The future of extreme climate in Iran

Iran is experiencing unprecedented climate-related problems such as drying of lakes and rivers, dust storms, record-breaking temperatures, droughts, and floods. Here, we use the ensemble of five high-resolution climate models to project maximum and minimum temperatures and rainfall distribution, calculate occurrences of extreme temperatures (temperatures above and below the historical 95th and 5th percentiles, respectively), analyze compound of precipitation and temperature extremes, and determine flooding frequencies across the country. We found that compared to the period of 1980–2004, in the period of 2025–2049, Iran is likely to experience more extended periods of extreme maximum temperatures in the southern part of the country, more extended periods of dry (for ≥120 days: precipitation <2 mm, Tmax ≥30 °C) as well as wet (for ≤3 days: total precipitation ≥110 mm) conditions, and higher frequency of floods. Overall, the combination of these results projects a climate of extended dry periods interrupted by intermittent heavy rainfalls, which is a recipe for increasing the chances of floods. Without thoughtful adaptability measures, some parts of the country may face limited habitability in the future.

Floods and droughts have been occurring all the times in the past, but previous research shows that these occurrences are happening in increasing rates and that the changes can be dominantly due to anthropogenic activities 1 . A sequence of processes due to increasing greenhouse gasses, could be summarized as (i) increases in air temperature and its capacity to hold more water 2,3 , (ii) accelerated and irreversible melting of permafrost 4 , glaciers 5 , and ice caps 6 adding more water into the atmosphere, and (iii) increases in plant biomass and evapotranspiration 7,8 . The net result is a transfer of more surface water into an atmosphere capable of holding more water. More substantial accumulation of atmospheric water will cause a higher frequency of high-intensity and short-duration precipitations 3,[9][10][11] , significantly increasing the chances of flooding in different parts of the world.
In recent decades, observed climate data clearly show a warming trend in many parts of the world, resulting in a wide range of climatic impacts 3,9,[11][12][13] . In Iran, a country dominated by an arid and semi-arid climate, significant climate anomalies have been observed 14 . In combination with management related issues, Iran has been confronted with many disasters from shrinking a significant number of lakes and river, to land subsidence, floods, and droughts. Lake Urmia -the largest lake in the Middle East and one of the world's largest hypersaline lakeshas significantly shrunk 15 and given the status quo, it may completely dry up in 6-9 years 16 . Hamun lake in the east of Iran, Parishan and Shadegan lakes in the south 17,18 , and Zayandeh-Rud river in the center of Iran 19,20 are also at risk of disappearing due to mismanagement and climate change. Iran also has extremely critical conditions in groundwater resources because of overexploitation, and the country ranks among the top groundwater miners in the world 21 . In this critical water condition, increasing frequency of floods causes severe damages to different levels of water infrastructure, economy, and society at large. From 2015 to 2018, approximately six major floods occurred in unexpected regions located in arid and semiarid parts of the country 22 . In addition, floods in the northern edge of the country often cause substantial damage. The worst flooding disaster occurred on August 2001, where a once in 200-year-flood, affected more than 27,000 people, rendered 10,000 homeless, and killed about 247 people in Golestan province in northern Iran 23,24 .
In the studies of the impact of climate change, most previous projections have been conducted at the decadal time scale to show the trend of changes in temperature and precipitation at the mid to end of the 21 st century [25][26][27][28][29] . However, extremely dry and wet periods, as well as floods in intra-annual temporal scale in arid and semi-arid countries, like Iran, have not been thoroughly studied, even though their impacts have been severe. Recent studies  of extreme events have mostly focused on single-driver climate indices, such as annual precipitation, maximum one-day precipitation, number of days above a certain threshold, annual frequency of warm days and nights, etc., in arid and semi-arid regions [30][31][32][33] . However, there are disagreements on the value and frequency of these indices across the country attesting to the significant uncertainties in the past and future climate data and the period of study.
Although the risk of extreme precipitation or temperature events may extend over a large geographic area, the vulnerability to flooding or drought events is a highly local phenomenon. Hence, while commonly used indices mentioned above are useful in predicting the extreme trends, they are of little use for assessing local floods and droughts. In this article, we look at compound extremes (i.e., the simultaneous occurrence of multiple extremes) of dry and wet periods in Iran and also identify past floods and the associated climate conditions (in terms of duration, intensity, and extent of precipitation) in different locations of the country. Based on the learned knowledge in each location, we predict the frequency of future floods by searching for similar climate patterns in the future data. With no changes in the hydro-morphological regime of the region, we could expect a similar or worse climate condition could lead to similar or worse floods.

