Evaluating the effects of climate change on precipitation and temperature for Iran using RCP scenarios

Climate change (CC), driven by increases in greenhouse gases concentrations in the atmosphere, as a consequence of anthropogenic activities, has led to an increase in the global temperature of approximately 1 °C since the pre-industrial period (Dibike & Coulibaly 2005; Feng et al. 2014; Bekele et al. 2019). Moreover, CC can alter local climatic conditions and consequently accelerate hydrological processes (Kim et al. 2011; Thomas et al. 2018; Bekele et al. 2019). The integration of likely changes in hydrology is critically important for flood risk reduction and water resource management (Broderick et al. 2019; Toosi et al. 2019). Effective adaptation measures in these sectors require CC to be factored into an investment in long-lived infrastructure, upon which a society depends (Milly et al. 2008; Li et al. 2016; Taye et al. 2018; Bekele et al. 2019; Broderick et al. 2019). Several criteria determine the rainfall–runoff response, including the characteristics of the basin, vegetation density, soil type, basin shape and size, and antecedent moisture conditions (which vary seasonally and on an event basis), controlling the amount of rainfall that can permeate into the soil (Alaghmand et al. 2014; Jajarmizadeh et al. 2016; Toosi et al. 2019). Therefore, it is essential to understand how hydrological processes are expected to change, together with the degree of uncertainty in the hydrological response at the regional scale (Dessu & Melesse 2013; Bekele et al. 2019).

General circulation models (GCMs) are the most widely used tools for investigating CC at global/regional scales through the production of climate scenarios for present and future time horizons. They have been found to be valuable tools for identifying hydrologic consequences of changes in the climate variables (Ullah et al. 2019; Thomas et al. 2018; Phillips et al. 2019; Warnatzsch & Reay 2019; Yang et al. 2019). The evaluation of global trends in GCM outputs has suggested that drier areas will likely become drier and wetter areas will likely become wetter, with a notable expansion in arid and semi-arid climates (Feng et al. 2014; WWAP/UN-Water 2018). Nevertheless, a comprehensive investigation of regional and seasonal impacts of CC in the future is needed, particularly at regional scales (Aloysius et al. 2016).

The results of previous research have revealed that simulated CC impacts in different regions are diverse and are highly dependent on the climate models and emission scenarios employed (Dibike & Coulibaly 2005; Feng et al. 2014; Fiseha et al. 2014; Cousino et al. 2015; Wu et al. 2015; Bekele et al. 2019). A case in point is the region of the Middle East where CC impacts vary significantly with each season and the ‘wet gets wetter, dry gets drier’ paradigm is not necessarily followed (Taye et al. 2018). Zarghami et al. (2011) assessed the impact of CC on temperature and precipitation for six synoptic stations in a case study using the HADCM3 model and three emission scenarios, A1B, A2, and B1 for the years 2020, 2055, and 2090. Their results showed that under the A2 scenario, a 2.3 °C rise in average annual temperature and a 3% reduction in annual precipitation are expected in the middle of this century. Zhao et al. (2014) pointed out that arid and semi-arid areas globally will likely experience a significant increase in surface air temperature (SAT) under various Representative Concentration Pathways (RCP), that wet regions will likely become wetter, and dry regions will likely become drier. However, the extent to which future CC will affect regional shifts is still uncertain and differs from region to region. Taye et al. (2018) evaluated the possible impact of CC on water availability for a basin in Ethiopia using three climate models from Coupled Models Inter-comparison Project phase 5 (CMIP5) for three future periods. Their results revealed that water shortages are expected to become more severe for some parts of the basin. Mirdashtvan et al. (2018) characterized the changes in climatic variables for the period 2011–2040 under different RCP scenarios and quantified the uncertainty in future projections linked with the downscaling methods for one of the most vulnerable basins in Iran; the ‘Karaj-Jajrud’ located in the South Alborz range. In comparison with the baseline period, all scenarios showed a consistent growth in SAT and a reduction in precipitation, while precipitation-series uncertainty was found to be more than the air temperature series.

