Evaluating uncertainty in climate change impacts on peak discharge and flood volume in the Qaran Talar watershed, Iran

Surface water plays an important role in human life (Hasan et al. 2021; Ambade et al. 2022a). On the other hand, the occurrence of floods causes substantial economic loss to the residents of flood-prone areas (Kauffeldt et al. 2016). These damages are not only related to humid areas but also dry and semi-arid regions can be affected by flash floods (Niyazi et al. 2020). In addition, hydraulic structures are designed based on design discharge with different return periods, hence estimating flood characteristics is critical to preventing flood damage, and useful for the more efficient management of water resources (Youssef et al. 2020). Flood characteristics are strongly dependent on climatic variables (i.e., precipitation, evapotranspiration, etc.). On the other hand, the increase in daily emission of GHGs due to the rapid growth of industrialization caused global warming (Ambade et al. 2021). As a result, in the twentieth century, the earth experienced a steady rise in temperature (0.65 °C) owing to climate change (CC; Resende et al. 2019). It is expected that the increase in temperature will reach approximately 1.5–4.6 °C by the end of the present century (Georgoulias et al. 2022). CC can alter the hydrological cycle, cause changes in rainfall patterns, increase snowmelt, change river runoff and increase the occurrence of extreme events such as droughts and floods (Chu et al. 2019). Depending on region and weather conditions, CC can cause an increase (Zhang et al. 2017) or a decrease (Zhai & Tao 2017) in river runoff. Since river runoff mainly arises from rainfall, CC has a direct impact on the hydrological conditions of rivers (Alifujiang et al. 2021). In summary, water bodies are under serious stress due to anthropogenic activities (Ambade & Sethi 2021; Ambade et al. 2022b; Kurwadkar et al. 2022).

The AOGCM models, which are widely applied to generate CC scenarios, are the primary tools to predict climate variables for future periods (Tsujimoto et al. 2022). To study the effects of AOGCM outputs on river runoff, several methods such as hydrological modeling, statistical methods and Budyko hypothesis are used (Xue et al. 2021), the most common of which is the use of calibrated hydrological models (Li et al. 2020; Yan et al. 2020). For example, Gao et al. (2020) employed hydrological models to evaluate the CC impacts on runoff change. Yang et al. (2020) applied the WEP-L model to assess the contribution of human activities and CC on river discharge in the Northeast Tibet Plateau. According to the results, CC accounted for 81.7 and 71.5% of the variations in the runoff for two sub-basins, respectively. The Kan basin in Iran was evaluated using the outputs of Can ESM2 climatic model and IHACRES hydrological model by Ahmadi et al. (2019). They concluded that surface runoff under RCP2.6, RCP4.5 and RCP8.5 scenarios increased by 26, 28 and 33%, respectively, with respect to the baseline period. Runoff simulation of a mountainous basin (located in Swiss) using the HBV hydrological model showed that under changing climate, runoff tends to decrease in summer but increase in winter (Etter et al. 2017). Assessing the CC impacts on runoff in the Luanhe River basin using the SWAT model indicated that the amount of runoff under different emission scenarios will increase by 39–58% with respect to the base period (Yang et al. 2019).

Most studies have assessed CC impacts on a large scale. However, few studies have been conducted on a sub-basin scale. In most cases, regional hydrological processes are not similar to global ones. Therefore, studying the relationship between runoff and CC in specific basins can provide us with a largely complete understanding of water management in a region (Wang et al. 2019). The CC impacts on the local extreme hydrological events (e.g., peak discharge) are uncertain and complicated (Meresa & Zhang 2021). Meresa & Romanowicz (2017) assessed the main uncertainty on the issue of CC in the projection of hydrological extremes. Hydrological parameters and emission scenarios were considered for uncertainty assessment in seasonal flow projections by Joseph et al. (2018). By considering the uncertainty of bias correction methods on flood frequency, Soriano et al. (2019) stated that the magnitude of flood design was strongly affected by the climate model's biases. Therefore, assessing the effects of CC on flood design is associated with various sources of uncertainties (Meresa et al. 2022), such as AOGCMs, downscaling methods, emission scenarios and hydrological models (Zhang et al. 2014). This diversity in hydrological models, AOGCM, downscaling methods and different study areas usually leads to different and sometimes contrary results (Gao et al. 2020). Thus, it is momentous to consider the main sources of uncertainty in flood management under CC conditions.