Data and statistics
The elevation of Iran ranges from less than −28 m at the Caspian sea to 5,610 m in the Damavand peak of Alborz Mountain chain. Alborz Mountain in the north and Zagros Mountain in the west play a significant role in dissecting the country into various climatic zones. These mountainous regions block moisture to reach the central part of the country, which receives little rain and hosts one of the hottest deserts in the world, the Lut Desert. Approximately 88% of Iran is located in arid and semi-arid regions ( Fig. 1) 34 . The mean annual precipitation during the study period  is 253 mm for the entire country, ranging from 144 mm to 342 mm per year. Iran receives less than a third of the world's average precipitation 15 .
Despite the generally low precipitation, there have been large floods recorded in Iran. We used 6 floods, which occurred during 2015-2018, for the analysis of their future probable occurrences in this study (Table 1) 35,36 . Several physical factors are involved in the generation of floods such as slope, geology, landuse, soil, climate, and topographic wetness index 37 . It is thus often very difficult to quantify the contribution from individual variables 38 . Many model simulations, therefore, must be carried out with different variable settings 39 . In such cases, an ensemble of model runs needs to be executed for long time periods to reduce model uncertainty 40 .
To avoid such tedious and uncertain calculations, in this study we suggest a more straightforward approach for parameterizing a flood condition by using statistics of previous floods. In this approach, we quantify past floods considering the volume of water generated by precipitation at different locations within the flooded basin every day before flooding and project this data into the future by a search routine that identifies similar or worse conditions. We define the basin as the area that contributes to flooding water at its outlet (or flood point). In this calculation, we are not assuming that floods would not occur in any other way, but merely calculate a likely flood if similar conditions of a past flood or worse happen in the future (see Methods for more detail).
We cannot evaluate the future climate predictions by Atmosphere-Ocean General Circulation Models (GCMs) for their correctness. For this reason, it is necessary to run a large number of models and Representative Concentration Pathways (RCPs) to quantify the uncertainties. Flood and extreme climate studies have widely used GCM data [26][27][28]41 . In this study, we used four sets of climate data that included: (i) 122 observed station data   42 , (ii) 0.5° gridded historical (1980-2004) data 43 , (iii) historical parts of five widely used global 0.5° GCM models   (Table 2) 44,45 , and (iv) the future simulations of the five GCM data (2025-2049). We used the observed data in (i) to bias correct the 0.5° gridded data in (ii) to obtain a uniformly distributed grid across the country. We then used the latter to bias correct the GCM data for RCP4.5 and RCP8.5 scenarios (see Figs S1-S3 in Supplementary Material for bias correction results). The grid data is often successfully used as observed data in places where sufficient observation is lacking 46,47 (see Methods for more detail).

Results and Discussion
Future maximum and minimum temperatures. The ensemble of 5 GCMs shows an increase of 1.1 to 2.75 °C in maximum temperature across Iran (Fig. 2). The two RCPs show similar patterns of change with differing magnitudes. RCP8.5 shows a sharp temperature rise of 2 to 2.75 °C in most parts of the country. The coefficients of variation show the degree of agreement between the five GCMs. As illustrated in Fig. 2, there are greater agreements in the models in the central to southeastern parts of the country, mostly in the Hot Dry Desert and Hot Semi-Desert climate zones. Results also indicate a moderate agreement in the west and the east in the primarily Mediterranean climate.
Extreme temperatures are wreaking havoc in many countries around the world with an increasing number of deaths 48 . We calculated the 95 th percentile of maximum (Fig. 3 top)   of extremely hot temperatures in the Desert climates. Nonetheless, many regions across the country experience extremely hot temperatures of up to 30 days across different climate zones. The number of days of extreme cold temperature in the future, however, drops across the country and all areas are predicted to experience a fewer number of extreme cold temperatures (Fig. 4).