Although previous studies provide essential information about potential CC impacts (Zarghami et al. 2011; Sabeerali et al. 2013; Mirdashtvan et al. 2018; Hosseini et al. 2020), our understanding of CC impact on the hydrological characteristics in Iran is lacking. In recent years, heavy rainfall events, flash floods, droughts, and extreme temperatures have been widely observed in Iran (Madani 2014; Rahimi et al. 2019; Toosi et al. 2019; Vaghefi et al. 2019). Given the variability of hydro-climatological conditions across the country, adopting a national (or regional) allowance for the realization of future climate may fail to address the actual emergent risk locally (Scussolini et al. 2016; Broderick et al. 2019). There is, therefore, a need to assess likely CC impacts for specific stations to help develop adaptation measures (Ullah et al. 2019). Furthermore, in Iran, most recent studies have used a limited number of GCMs, meaning that uncertainty in the results of these models has been neglected. Although the choice of suitable climate models, that can capture basic facets of Iranian climatology, is of considerable importance, few studies have aimed to assess climate models in this regard. The current study aims to evaluate the impacts of future CC on seasonal SAT and precipitation for distinct climatological regions in Iran using six long-term, high-quality synoptic weather stations. The specific objectives are (1) to assess the ability of 25 GCMs from the CMIP5 archive to capture the observed seasonal variability in each variable for different regions of Iran and (2) using selected GCMs, to investigate the impacts of CC on future precipitation and SAT under various RCP scenarios.

The flowchart of the methodology adopted is shown in Figure 1. Precipitation and temperature simulations from 25 GCMs from CMIP5 are compared with observed data from six stations for each season. Models that best capture the seasonal cycle are selected for further use. Using the change factor method, the outputs of GCMs were downscaled for the future period (2046–2065). Finally, the impacts of CC are discussed on a station by station basis.

Located in Western Asia, Iran has an area of 1,640,195 km², a population of 81 million inhabitants, and is the second major country in the Middle East. It lies between 24°N and 40°N latitude, and 44°E and 64°E longitude. Iran has diverse climates: mild and wet on the coast of the Caspian Sea, continental and arid in the plateau, cold in the high mountains, and dry and hot in the deserts of the southern coast and southeast. These distinct climate zones make it challenging to assess CC impacts in Iran and provide a stern test for climate models (Nicholson 2011; Mansouri et al. 2014). In this research, six synoptic stations were selected to be representative of specific climatic regimes (Figure 2). A description of the selected stations is presented in Table 1. Total monthly precipitation and mean monthly SAT data for these stations for the 20 years (1981–2000) were collected from the Iran Metrological Organization (IRIMO) (IRIMO 2016).

There are many factors to consider when selecting a model (Gleick 1986; Bekele et al. 2019; Gorguner et al. 2019). The purpose of the study and data availability are the dominant factors responsible for the choice of a particular model (Thomas et al. 2018). The most important base for obtaining CMIP5 data is ESGF, which is also the official website of the Intergovernmental Panel on Climate Change (IPCC). It provides users with the raw output from models for the historical period and future scenarios for a variety of time scales and a large number of climatic parameters such as temperature and precipitation. The main problem in using the output of GCMs is the low resolution of output grids which is too coarse to be useful for regional studies. Consequently, their outputs should be downscaled before being used in local studies. Given the importance of these models, some research centers have downscaled the output of GCMs to smaller grids and made the results accessible for all researchers worldwide. In the present study, the output from CMIP5 GCMs with 0.5° × 0.5° grid size (downscaled using a bias-corrected statistical downscaling (BCSD) method) was used. Data were downloaded from http://gdo-dcp.ucllnl.org. Among available downscaled GCMs, 25 models which had projected climate under three RCP scenarios (RCP2.6, RCP4.5, and RCP8.5) were selected. These scenarios describe different climate futures according to the quantity of greenhouse gases emitted in the coming years (IPCC 2014).