As a semi-arid region, Iran receives an average annual rainfall of 250 mm. However, the country is one of the most flood-prone regions of the world. Each year, flash floods in Iran lead to hundreds of deaths, and the floods sometimes cause serious damage to various sectors, including agriculture and hydraulic structures around rivers. These damages occur mostly in the northern regions of Iran (i.e., the Caspian Sea coast), which receive higher annual rainfall. Given the importance and prevalence of the risks, it seems important to study the CC impacts on the Caspian Sea coast. Therefore, this research was designed to assess the changing climate impacts on the Qaran Talar watershed floods located in northern Iran. The region has an average annual rainfall of 950 mm and experiences yearly flood events. For example, the floods in the early spring of 2019 caused extensive damage to agriculture, buildings, hydraulic structures, roads, railways, etc., and caused waterlogging in some parts of the flood-affected regions for 10–20 days. While hydraulic structures are designed according to design discharge, one of the methods to determine design discharge is to use a combination of rainfall-runoff and design precipitation (DP) models. Since CC affects the amount of precipitation, and hence DP, it is important to determine the amount of DP under CC conditions in order to apply a more reliable design discharge when designing hydraulic structures.

In general, the major objective of this study is to determine the changes in the design discharge with different return periods across the upper reach of the Babolrood basin (Qaran Talar watershed) during 2021–2050 and 2070–2099. So far, few studies have investigated a combination of CC impacts and uncertainty sources on extreme flows. Therefore, our further aim was to examine two main uncertainty sources (uncertainties in future GHG emission scenarios and GCMs) in generating CC scenarios.

Considering the uncertainties of GCMs and emission scenarios, CC impacts on river floods were evaluated in four steps, as follows:

• 1.

  The Watershed Modeling System (WMS) runoff model was calibrated and validated using six rainfall-runoff events.

• 2.

  The 22 AOGCM models following two emission scenarios were used for the generation of CC scenarios in two future periods, namely, the near-future period (2021–2050) and the far-future period (2070–2099).

• 3.

  The design precipitations of the 2–500-year return period were determined for the recent past (1991–2020) and two future periods, and the runoff corresponding to each DP was simulated using the evaluated WMS model.

• 4.

  The shape of the hydrograph arising from design precipitations in the future periods was compared to those in the recent past period.

Hydrological models evaluate the impacts of climate and environmental factors on hydrology, water quality, etc. (Kim et al. 2021), and can solve the governing equations of water flow using the necessary information (Marko et al. 2019). The WMS is one of the comprehensive and new hydrological models developed by Brigham Young University (Samadi et al. 2019). The WMS model consists of several modules that can estimate surface runoff and hydrograph shape (Srinivas et al. 2018). To simulate rainfall-runoff, the drainage network and the physiographic characteristics of the watershed, including the slope, the area and the length of the drainage, are calculated from the Digital Elevation Model (DEM). Among the rainfall-runoff models available in the WMS model, the HEC-1 model can be run independently. Therefore, in this study, the HEC-1 model was used for simulating runoff. For the estimation of rainfall loss and runoff hydrograph, the Soil Conservation Service (SCS) and Snyder Unit Hydrograph methods were used, respectively. The weighted average of CN was estimated according to the land cover and land use.

A hydrological model consists of different parameters that define the characteristics of the model. Model evaluation is considered as the estimation of these parameters to match the output of the model to the observed values as far as possible (Devia et al. 2015). Some models use parameters such as soil and vegetation maps for the evaluation, but model evaluation with precipitation and discharge data usually increases the efficiency of the models (Kauffeldt et al. 2016). Among all the required data, the most important challenge in using hydrological models is a lack of past rainfall-runoff data (Youssef et al. 2020).

In this study, by examining daily and hourly rainfall, it was found that most of the historical daily and hourly rainfall are not proportional to each other. Therefore, from the precipitations of the past period, only six events were selected which showed consistent trends in daily and hourly precipitations; moreover, flood hydrograph was recorded in the gauging station. Three events were used for calibration, and a further three events were used for model validation. The parameters used for calibrating the model included the curve number (CN), percent impervious (PI), lag time (TP) and peaking coefficient (CP).

