Assessment of impact of climate change on the streamflow of Idamalayar River Basin, Kerala

The availability of freshwater remains one of the essential perquisites of human existence on the planet. The diminishing quantity and quality of freshwater and its high sensitivity to climate change call for effective and sustainable utilization of this resource in the present and the future (Mandal & Simonovic 2017). Therefore, it is essential for water resources managers and decision-makers to take climate change into account. Several studies have investigated the impact of climate change on water availability in different parts of the world. The general approach to investigate the effect of climate change on the hydrological regime is to simulate the hydrological response of the catchments using a hydrological model utilizing the climatic variables projected by general circulation models (GCMs) (Zhang et al. 2016).

Hydrologic models are generally classified into lumped, semi-distributed, and distributed models. Soil and Water Assessment Tool (SWAT) and Hydrologic Engineering Centre – Hydrologic Modeling System (HEC-HMS) are the two semi-distributed hydrological models that are widely used to simulate the responses of a basin. SWAT has been extensively used in modelling streamflow, water quality, and sediment yield. This model has been successfully configured for different river basins across the globe. It has been identified as a valuable and popular tool for environmental and climate change studies. Yaduvanshi et al. (2018) used SWAT to model the streamflow under extreme rainfall events. They found that it performed well in simulating hydrologic responses under extreme events. The SWAT model has also been used in studies investigating the impact of land-use change and climate change on streamflow. For instance, Bhatta et al. (2019) examined the effect of climate change on a Himalayan river basin using SWAT. They compared the impact of the complexity of the model on the results based on the number of sub-basins, hydrologic response units (HRUs), and elevation bands. They successfully configured SWAT for the river basin and concluded that including different elevation bands improved the simulation results. In a similar study by Sharma et al. (2022), the SWAT model was used for modelling the impacts of land-use and climate changes on the water availability of the Dharoi River basin, India. They compared the model's performance based on different calibration approaches and concluded that the SWAT model was satisfactory in simulating monthly streamflow.

HEC-HMS is another semi-distributed hydrological model widely used for event-based as well as continuous simulations. Oleyiblo & Li (2010) have used the HEC-HMS model for Misai and Wan'an catchments in China to study its applicability in flood forecasting. The results showed that HEC-HMS is suitable for the selected catchments, with a satisfactory prediction of the time of occurrence and volume of flood events. Wijayarathne & Coulibaly (2020) used the HEC-HMS model along with other hydrologic models for the Waterford River watershed, Canada, to study its applicability in streamflow simulation and flood forecasting. They concluded that the model reasonably simulates the streamflow and recommends it for operational use in predicting floods. Palacios-Cabrera et al. (2022) used the HEC-HMS model to study the hydrodynamic response of a semiarid Mediterranean karstic watershed based on land-use change and precipitation cycles. They reported that HEC-HMS gave satisfactory values for the performance indices considered. In general, both the SWAT and HEC-HMS models have been successfully applied to different watersheds and are suitable for streamflow simulation.

In addition to hydrological models, downscaled and bias-corrected GCM outputs are also required for assessing the impact of climate change. The GCMs developed in the sixth phase of Coupled Model Inter Comparison Project (CMIP6) of the Intergovernmental Panel on Climate Change (IPCC) are the most advanced tools for projecting future climate. The CMIP6 projections are available under different shared socio-economic pathway (SSP) scenarios.

By using GCMs, studies have been conducted worldwide to examine the impact of climate change on water resources. In a study by Zhang et al. (2016), the hydrologic responses of the Heihe River basin in China were investigated for future climate change scenarios using SWAT and the statistically downscaled climate change data. They found that warmer and wetter trends will be continued in the future. In a similar study by Luo et al. (2019), the impact of climate change in nine high-alpine catchments across Xinjiang, China, under two representative concentration pathway (RCP) scenarios, RCP 4.5 and RCP 8.5, was investigated using the SWAT model. It was reported that the future water resource changes in Xinjiang are not sustainable and will lead to a decrease in discharge. Gao et al. (2021) studied the impact of climate change on the future streamflow of the Laixi River basin, China, using statistically downscaled CMIP5 GCM outputs and SWAT model under three RCP scenarios (RCP 2.6, RCP 4.5, and RCP 8.5). The streamflow was projected to rise gradually during future periods (i.e., 2050s and 2080s) under all three scenarios. By using bias-corrected Regional Climate Models (RCM) outputs and the Hydrologinska Brans Vattenbalansavdelning (HBV) hydrological model, Abdulahi et al. (2022) investigated the effects of climate change on streamflow in the Upper Awash River basin, Ethiopia, under the RCP 4.5 and RCP 8.5 scenarios. The findings indicated that streamflow would rise under both RCPs following the precipitation increment.