RCPs explore credible future options by considering the uncertainties associated with future developments. RCP 2.6 is the most optimistic mitigation scenario, in which global annual greenhouse gas emissions peak to 440 ppm between 2010 and 2020 and then reduce considerably (van Vuuren et al. 2011). It implicates a global turnaround in environmental policies and collaborative actions from all emitters in the next few years for the active clearance of carbon dioxide from the atmosphere (van Vuuren et al. 2007). Under RCP 4.5, which is a stabilization scenario, emissions peak around 2040 then decrease and total radiative forcing reaching 540 ppm by 2100 before leveling off (Clarke et al. 2007; Thomson et al. 2011). RCP 8.5 is a pessimistic scenario in which emissions rise steadily over the 21st century, reaching 940 ppm by 2100, and continue increasing for another 100 years (Riahi et al. 2007) (Figure 3).

We used monthly values of total precipitation and average SAT for the historical (1981–2000) and future (2046–2065) periods. Table 2 shows the summary characteristics of the GCMs used in the current study.

The Taylor diagram has become a useful tool in the evaluation of the overall performance of climate models (Kim et al. 2014; Loikith et al. 2015; Warnatzsch & Reay 2019). The annual performance of models was assessed statistically using the Taylor diagram (Taylor 2001). It provides a brief statistical analysis of the degree of pattern correspondence between the modeled and observed data in terms of their Pearson's correlation, root-mean-square error (RMSE), and the ratio of their variances which are simultaneously indicated by a single point on the plot (Taylor 2001). The diagram is particularly useful in assessing the relative merits of competing models and monitoring the overall performance of a model as it evolves.

Although the annual analysis illustrates a good overall picture of models' performance, the results can vary significantly in other temporal scales (e.g. seasonal). Therefore, the seasonal performance of the precipitation and SAT for all selected GCMs was examined using the RMSE and skill score (SS), and the best models were selected based on their seasonal performance. RMSE is always non-negative, and a value of 0 would indicate a perfect fit to the data. The SS index is used to evaluate the goodness of fit of a model prediction. It ranges from −∞ to 1 with SS = 1 reflecting the perfect match of simulated and observed data, and SS = 0 shows that the model predictions are as accurate as of the mean of the observed data.

Several papers have previously indicated that the direct application of GCMs is limited due to their coarse resolution and systematic bias (Park et al. 2016; Yang et al. 2019). The coarse resolution prevents the models from sufficiently representing regional climatic processes. The biases potentially grow when used for CC simulations under global warming conditions. Bias correction should be applied to each model (Christensen et al. 2008; Yang et al. 2019). Downscaling is one of the approaches where GCM outputs are interpolated to meet local scale requirements and reduce bias (Mujumdar & Nagesh Kumar 2012; Raju & Kumar 2018). The change factor method is the most straightforward technique and is suitable for downscaling the mean value of climatic variables. In this study, to capture higher-resolution features and preserve the monthly variability of stationary observed data, the bias-correction change factor technique was applied to 0.5° × 0.5° gridded GCMs to map the projections to weather stations' scale (Hansen et al. 2017; Ciscar et al. 2019).

We used the Taylor diagram to evaluate the overall performance of GCMs for annual precipitation. Figure 4 illustrates the performance of models for each station. The radial distance from the REF point, which is the indicator of observed values, is the centered pattern root-mean-square deviation (CRMSD), and the radial distance from the origin is the standard deviation, while the angle shows the correlation coefficient. Seasonal Taylor diagrams are presented in Figure S1 in the Supplementary section.

As illustrated in Figure 4, some models perform better than others; however, their overall performance is similar. The performance of the CMIP5 models for the Babolsar station is weaker, but all other stations illustrate the correlation value above 0.7. At the seasonal time scale, the performance of the simulated data for the historical period varied significantly (Figure S1), which highlights the importance of using seasonal data in choosing an appropriate model (Sabeerali et al. 2013; Li & Mao 2016). The acceptable performance of GCM models at the annual time scale cannot guarantee a good performance at monthly or seasonal time scales.