Since the outputs of AOGCMs are coarse at the spatial scale, it is not possible to use the output of these models directly in hydrological models (Cammarano et al. 2017). To overcome this limitation, downscaling methods can increase the spatial resolution of AOGCMs outputs (Pichuka et al. 2017). In the present study, the Long Ashton Research Station Weather Generator (LARS-WG) model was used for spatial and temporal downscaling of precipitation data, the ability and efficiency of which have been reported in previous literature (Khalaf et al. 2022).

Uncertainty in CC studies arises when several climate models or impact models are used, or when different hypotheses and methods are applied in employing these models (Senatore et al. 2022). Therefore, uncertainty sources in CC studies are substantially diverse, and it may not be possible to account for all uncertainty sources. Among all the uncertainty sources, AOGCMs and GHGs scenarios are of vital importance (Rahman et al. 2018). Hence, the uncertainty of these two sources was evaluated.

To generate daily precipitation scenarios, ΔP values (Equation (5)) were calculated for each of the AOGCMs following two emission scenarios (RCP4.5 and RCP8.5). Then, the best distribution function was fitted to the values of ΔPs. The ΔP values were extracted at probability levels of 0.50, 0.75 and 0.90 (at risk levels of 0.5, 0.25 and 0.1, respectively) from the cumulative distribution function (CDF). By importing observed daily precipitation of the baseline period (1971–2000) and ΔP values (at different risk levels) into the LARS-WG model, daily precipitation scenarios were generated for the two future periods.

The rainfall pattern for each basin should be determined using its rainfall data (Pani & Haragan 1981). Therefore, by examining the daily precipitation of the region, the general daily rainfall pattern was determined. Then, this pattern was used to convert the DR into hourly rainfall. Ultimately, by entering the design precipitations into the WMS model on a 15-min time scale, the flood hydrograph of these precipitations was simulated.

By determining the CN map, the weighted averages of CN and PI were calculated as 74 and 10%, respectively, and these values were used as a preliminary estimate in the calibration of the model. Using three storm events, the WMS model was calibrated (Table 2).

The relative error for peak discharge in the three rainfall events was about 10, 4 and 1% (averaging at 5%), but regarding FV, the model error was higher, and the average absolute value of the model error was about 10% in estimating the FV, which is an acceptable level of error (Table 3). Generally, considering the model error in estimating and FV, it can be said that the model simulated these two parameters with remarkable accuracy.

NRMSE and Er (Table 4) determined that the model error in hydrograph estimations was about 16 and 11%, respectively, indicating a high accuracy of the model in flood hydrograph simulation. The results of CRM also showed that the model generally tends to overestimate flood hydrograph for all three events. The R² value of the plot 1:1 for observational discharges and simulations was also in an adequate range. In general, it can be said that the model was able to estimate the and FV with an acceptable level of accuracy, and it showed good accuracy in estimating the flood hydrograph.

By fitting the best probability distribution function for ΔP data, ΔP values in the three probability levels of 0.5, 0.75 and 0.90 (risk levels of 0.50, 0.25 and 0.1, respectively) were extracted from CDF for the two future periods under each of the emission scenarios.

By importing the required data (ΔP values in the three levels of risk and daily rainfall of the baseline period) into the LARS-WG model, daily rainfall scenarios for the future periods were generated. Using daily CC scenarios, DP was calculated with different return periods (2, 5, 10, 25, 50, 100 and 500) for the future and the recent past periods (Table 5). In all CC scenarios, the DP (with different return periods) increased compared to the recent past period.

By entering the design precipitations of the past period (1991–2020) into the evaluated WMS, the two main characteristics of the hydrograph (i.e., and FV) in design floods were determined. Peak discharge varied from about 50 m³/s in the case of the DP for the 2-year return period to 300 m³/s for the 500-year return period. The values of FV ranged between 6.6 and 32.4 Mm³ for DP from 2 to 500 years of the return period (Table 6).