Future precipitation.
The ensemble of all models for both scenarios predicts no significant change in the entire country's average annual precipitation during the future study period. However, spatially, there may be a considerable change of up to ±100 mm year −1 (Fig. 5). Unlike temperature, precipitation is projected quite differently by the two RCPs. RCP4.5 predicts a rise of up to 100 mm year −1 in the Mediterranean and Semi-Desert climates and the Caspian zones and a modest increase in the central Cold Semi-Deserts. In this scenario, similar to a previous study 29 , the wet regions of the country get wetter, while the dry areas get drier. In contrast, scenario RCP8.5 predicts a significant decrease in the precipitation of about 100 mm year −1 in the Hot Semi-Desert areas and a relatively stable Caspian zone climate.
Compound analysis of dry and wet conditions. The compound analysis allows for consideration of several variables simultaneously. Here, we used the compound analysis to identify extreme dry and wet periods across Iran using precipitation and temperature. Compound extremes exert the most substantial impacts on the environment. To demonstrate, for an extremely dry period, we assumed a condition where (for ≥120 consecutive days: precipitation <2 mm day −1 and Tmax ≥30 °C), and for an extremely wet period, we assumed a condition where (for ≤3 consecutive days: the total amount of precipitation ≥110 mm). These conditions were chosen subjectively here and could differ for different regions. The compound analysis could include other variables such as soil moisture, humidity, evapotranspiration, or crop yield, subject to data availability.
Future analysis of extreme dry periods showed a significant 16-fold increase in most of the country south of the Alborz Mountain chain (Fig. 6). The increase in dry periods corresponds well with the increases in the extremely hot temperatures illustrated in Fig. 3. The Caspian and Wet Mediterranean zones in the north may, however, experience slightly fewer dry periods than before. Except for the Hot Dry Desert, the rest of the country is also projected to experience significant increases in extremely wet periods (Fig. 7). The significant increases in both extremely hot and wet periods simultaneously indicate long dry periods intermittently interrupted with precipitations of high intensity and short duration, which is a recipe for a higher probability of drought and flood conditions. Future Floods. We simulated the conditions of the most recent floods (Table 1)  What about natural climate variability? Simulation of mid-latitude atmospheric blocking has always been a weak point of GCM models 49 . Although some improvements have been reported 49,50 , still considerable errors exist in GCMs as a result of blocking. Iran, because of its location in the mid-latitudes, is affected in different seasons by anticyclones and blocking patterns 51 whose frequency can be expected to increase under conditions of global warming 52 . For instance, mountainous regions in western Iran (Zagros Mountains range) are influenced by the Mediterranean cyclones 53,54 , whereas southeastern fringes of Iran are affected by the Monsoon weather phenomena 55 . Also, El Niño-Southern Oscillation is reported to be responsible for intensification of March-April floods compared with normal conditions 56 . Blocking causes alteration of weather patterns leading to floods, droughts, unusual temperatures, and other weather extremes. The results reported in this study, therefore, should be taken as likely events under the 'new normal' future climate that could be modified by various occurrences of natural climate variability.
In summary, our study of future climate in Iran depicts a grim picture concerning more frequent extreme wet and dry periods, more extended periods of extremely hot temperatures, and higher frequency of floods across the country. Combination of these events, especially in the three Desert climate zones, may create an uninhabitable living condition as also suggested by other studies 48 . More resilient multipronged adaptive measures, therefore, must be taken to protect the people and the cities from these disasters. Methods Flood volume calculation. We first identified the flooded points, then used GIS to draw the contributing basin boundary. After determining the climate grids that fell in the corresponding basins, we calculated weighted-average precipitation multiplied by basin area to obtain the total volume of water in the basin going through the flooded region in a day using the following expressions: where V j is volume of water generated by the jth climate grid in the basin in a day (m 3 ), R is the precipitation (m day −1 ), A is the tributary area of a precipitation grid (m 2 ), V d is the volume of water generated in one day in the basin by precipitation from all grids (m 3 ), and V T is the total volume of water generated in D days of rain. If a historical flood occurred after D rainy days, then a future flooding condition is formulated as: Future Flood = True, if: AND ( , ) AND ( ) where F and H stand for future and historical, respectively.  Table 2). Based on preliminary analyses and comparison tests, the CFSR data, after bias correction, provided a reasonable estimate of precipitation across the country. First, we bias corrected the CFSR data based on the 122 observed metrological stations, then GCMs were bias-corrected based on corrected CFSR data. For bias correction and pattern recognition of a historical flood condition, we used the Climate Change Toolkit (CCT) program 57 .
A key aspect of the climate change impact study is the spatial and temporal downscaling of the GCM results. In this study, the GCM data were downscaled using the nearest observation station. For precipitation, we used a linear correction method. GCM daily precipitation amounts, p, were transformed into p* using a scaling factor α such that p* = αp, where α = O P O / , , is the average monthly observed precipitation, and P is the average monthly GCM precipitation. Here, the monthly scaling factor was applied to each uncorrected daily observation of that month, generating the corrected future daily time series.
For temperature, we tested linear and nonlinear models as used in the literature 29,58,59 and chose a fourth-degree regression model based on the calibration and validation results of stations in different regions. In general, the results of a first-degree linear and a fourth-degree nonlinear model were similar except for small and large temperature values 29 , where the nonlinear model performed systematically better, especially for the validation data set. Hence we opted for the nonlinear model expressed as: