamflow delineations, ArcGIS 10.3 software has been applied.
(2) For the analysis of the daily hydro-climatic datasets such as trend, monthly and annual amounts, modifications and filling of data gaps were estimated by the Statistical Program for Social Sciences program (SPSS) 23 (ITS 2016

Representative study example
The proposed methodology was applied to the Lower (Lesser/ Little) Zab River, which is considered one of the most important tributaries of the Tigris River stream. The Lower Zab River is situated in Erbeel Governorate. The watercourse network is located between latitudes 36°50′ N and 35°20′ N and longitudes 43°25′ E and 45°50′ E  as shown in Fig. 2. Dokan is the most important dam that has been constructed on the Iraqi part of the shared basin. The dam is a multi-function arch dam with an extreme storage size of approximately 6.970 billion cubic meters (BCM), a top elevation of 116 m above the river bed (516 m) and a stretch of 360 m ). The LZRB has been selected as an example basin for arid and semi-arid areas. The nature of the study is common to other shared river basins such as the Senegal, Volta and Rhine, where challenges of shared sustainable water resource use are anticipated to exacerbate under the collective impact of climate change (Mohammed and Scholz 2017a, b). The upstream and downstream developments vary widely. This suggests a varied range of uncertainties in climate change impacts on water resources availability.

Downscaling model
The Long Ashton Research Station Weather Generator (LARS-WG5.5) has been used to predict meteorological variables at a single site under the current and future climate conditions. The process of artificial climate generation is divided into three stages (Semenov and Stratonovitch 2010): (1) Site analysis: this function is used for the model calibration by analysis of the statistical properties of the observed weather variables. The statistical properties are saved in two files; (2) The Q Test function is used for model validation during which the statistical properties of the observed and the generated weather variables are used to examine if there are any statistically significant variations; and (3) The generator: this function is used for the generation of artificial weather variables. The files resulting from the calibration of the model are used to produce synthetic weather variables having the same statistical properties as the observed ones, but are different on a day-to-day basis. The climate scenarios in this research are based on The AR5 report brings decision-makers up-to-date on the state of climate science. Since it is released in stages, it is the most complete evaluation of existing climate change studies and gives a baseline for understanding and future actions. Moreover, among the highlights of the AR5 report, the conclusion that much of the warming over the past five decades resulted from anthropogenic intervention is now Bextremely likely^(Bvery likely^in the AR4 report). Predictions of the future rise in sea level have been considerably improved, because of the enhancement in the understanding of the ice sheet movements in temperate weather. Additionally, the Arctic Ocean is now projected to be ice-free during the summer by the mid-century (end-of-century in AR4 report) under a high greenhouse gas emission scenario.
However, there are no remarkable changes in the AR5 report, which are linked to previous models released as part of the AR4 report and earlier assessment reports Nevertheless, there is a noticeable rise in the volume of data and a steady  Fig. 4b-h): b 10% reduction in precipitation (P); d 20% reduction in P; f 30% reduction in P; and h 40% reduction in P, compared to the baseline 2 (1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)  increase in the number of modelling research groups providing their technical perspectives to the modelling community. Accordingly, this research utilised LARS5.5 for downscaling weather data over the study area based on the reasoning of the IPCC AR4 report.

Weather parameters and reservoir inflow
Thirty hydrological years were utilised for the calibration and validation of the LARS-WG5.5 model. To assess the model performance, some graphical comparisons and statistical analysis were applied. For the analysis of the equivalence of the wet/dry series periodic distributions, daily rainfall distributions as well as daily minimum and maximum temperature distributions, the Kolmogorov-Smirnov (K-S) method (Chen et al. 2013) was applied. The p value was used as an indicator, if there is a significant change between the simulated and the observed climatic parameters. A high K-S value and a very low p value mean that the simulated climate is unlikely to be identical to the observed one (Semenov et al. 2013). The obtained results from the LARS-WG model proved that the model performs well in producing weather data for most stations. Accordingly, the model can be applied to predict daily meteorological parameters for the stations for the time periods 2011-2030, 2046-2065 and 2080-2099, subject to seven ensembles of GCM as well as the SRA2 and SRA1B scenarios. Compared to the 1980-2010 baseline period, there is a rising trend in T min and T max and a declining trend in P values. The corresponding monthly values vary from month to month. The maximum increases in the predicted variables were 3.02 and 3.33°C, 3.17 and 3.70°C and 17.33 and 21.93 mm for the two considered emission scenarios, respectively. Figures 4 and 5 show the timing and the magnitude of the streamflow hydrographs simulated with meteorological data predicted by seven GCM that were downscaled using LARS-WG5.5 and DP for climate change scenarios, respectively. In order to avoid any bias resulting from the hydrological simulation process, the streamflow for the baseline periods, whether it is baseline 1  or baseline 2 (1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000), is represented by modelled flow. The results indicate how climate change might cause a reduction in both timing and magnitude of the inflow hydrograph to the reservoir. Both the GCM and DP climatic scenarios predict nearly the same declines in the mean monthly flows, and subsequently, their peak points are approximately the same. Figure 4 shows a declining trend in the inflow peaks fluctuating from 3% (INCM4) to 21% (GFCM21) as shown in Fig. 4a, 9% (CNCM3) to 39% (GFCM21) as presented in Fig. 4c and 21% (NCCCM) to 42% (GFCM21) as highlighted in Fig. 4e for the three future time periods, respectively.
Further, to compare the results of the GCM and DP scenarios, the following has been highlighted: Firstly, based on the SRA1B emission scenario, Fig. 4g shows that the inflow to the reservoir is expected to decrease by about 10% for the 2011-2030 horizon, which is identical to the anticipated reservoir inflow applying DP (10% P decrease and 0% PET) as Fig. 4b shows. Figure 4g highlights that the equivalent decrease for the time period 2046-2065 is expected to be approximately 25%, which is identical to the predicted decrease in streamflow applying DP (20% decrease in P and 0% PET) as indicated by Fig. 4d. However, the predicted decline in the reservoir inflow for the 2080-2099 horizon is almost 32% (Fig. 4g), which is identical to the hydrograph that is shown in Fig. 4b (10% P reduction and 30% increase in PET).
Secondly, and based on the SRA2, all GCM show decreases in peak values ranging between 4% (IPCM4) and 19% (GFCM21) as presented in Fig. 5a, 16% (HADGM3) to 40% (GFCM21) as Fig. 5c shows, and 13% (MPEH5) to 56% (GFCM21) as highlighted in Fig. 5e for the three time horizons in this order. It is expected that the reservoir inflow will decline by approximately 13% by the 2020 horizon, Fig.  5g shows that the reservoir inflow is equal to the anticipated inflow by DP (10% P decrease and 0% PET) as Fig. 5b demonstrates. The equivalent decrease for the time period 2046-2065 is expected to be nearly 25%, which is identical to the predicted decrease in streamflow by DP (20% decrease in P and 0% PET) as indicated in Fig. 5d. However, the predicted decline in the reservoir inflow for the 2080-2099 horizon is almost 36% (Fig. 5g), which is identical to the hydrograph shown by Fig. 5f (30% P reduction and 0% increase in PET). Moreover, Figs. 4 and 5a, c, e, and g confirm that maximum discharges were observed earlier than for the reference period (1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000), since the lag was about 8 days for nearly all the future time periods. Whereas there was not any variation in the time to maximum flowrate that was anticipated by the DP scenario, as indicated by Figs. 4 and 5b-h.  Figure 6a shows that during the 2020 time horizon, the seven GCM predict the inflow similarly. However, there is an evident  (Fig. 6b, c), and all the time periods (Figs. 7a-c and 8a-c), respectively. This uncertainty stems from both GCM and emission scenarios.

