Projected changes in extreme precipitation indices over the Lake Urmia basin in Iran

Climate change is already affecting global ecosystems, biodiversity, social economics, and human health (IPCC 2013; Ashktorab & Zibaei 2022; Esa et al. 2022). Therefore, it is important for scientists to assess the effects of climate change in each region and provide proper solutions to adapt and reduce the consequences of climate change (Fuhrer 2003). The Intergovernmental Panel on Climate Change (IPCC) in the Fifth Assessment Report (AR5) reported that climate warming is quite clear, and the continued greenhouse gas emissions cause more warming and change in all parts of the climate system (IPCC 2013). One of the important consequences of climate change will be the increase in extreme climate phenomena such as droughts, floods, hail, heat waves, rising sea levels, and cold waves (Hu et al. 2012). The previous study indicates that the intensity and the frequency of extreme precipitation events in the mid-latitudes may further increase in the future (IPCC 2013). Any change in the intensity or frequency of climatic extremes can have great effects on the natural environment and human societies. Therefore, analyzing extreme events is very important.

In this regard, a set of climate extreme can be used for the analysis of extreme precipitation amounts (e.g., Sillmann et al. 2013; Xu et al. 2018; Lee et al. 2020; Grover et al. 2022; Hong et al. 2022). Some researchers have conducted various studies on changes in climate extremes for both observational and future periods. On the other hand, many studies have focused mainly on assessing changes and trends in mean observational climate values in the last decades. On the basis of studies on observed data, Rahimzadeh et al. (2009), Soltani et al. (2016), Azizzadeh & Javan (2018), and Fathian et al. (2020) showed that the extreme precipitation indices in some regions of Iran, such as dry and wet days, have increased and decreased, respectively.

Climate change forecasts rely on general circulation models (GCMs). The GCMs, as numerical models representing the physical processes of the atmosphere, can simulate the current climate and visualize future climate situations based on various scenarios (Sillmann et al. 2013; Lee et al. 2020; Abbasi et al. 2022). As a collection of GCMs, CMIP5 (Coupled Model Intercomparison Project Phase 5) models have higher horizontal resolutions than previous CMIPs and also have much more comprehensive behavior in relation to physical processes such as vegetation feedbacks, aerosols, and land cover types (Aloysius et al. 2016). Unlike the Special Report on Emissions Scenarios (SRES), in the representative concentration pathway (RCP) scenarios, the forcing radiation path is not predefined, and the RCPs are able to represent different combinations of economic status, demographics technology, and political development (Moss et al. 2010). It is, therefore, expected that the errors of the CMIP5 models will be reduced compared to the previous models and also provide more realistic simulations of the future (Taylor et al. 2012).

Because the outputs of GCMs are large-scale gridded, there is a discrepancy between the scale of these models and the scale needed to study the effects of climate change. This has led to the development of several downscaling methods (Graham et al. 2007). LARS-WG (Long Ashton Research Station Weather Generator), which is employed in this study, is one of the statistical models that many evaluations have been made with this model and acceptable results have been obtained (Reddy et al. 2014; Kavwenje et al. 2022).

In Iran, Zamani Noori et al. (2014) investigated the uncertainty of precipitation and temperature parameters simulated by two downscaling models, namely, LARS-WG and statistical downscaling model (SDSM). The results showed that the LARS-WG performs better than the SDSM in precipitation simulation and shows a decrease in precipitation in the near future. Goudarzi et al. (2015) in simulating climate change in the Lake Urmia basin with LARS-WG and SDSM showed that the SDSM model is more successful in simulating temperature parameters and has less uncertainty, while the LARS-WG model is better in simulating rainfall periods. Lotfi et al. (2022) analyzed the performance of LARS-WG and SDSM in simulating temperature and precipitation changes in the West of Iran. They showed that LARS-WG is more efficient in simulating annual precipitation and is simpler with a higher performance speed. Zarrin et al. (2022) evaluated future changes in some precipitation extremes over Iran for the historical and future periods based on five CMIP6 bias-corrected models.