Table 6

Peak discharge (⁠⁠) and flood volume (FV) for the design precipitation in the recent past period

Return period (year) .	(⁠⁠) .	FV (⁠⁠) .
2	49.8	6.6
5	90.7	10.8
10	120.2	13.8
25	159.8	17.9
50	190.5	21.2
100	222.1	24.5
500	298.4	32.4

Return period (year) .	(⁠⁠) .	FV (⁠⁠) .
2	49.8	6.6
5	90.7	10.8
10	120.2	13.8
25	159.8	17.9
50	190.5	21.2
100	222.1	24.5
500	298.4	32.4

View Large

Figures 3 and 4 illustrated that the results of the RCP4.5 and RCP8.5 scenarios in the near-future are not very different so that the difference in the percentage changes in peak discharge in these two scenarios is less than 10% compared with those of the past period. For example, in the far-future, under the RCP8.5 scenario, and at two probability levels of 0.75 and 0.90, the changes in discharge and FV were higher than that of RCP4.5.

In all different scenarios and levels of probability, by increasing the return period, the percentage increase of and FV decrease compared to the recent past period. This is due to the use of a nonlinear rainfall-runoff relationship (SCS) in the WMS model. Since precipitation loss is a constant value (according to the constant CN number for the study area), it exerts a greater effect on precipitations with shorter return periods. The more the precipitation increases, the less strong this effect tends to be. On the other hand, as the level of risk decreases (i.e., by increasing the level of probability), and FV also show a significant increase.

In the RCP4.5 scenario and the near-future period at the probability level of 0.50, the increase in peak discharge in the return period of 2 years was about 35%, which decreased to 30% with the increase of the return period to 500 years, but by reducing the risk level to 0.10 (with a probability level of 0.90), the increase in peak discharge for the precipitation with the 2-year return period of was about 85% compared to the preceding period. This increase for discharge in the return period of 500 years reached about 70%. The trend of changes in FV was almost similar to that of peak discharge. The same was true for the RCP8.5 scenario for the near-future period.

For the far-future, the RCP4.5 scenario was somewhat similar to the near-future, but in the RCP8.5 scenario, the changes in discharge were greater than those of the near-future. In other words, if this scenario occurs, a larger increase in discharge and FV can be expected.

Regarding the CC in the study area (under two emission scenarios), it can be said that, in general, the amount of discharge and FV will increase, and the design of hydraulic structures without considering this issue would be likely to cause damage. Research conducted in humid areas also confirms the results of this research. For example, Hlavčová et al. (2016) studied a basin in central Slovakia with an annual rainfall pattern similar to that of the study area in the current research. They concluded that monthly discharge will increase (due to CC) in winter and spring. Pichuka et al. (2017) also predicted that CC will increase the number of extreme events and the resulting discharge. Etter et al. (2017) studied a mountainous basin and concluded that the discharge will increase in the winter months, but decrease in the summer months. Zhang et al. (2014) also examined three emission scenarios and concluded that under CC conditions, the discharge that would result from DP with different return periods would increase under all three emission scenarios for a basin in eastern China (with a humid climate). Sarkar & Maity (2020) showed that under the RCP8.5 scenario in the 2071–2100 period, the probable maximum precipitation will increase in 70–80% of regions of India.

To assess the uncertainty of CC scenarios, the percentage changes in and FV in different return periods were averaged (Table 7). Table 7 reveals that in the near-future, at all three levels of probability, the RCP4.5 scenario estimated the discharge and FV to be slightly higher than that of the RCP8.5 scenario (about 2%). This indicates that RCP4.5 and RCP8.5 scenarios yielded slight differences in the results and are characterized by very low uncertainty, but the difference in the results at different levels of probability (risk) that arise from the outputs of AOGCM models is high. So, by changing the probability level from 0.5 to 0.9, the peak discharge increases above 100%. Therefore, it can be concluded that in the near-future, the uncertainty of the results is only due to the uncertainty of the output of AOGCM models. These results hold for FV as well.