Uncertainty of reservoir inflow
Furthermore, the simulation results discussed how the SRA2 and SRA1B emission scenarios predict approximately the same decreases in the mean monthly flows, and consequently, their peak values are almost the same, in particular for the 2011-2030 and 2046-2065 time periods (Fig. 9a, b). There is no great variation in the predicted values of the reservoir inflows for the 2020s and 2050s, since the two emission scenarios show decreases in peak values ranging between 6% (SRA2) and 10% (SRA1B) as shown in Fig. 9a, and between 21% (SRA2) and 25% (SRA1B) as indicated by Fig. 9b. However, there is a clear variation in the predicted inflow using the two considered emission scenarios for the 2080-2099 time period. The variation altered from 31% (SRA1B) to 49% (SRA2) as illustrated in Fig. 9c.
By the 2020 horizon, there will be a decrease of between 6 and 13% in the mean monthly basin runoff (Fig. 10a) based on the baseline time periods 1980-2010 and 1988-2000, respectively, which is identical to the predicted values by DP (10% P reduction and 0% PET) as indicated by Fig. 5b. The corresponding decrease for the time horizon 2046-2065 will be between 21 and 25%, which is identical to the predicted decrease in streamflow by DP (20% reduction in P and 0% PET increase) as highlighted in Fig. 5d. However, the anticipated runoff decrease for the last time horizon is between 31 and 36% (Fig. 10c), which is identical to the value obtained via DP as highlighted by Fig. 10f, (30% P reduction and 0% PET increase). It is expected that the reservoir inflow peak point will decrease, and there is likely to be a noticeable change in the flow amount, which may lead to a considerable impact on water resources management of this example basin.

Conclusions and recommendations
In order to enhance adaptation strategies to climate change, this study proposed a simple and generic methodology. The following conclusions can be drawn: & The downscaling model LARS-WG5.5 predicted meteorological data in arid and semi-arid areas well and should be applied for prediction of daily future values. & There will be a likely increase of about 3.02 and 3.33°C and 3.17 and 3.70°C in T min and T max for the 2046-2065 time period and the two considered baselines, respectively. However, there may be a decreasing trend of nearly 17.33 and 21.93 mm in the corresponding P values and the relative variations will fluctuate monthly. & Both 1988-2000 and 1980-2010 baseline time periods anticipated almost similar reductions in the average monthly inflow to the reservoir, and subsequently, their peak points are nearly the same. For example, the predicted reduction in the reservoir inflow for the 2080-2099 period was between 31 and 36%, which is equivalent to the assessment from the DP scenario (30% P decrease and 0% PET rise). & The SRA2 and SRA1B scenarios forecast approximately the same reductions in the mean monthly reservoir inflows; accordingly, their maximum values are almost the same, in particular for the 2011-2030 and 2046-2065 time periods. The two emission scenarios show decreases in peak values ranging between 6% (SRA2) and 10% (SRA1B), 21% (SRA2) and 25% (SRA1B) and 31% (SRA1B) and 49% (SRA2) for the three future time periods, respectively. & There will be a reduction in the average monthly reservoir inflow as a result of P reduction.
By considering as many climate change and emission scenarios as well as baseline time period as practically possible, the water resource policy makers can adapt to the expected impacts of climate change efficiently. This study should be undertaken again for other areas and climate types using different baseline time periods, which should lead to a further generalisation of the study conclusions.