Lake Urmia, the largest inland lake in Iran, plays a great role in moderating the climate of the region. In the basin of this lake, agricultural and industrial activities, as well as infrastructure projects and water resources development, have expanded in recent decades (Taheri et al. 2019). These actions have had important effects on the ecological conditions and level of Lake Urmia. Climate change, on the other hand, plays an important role in Urmia Lake's water level and water resources (Abbaspour et al. 2012). Research has shown that water resources have declined in watersheds such as Lake Urmia (Delju et al. 2013). Therefore, it is necessary to examine climate change, especially in the Lake Urmia basin, which has become an environmental crisis (AghaKouchak et al. 2015). Attention to the new scenarios and models is also necessary to better understand the behavior of climate change in the basin in the future.

The purpose of this study is to investigate the change of extreme precipitations in the Lake Urmia basin using the output of CMIP5 models and RCP in the base period (1987–2016) and the future periods (2021–2100). This study attempts to:

• select the most suitable model for the basin from the CMIP5 models and to construct an ensemble model to reduce uncertainty,

• downscale the precipitation during the reference period, and

• provide a prospect of changes in extreme indices over the studied periods.

Considering the critical situation of the water resources of the Lake Urmia basin, we believe that this study is very necessary for the sustainable exploitation of the basin's water resources under climate change.

One of the main problems is the decrease in the water level of Lake Urmia in recent years, which seems to have affected the climate change in addition to the construction of numerous dams and the expansion of gardens and agriculture (Khazaei et al. 2019). At present, the lake's water level has decreased more than 6 m from the peak and nearly 3 m from the ecological level (Tourian et al. 2015). So far, research studies have been conducted on the future status of rainfall and runoff in the Lake Urmia basin using different models (Goudarzi et al. 2015; Davarpanah et al. 2021). But there has been no research on the impacts of climate change on extreme precipitation indices in this basin, as far as the authors know.

The data used in this research include daily precipitation data, minimum temperature, maximum temperature, and number of sunshine hours at eight synoptic stations in the basin that have long-term and reliable data. These data are for the period 1987–2016. Missing data for each station were replaced with daily averages from other years obtained from the same station (Liu et al. 2011). The distribution of the synoptic stations is shown in Figure 1, and the characteristics of these stations are provided in Table 1. In the first step, data quality control and homogeneity were investigated using the RClimDex package.

Future period data are based on the output of CMIP5 models. Despite the development of climate models and the advancement of science, there are some biases and uncertainty in models (IPCC 2013). Various methods have been proposed by IPCC to reduce these uncertainties. One of these methods is to identify suitable GCMs that reduce uncertainty in future simulations. Table 2 presents the details of 16 models from the CMIP5, which were used for this study.

Considering the complexity of the climate system, it is hard to adequately describe climate change with a single model. To reduce uncertainty, the average or a combination of results from several models has been used for climate projections. Model averaging is a statistical method in which unequal or equal weights are assigned to those models (Hong et al. 2021). In this study, equally weighted model averaging was used to ensemble the selected models.

Because the output of GCMs does not have the necessary capability at the local scale, downscaling methods should be used to compensate for this shortcoming. In this regard, the LARS-WG downscaling method was used based on the scenarios of RCP4.5 and RCP8.5. The RCP4.5 represents the intermediate scenario, while the RCP8.5 is the most pessimistic scenario (Moss et al. 2010). In RCP4.5 (RCP8.5), the CO₂ concentration by the year 2100 is 650 ppm (1,370 ppm), and the effect of greenhouse gases on radiative forcing is 4.5 W/m² (8.5 W/m²) (Van Vuuren et al. 2011).

Eight extreme precipitation indices introduced by ETCCDMI (Expert Team on Climate Change Detection, Monitoring and Indices) were used in this study. The list of these indices is presented in Table 3. All of these indices are estimated by RClimDex (Zhang & Yang 2004). These indicators can be grouped into four classes (Alexander et al. 2006):

• 1.