In Figure 5, boxplots depict the average range of seasonal results obtained from GCMs and red points show the average observation values. As can be seen, almost all models struggle to simulate the seasonal mean precipitation. This is particularly the case for the Babolsar station, where the model's results have a significant deviation from the observed values. For some stations, models tend to overestimate the seasonal precipitation, for example, Tabriz and Yazd in all seasons. However, model biases in different seasons vary considerably; for example in the Yazd station where biases in JJA and SON are relatively small, they are higher for MAM and DJF. Conversely, at the Babolsar station, all models underestimate seasonal precipitation. In particular, in DJF and SON, the model biases are considerably greater than in other seasons. The results for other stations were reasonably acceptable for all seasons, excluding for DJF at Torbat Heidarieh, JJA at Iranshahr, and SON at the Abadan station, where all models underestimated seasonal precipitation totals.

The ranking of models based on SS is presented in Table 3. In this table, the relative values of RMSE are shown as the color spectrum in which darker colors represent better performance (smaller RMSE). It can be seen that RMSE and SS rankings of individual models are quite similar. Tables S1–S6 in the Supplementary section contain the value of the SS score for different GCMs as well as additional indices, which were calculated to support our findings.

For the Tabriz station, four models, including CSIRO-Mk3-6-0 for DJF, IPSL-CM5A-MR for MAM, FIO-ESM for JJA, and MIROC-ESM for SON, showed better performance in terms of simulating the seasonal mean precipitation, and these models were used to simulate future changes under the RCP2.6, RCP4.5, and RCP8.5 scenarios. Likewise, for the Abadan station, three models including CESM1-CAM5, NorESM1-M, and CanESM2 were selected. At the Babolsar station, four models were selected, including EC-EARTH, FIO-ESM, CanESM2, and IPSL-CM5A-MR, while for the Torbat Heidarieh station, four selections were made, including GFDL-ESM2G, IPSL-CM5A-MR, BNU-ESM, and HadGEM2-AO. At the Iranshahr station, the selected models include HadGEM2-ES, MPI-ESM-LR, CESM1-CAM5, and CSIRO-Mk3-6-0 and lastly for the Yazd station, three models were selected, including HadGEM2-AO, EC-EARTH, and BNU-ESM.

The change factor downscaling technique was applied to six synoptic stations. The precipitation change factors were estimated for those grid cells containing the stations to produce future projections. The obtained change factor value for each selected model over the baseline period was applied for relevant months for the future period. The projected precipitation under all scenarios for the years 2046–2065 was calculated and is depicted in Figure 6. The results for different stations show different behavior under various scenarios. Overall, the average annual precipitation is simulated to increase under all scenarios compared to the baseline period for the Abadan and Yazd stations. Similarly, at the Tabriz station under RCP2.6 and RCP8.5, annual precipitation is expected to increase. In general, simulations show that precipitation is likely to increase for the western stations, while projected decreases are apparent for the eastern stations. It is also expected that the greatest increase in precipitation under all RCPs is likely for the Yazd station, while the highest reduction occurs for the Iranshahr station, which has a dry and warm climate. Moreover, for the Iranshahr station, when the scenarios become more pessimistic, declines in precipitation are even greater.

The projection of seasonal precipitation changes under all emission scenarios is given in Table 4. The results show the greatest decrease (93% in JJA) under RCP4.5 at the Iranshahr station and the greatest increase (222% in JJA) under RCP8.5 at the Yazd station. The results of various scenarios show that there is not any specific pattern in precipitation change over different seasons but, for some stations, under all scenarios, the largest increase/decrease occurs in one season (e.g. for the Abadan and Iranshahr stations, the greatest decrease occurs in JJA). It is expected that in all stations, precipitation changes exhibit different behavior at different seasons when scenarios become more pessimistic.

Similar to precipitation, we used the Taylor diagram to assess the overall performance of GCMs in the simulation of the annual SAT. Figure 7 shows the annual performance of models for all six stations. It can be seen that for all models, the results are almost the same, and their overall performance is very close. The correlation coefficient for all GCMs is above 0.95 in all stations. Corresponding Taylor diagrams of the seasonal cycle are summarized in Figure S2 of the Supplementary section.

From Figure 8, the performance of different models in capturing the seasonal temperature varied significantly across each station. While most GCMs give a reasonable prediction in terms of the mean SAT, at some stations, the models overestimated the seasonal SAT, e.g. in all seasons at the Babolsar station except DJF; in JJA and SON at the Abadan station, and all of the seasons except JJA at the Torbat Heidarieh station. Conversely, at the Iranshahr, Tabriz, and Yazd stations, all models underestimated the seasonal SAT. Overall, the model biases in the simulation of the SAT are relatively small for all stations.