Table 7

The average percentage changes in peak discharge and runoff volume in different CC scenarios

Parameter .	Probability level .	Near-future .		Far-future .
		RCP4.5 .	RCP8.5 .	RCP4.5 .	RCP8.5 .
	0.50	30.7	29.0	27.7	26.0
	0.75	48.4	46.5	50.5	57.2
	0.90	70.3	65.9	75.8	94.3
FV	0.50	30.5	28.8	27.5	25.8
	0.75	48.1	46.2	50.2	56.8
	0.90	70.0	65.6	75.4	93.9

Parameter .	Probability level .	Near-future .		Far-future .
		RCP4.5 .	RCP8.5 .	RCP4.5 .	RCP8.5 .
	0.50	30.7	29.0	27.7	26.0
	0.75	48.4	46.5	50.5	57.2
	0.90	70.3	65.9	75.8	94.3
FV	0.50	30.5	28.8	27.5	25.8
	0.75	48.1	46.2	50.2	56.8
	0.90	70.0	65.6	75.4	93.9

View Large

In the far-future, the average percentage changes in discharge and flood volume (FV) in the RCP4.5 scenario were estimated to be about 27, 50 and 75% for the three probability levels of 0.5, 0.75 and 0.9, respectively. In the RCP8.5 scenario, however, these numbers were estimated to be around 26, 57 and 94%, respectively. Compared with the near-future, uncertainty related to emission scenarios increased in the far-future, which is attributed to different levels of greenhouse gas (GHG) emissions in these two scenarios. Toward the end of the century, the difference between GHG emissions in these two scenarios increases. However, the uncertainty of the results is mostly caused by the output of AOGCM models, so that the change in the emission scenario leads to an increase of about 10% in the changes of discharge and FV as compared to the past period; but, the change in probability level causes 200 and 300% increases in the changes of discharge and FV by the RCP4.5 and RCP8.5 scenarios, respectively.

Considering the mentioned results for the two future periods, generally, it can be said that the uncertainty of the results is mostly due to the output of AOGCM models, compared to the emission scenarios. Thus, in CC studies, the output of several AOGCM models should be used so that by averaging or evaluating the probabilistic outputs of these models, water structures can be designed with greater confidence.

In the present study, the WMS model was used to investigate the CC impact on the Qaran Talar watershed runoff in the northern of Iran. The WMS model was accurately calibrated and validated using observed data. By simulating hydrograph DP of 2–500 years, for the future periods, and FV were estimated to be in the ranges of 50–300 m³/s and 6.6–32.4 Mm³, respectively. CC scenarios were generated using 22 AOGCMs following two emission scenarios (RCP4.5 and RCP8.5) at three risk levels (0.10, 0.25 and 0.50) for two future periods ((2021–2050 and 2070–2099).

By determination of design discharge that resulted from design precipitations with 2, 5, 10, 25, 50, 100 and 500-year return periods, it was revealed that the and FV were not significantly different when comparing the outputs of the RCP4.5 and RCP8.5 scenarios for the near-future period. The difference in the percentage changes in discharge by these two scenarios, compared to the past period, was estimated to reach less than 10%. In the 2021–2050 period, under the RCP4.5 scenario, and the risk level of 0.50, the increase in the peak discharge for the 2-year return period has been estimated to reach about 35%, although it would get 30% for the 500-year return period. However, with a decrease in the risk level (risk level of 0.10), the increase in peak discharge for the DP of 2 and 500-year return periods were estimated to reach about 85 and 70%, respectively. The changes in FV were almost similar to those of the peak discharge. The same was estimated to apply for the RCP8.5 scenario in the near-future.

In the far-future, the RCP4.5 scenario performed almost similar to the near-future, although the changes in discharge were greater by the RCP8.5 scenario compared to the near-future. In other words, it is expected that in the pessimistic scenario (RCP8.5), there will be a greater increase in peak discharge and, thus, a greater amount of FV.

Assessments of the two important sources of uncertainty in CC studies also revealed that the uncertainty caused by AOGCMs is higher compared to GHG emission scenarios. As mentioned in the case of the near-future, the results of the two emission scenarios were similar, but regarding the end of the century, the difference between the emission scenarios increased, and the uncertainty related to the emission scenarios entered in the results of CC scenarios. Nonetheless, the uncertainty of the output by AOGCM models is still much higher than that of the emission scenarios.

In general, regarding the study area, it can be said that under CC conditions (two emission scenarios RCP4.5 and RCP8.5), the amount of discharge and FV will increase and the design of hydraulic structures is recommended to be updated according to the future-projected DP.

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

The authors declare there is no conflict.