  Percentage indices: Precipitation over the 95th percentile (R95p) and 99th percentile (R99p), and 1-year extreme precipitation events.

• 2.

  Absolute indices: Maximum 1-day precipitation (RX1day) and maximum 5-day precipitation (RX5day).

• 3.

  Periodic indices: Consecutive wet days (CWDs), and consecutive dry days (CDDs).

• 4.

  Other indices: Simple daily intensity index (SDII), and the annual wet day precipitation (PRCPTOT).

The statistic Z follows the standard normal distribution. If the Z statistic is positive, the trend of the data series will be upward, and if it is negative, the trend will be decreasing.

The performance evaluation and comparison of the proposed models are reported in Table 4. The results showed that HadGEM2-ES, GFDL-CM3, NorESM1-M, and MIROC5 models have a high correlation with the observed precipitation in the basin. The results showed that based on the error statistics (RMSE, MAE, MBE, and MPE) and R², HadGEM2-ES, GFDL-CM3, NorESM1-M, and MIROC5 models are the most suitable models for the Lake Urmia basin. In terms of the MPE, these models have the lowest error percentage (between 9 and 12%). Positive values of MBE indicate an overestimation of GCMs, and its negative values indicate an underestimation of these models. Therefore, these four models were used to create the ensemble model and to investigate precipitation changes.

The adequacy of the LARS-WG model to predict the precipitation was tested using the p-value of the Kolmogorov–Smirnov (K-S) test. The K-S test is performed to check the equality of the seasonal distributions of wet and dry series and distributions of daily rainfall calculated from observed data and downscaled data (Wilks 2011). This test calculates a p-value, which is used to accept or reject the hypotheses that two datasets could have come from the same distribution (i.e., when there is no difference between the observed and predicted data). A very low p-value means the predicted data are unlikely to be the same as the observed data and hence should be rejected. The results of the K-S (Kolmogorov–Smirnov) test show that there was no significant difference between the distributions of the daily observed and modeled series for all weather stations. The acceptable p-value from the LARS-WG manual is more than 0.01 (Semenov & Barrow 2002). In most months, the p-values are above this range, indicating that the data follow a normal distribution (Table 5). The LARS-WG model was thus judged to be acceptable to generate synthetic weather series based on projected climate scenarios.

In general, it can be said that in the base period, in January, February, March, April, June, August, and December, there is a decreasing trend (Table 6). Moreover, in May, July, September, October, and November, there is an increasing trend in precipitation. In the coming periods and in January, April, May, June, July, August, November, and December, a decreasing trend is observed. In February, March, September, and October, an increasing trend is observed in the basin. However, the trend slope based on RCP8.5 is higher than RCP4.5.

The results of previous studies also show a decrease in precipitation in future periods in Iran and the Lake Urmia basin (Sayadi et al. 2019; Motiee et al. 2020; Davarpanah et al. 2021). This result is consistent with the results presented in the study by Abbasian et al. (2021), who showed an 8% decrease in precipitation in 2060–2080 based on RCP8.5. Naderi & Saatsaz (2020) also reported a decrease of 10% in annual precipitation.

Because Rx1day and Rx5day indices indicate the continuity of cyclonic systems or instability in 1-day and multiday periods, they are suitable indicators for the flood study. Considering that the amount of these indices will decrease in the future in the Lake Urmia basin, it can be concluded that the potential for floods and the resulting damage in the region will be reduced. Fathian et al. (2022) obtained similar results on the behaviors of changes in RX1day and RX5day.

Table 7 shows the results of the trend test and the slope estimates of the predicted indices based on the RCP4.5 and RCP8.5. Rx1day has an upward trend, whereas Rx5day has a downward trend based on both scenarios, but the trends of the indicators are not significant. Zarrin & Dadashi-Roudbari (2021) reported an increasing trend in Rx1day, which is consistent with the results of this study.