Similar to precipitation, we examined the ability of the GCMs to simulate the seasonal SAT. The same statistical indices, i.e. RMSE and SS, were used together to select GCMs with the best performance in the baseline period. The selected models were utilized to project future SAT changes at each station (Table 5). It is noteworthy that similar to precipitation, additional indices were also calculated for the SAT but were not used in model selection and are presented in the Supplementary section (Tables S7–S12).

For the Tabriz station, three models, i.e. CESM1-CAM5 for DJF and JJA, IPSL-CM5A-MR for MAM, and CanESM2 for SON were selected to examine future SAT changes under the RCP2.6, RCP4.5, and RCP8.5 scenarios. Likewise, for the Abadan station, four models including GISS-E2-R, IPSL-CM5A-MR, EC-EARTH, and GFDL-CM3 were selected. For the Babolsar station, three models including MPI-ESM-MR, GISS-E2-R, and CSIRO-Mk3-6-0 were identified, while for the Torbat Heidarieh station, three models including CNRM-CM5, GFDL-ESM2G, and CCSM4 were selected. For the Iranshahr station, four models including FGOALS-g2, CESM1-CAM5, CanESM2, and BNU-ESM were chosen, and lastly for the Yazd station, three models including BNU-ESM, CESM1-CAM5, and bcc-csm1-1 were selected.

Again, the change factor method was used as the downscaling method over six grid points in Iran. The simulation of SAT under all RCP scenarios for the period 2046–2065 was made, with results presented in Figure 9. The results indicated a positive trend in SAT in all months over all stations for the time period considered. The smallest and largest increases in SAT were derived under the RCP2.6 and RCP8.5 scenarios, respectively. For all six stations, there is an increase in the projected SAT, as the severity of the greenhouse scenario increases. The greatest increase in SAT under all RCPs is projected for the Abadan station, which has a warm and desert climate. For all stations, increases in SAT get larger as scenarios become more pessimistic.

Table 6 contains the seasonal projections of SAT changes over different stations under all RCP scenarios based on selected models. The results show the lowest increase (0.26 °C in JJA) under the RCP2.6 for the Torbat Heidarieh station and the highest increase (5.69 °C in SON) under RCP8.5 for the Abadan station.

CMIP3/CMIP5 simulations have long been used in various studies to assess the impacts of CC on humans and the environment. Many GCMs have been developed to represent future climate conditions at the global scale under different scenarios (Zhang et al. 2013; Mallakpour & Villarini 2015; Najafi & Moazami 2016; Ahmadalipour et al. 2017). Uncertainty is an indispensable part of GCM predictions which can be derived from their natural variability and coarse resolutions (Hawkins & Sutton 2011; Ahmadalipour et al. 2017). For that reason, selecting models that appropriately represents the regional-scale climate is an essential step before performing a regional CC impact assessment (Ahmadalipour et al. 2017). The better performance of the mean feature over the baseline period indicates the better simulations of the model (Sabeerali et al. 2013). Monthly datasets from CMIP5 can help researchers derive a more robust analysis and more reliable model comparison.

Evaluation of the results of the present study illustrates that models generally perform better in simulating SAT in comparison with precipitation. The majority of the models was unable to simulate the temporal pattern of precipitation at the seasonal scale in all stations. This is primarily due to the fact that the average rate of monthly variations in precipitation is higher in comparison to SAT (e.g. assume two following days, one with heavy rainfall and the other one with no rainfall (Ahmadalipour et al. 2017). Also, it can be attributed to the more stable nature of SAT which makes it easier to predict. Other research also confirms this issue (Samadi et al. 2013; Etemadi et al. 2014; Mirdashtvan et al. 2018).

Despite the fair simulation of GCMs in terms of the mean seasonal precipitation at most stations, all the models have difficulties in simulating precipitation at the Babolsar station, particularly in DJF and SON, where the models failed to simulate a reasonable correlation pattern with observations and underestimate totals during all seasons. Out of the 25 models analyzed, none gives a correlation of greater than 0.5 in simulation annual precipitation amounts at this station. Since GCMs produce results on the global scale (coarser resolution, Table 1), they tend to over/underestimate climatic variables on regional and global scales, failing to resolve the microscale climate (Ahmadalipour et al. 2017). However, the outcome of our study revealed that models with finer resolution do not always perform better than those with coarser resolutions (e.g. CCSM4, CNRM-CM5, MIROC5, and MRI-CGCM3). Regarding other stations, some models provide realistic figures for some seasons in the baseline period. However, they failed to provide reasonable outputs for all seasons over the years.

At the Tabriz station, seasonal results show that the total mean precipitation under RCP8.5, which is a pessimistic scenario, increases in the future over wet seasons and decreases over dry seasons compared to the historical observations. These results agree with Zarghami et al. (2011) in the Northwest region of Iran, which were derived from the HADCM3 GCM model under the A2 scenario. Future simulations of SAT for most of the world show that there will be consistent warming in SAT with different magnitudes under all RCP scenarios (Mirdashtvan et al. 2018). SAT changes may have an impact on the hydrology of a region by increasing evapotranspiration rates even when no significant changes occur in precipitation amounts (Mirdashtvan et al. 2018). The results of this study on consistent warming in future periods are verified by the IPCC assessment reports, which state that the average SAT is rising globally (Samadi et al. 2013; Etemadi et al. 2014; IPCC 2014; Mirdashtvan et al. 2018).

It is noteworthy to mention that the selection of models in the current study is dependent on which metrics/skill scores you assess against. It means that if you were to look at how GCMs capture extremes or modes of variability that affect a region, you would most likely get a different set of climate models. Also, the models were chosen according to their seasonal performance, and there is no guarantee that this will remain the best model for other time scales. Therefore, there is a risk that uncertainties in future projections under-representing and should be considered in future studies. The question that arises here is that if the selected models for different seasons affect the physical integrity of the simulations. These models might be right for the wrong reasons or vice versa. This makes the future simulations lack internal consistency, which is suggested to be analyzed more in future works.

In this study, utilizing monthly data for 20 years (1981–2000), the outputs of 25 GCMs for SAT and precipitation were evaluated using observations for six synoptic stations over Iran. The performance of these models was evaluated using RMSE and SS, and the best models were selected at seasonal time scale and accordingly, the future data were generated. Although the annual performance of models was different, most models show an acceptable representation of the annual cycle of each variable at most stations. The GCM predictions at different time scales showed dissimilar uncertainties. Evaluation of the results of the present study illustrates that models generally perform better in simulating SAT compared to precipitation. The majority of the models was unable to simulate the temporal pattern of precipitation at seasonal scales at all stations. Therefore, there is less confidence in precipitation projections in comparison with SAT projections due to the unpredictable nature of the former. The results showed that the GCMs tend to over/underestimate climatic variables on regional and global scales, failing to resolve the microscale climate. Also, the results revealed that models with finer resolution do not always perform better than those with coarser resolutions. In most stations, some models provide realistic figures for some seasons in the baseline period. However, they failed to provide reasonable outputs for all seasons over the years. Overall, the results showed that the mean SAT is expected to increase in all seasons, while the precipitation change did not follow a specific trend. The outcomes of this study and related research can be a stimulus for the government to find new and sustainable adaptation strategies for the water sector.

We acknowledge the World Climate Research Programme's Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Table 2 of this paper) for producing and making available their model output. For CMIP, the U.S. Department of Energy's Program for Climate Model Diagnosis and Inter-comparison provides coordinating support and led the development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. We would also like to thank the Downscaled CMIP3 and CMIP5 Climate and Hydrology Projections archives at http://gdo-dcp.ucllnl.org/downscaled_cmip_projections/ for providing the BCSD CMIP5 downscaled data, as well as the re-gridded to 0.5° × 0.5° resolution GCMs data.

The Supplementary Material for this paper is available online at https://dx.doi.org/10.2166/wcc.2020.114.