The overall rate of R95p based on RCP4.5 and RCP8.5 scenarios for the Lake Urmia basin is −2.7 and −0.4 mm per decade, respectively. The overall rate of R99p based on RCP4.5 and RCP8.5 is 0.2 and 2.7 mm per decade, respectively (Table 7). Darand (2020) and Soltani et al. (2020) reported an increase in the heavy rainfall over Iran in the future. Fathian et al. (2022) showed that the number of stations with a positive trend in R95p will increase in the future in Iran. However, most stations situated in the north and northwest regions show negative trends that are in agreement with our findings.

The results of downscaling based on RCP4.5 and RCP8.5 in the Lake Urmia Basin show insignificant changes in SDII in all three periods (Figure 8(c)). Examining the trend of this index also shows a downward trend in both scenarios, but the slope of this trend is small (Table 7). Darand (2020) showed that annual total precipitation decreased in the future period, and the projected decreases were larger in the northwestern parts of Iran.

There are a few limitations to this study. First, the number of synoptic stations in Iran, especially in the Lake Urmia basin, is very low and scattered, and most of the stations are newly established and their period is short. As a result, the zoning of climatic variables and extreme indicators in the basin is difficult. Furthermore, the ETCCDI indices are defined generally at annual timescales, and some are defined at monthly timescales as well. For some sectoral applications (e.g., in agriculture water resource management and energy), the current set of monthly/annual indices may prove less helpful, as climate anomalies need to be computed over different timescales. Thus, the results of this study are subject to some uncertainties in the prediction of precipitation and especially in extreme values (Najafi & Hesami Kermani 2017; Hong et al. 2022). We may need to employ other statistical methods such as based on return level estimation (EskandariPour & Soltaninia 2022; Rivoire et al. 2022).

This study was conducted to provide a perspective on future changes in precipitation extreme indices in the Lake Urmia Basin in the period 2021–2100. For this purpose, the output of 16 GCMs was examined. HadGEM2, GFDL, NorESM1, and MIROC5 models were selected as suitable models for the basin. The output of these four models was downscaled by LARS-WG. Then, an ensemble model was created by model averaging. Finally, precipitation extreme indices were calculated for the studied stations, and then the regional average for the whole basin was obtained. By comparing the changes in precipitation between the reference period and the future, we obtained a conclusion that the average precipitation of the basin in the future period will decrease in both scenarios.

The results also showed that the number of CDDs will increase with a possible decrease in precipitation during the next period. This can be considered as a drought potential index and based on increasing greenhouse gases (Sillmann et al. 2013). Other indices (Rx1day, Rx5day, R95p, R99p, CWD, SDII, and PRCPTOT) based on both scenarios are reduced compared to the baseline period. In the Lake Urmia basin, there is a positive and negative relationship between the increase in rainfall with the rate of CWD and CDD, respectively, which has been confirmed by many studies in other regions of the world (Sharma et al. 2020; Wang et al. 2021). Also, the slope of significant trends in RCP8.5 is higher than in RCP4.5.

The results of this study have important implications for the management of water resources in climate change conditions. Increasing the frequency of CDDs, decreasing precipitation, and decreasing wet days may make the soil drier. This may induce a serious effect on the hydrological cycle. Also, the increase in frequent dry periods causes a decrease in soil fertility, which in turn causes a decrease in agricultural production. Moreover, the decrease in agriculture and the ever-increasing population growth may pose a serious threat to food security in the Lake Urmia basin, which is the agricultural pole of Iran. Therefore, it is necessary to implement comprehensive policies to prepare for future water shortages in the basin, by quantifying the amount of precipitation decrease and temperature increase.

The authors are grateful to the Editor and to the reviewers for their valuable and constructive comments. We also gratefully acknowledge the project funding provided by Urmia Lake Research Institute, Urmia University, Iran. The work by Park was supported by the National Research Foundation of Korea (No. 2020R1I1A3069260).

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.