Similar studies have been conducted in the Indian subcontinent also. Mudbhatkal et al. (2017) assessed the possible impact of climate change on the hydrology of the Malaprabha and Netravati River watersheds in India. They reported that the effect of climate change is evident, but each river responds differently to climate change. In a study by Kumar et al. (2017) on the impact of climate change on different water balance components in the upper Kharun catchment, it is reported that the flow will decrease in the 2020s and increase in 2050s and 2080s. Bisht et al. (2020) assessed the impact of climate change on the streamflow characteristics of the Mahanadi basin. The mean monthly flow was found to be increasing in the far-future (FF) period (to the end of the century). Jaiswal et al. (2020) have conducted a similar study to analyse the impact of climate change on the command area of Tandula dam, India. They found that there would be a decrease in the reservoir water supply due to reduced rainfall. In a similar study by Sadhwani et al. (2023), the impact of climate and land-use changes in the Periyar River Basin, Kerala, was investigated using a SWAT model. They pointed out that there are possibilities of reduction in average annual streamflow.

The results of all these studies substantiate the necessity of regional-scale research on climate change impacts on hydrological regimes. For instance, Kerala has witnessed severe floods in the monsoon for two consecutive years, in 2018 and 2019. In the 2018 flood, all 14 districts in Kerala were affected (Pramanick et al. 2022). The heavy rainfall and the associated change in discharge complicated reservoir operation. Localized intense rainfall in the upstream and subsequent releases from reservoirs caused flooding in the downstream (Pramanick et al. 2022). The study by Hunt & Menon (2020) has linked the 2018 flood to climate change. It was reported that the future climate would possibly exacerbate the chances of flooding. Sudheer et al. (2019), in a study of the role of dams in flooding in the Periyar River Basin, Kerala, concluded that the operating rules of the reservoirs should be revisited to mitigate floods while serving the purposes for which they were designed. For this, knowledge of the future streamflow is required. Idamalayar dam is among the major dams in the Periyar River Basin, with a live storage capacity of 1018 MCM. Also, the Idamalayar catchment received the highest rainfall in August 2018 (Sudheer et al. 2019). The current investigation examines how climate change affects the streamflow from the Idamalayar catchment, Kerala, using a hydrological model.

The development of hydrological models often requires spatial and temporal data (Luo et al. 2019; Haro-Monteagudo et al. 2020). The spatial data requirement includes topography, land use, and soil maps. The temporal data required include hydro-meteorological data. Table 1 shows the different datasets and their sources used for the current study.

The daily meteorological data for 1987–2017 were collected from Indian Meteorological Department (IMD). The daily rainfall was acquired from IMD as 0.25 × 0.25° gridded data (Pai et al. 2014). The original resolution of daily temperature data was 1 × 1° (Srivastava et al. 2009). Using the bilinear interpolation, the daily temperature was regridded to 0.25 × 0.25° to match the grid points of rainfall data. The streamflow for 1987–2017 was collected from Kerala State Electricity Board (KSEB), Idamalayar division.

The geographical data used in this study include digital elevation model (DEM) data, soil data, and land-use data. The Cartosat DEM with a resolution of 30 m was obtained from Bhuvan. The elevation of the catchment varies from 53 to 2,575 m. The required land-use map was prepared from Landsat 8 images using the maximum likelihood classification tool in ArcMap. Training samples are created by visual inspection of the image. Signature file is created from these training samples and is used for classification. The primary land-use class in the catchment is forest (87.25%), followed by water (6.92%) and rangeland (5.83%). The required soil map was obtained from the Food and Agricultural Organization (FAO) of the United Nations. The soil classes in the study area include clayey loam (63.58%) and sandy clay loam (36.42%).

The primary tools for deriving future climate data are the GCMs. Due to their coarse scale, GCMs intended to project climate data cannot be used directly at the regional and basin levels (Jaiswal et al. 2020). They must be downscaled to a fine resolution (Luo et al. 2019). Different statistical and dynamical methods are available for downscaling (Shaaban et al. 2011). Along with downscaling, bias correction must be done to eliminate the systematic error caused by the coarse resolution (Mishra et al. 2020a; Kim et al. 2022). Mishra et al. (2020a) downscaled precipitation and maximum and minimum temperatures from 13 GCMS using empirical quantile mapping and developed a dataset at a daily timescale at a spatial resolution of 0.25 × 0.25° for South Asia and 18 river basins of the Indian subcontinent. The bias-corrected dataset (Mishra et al. 2020b) from the 13 CMIP6 GCMs for the historical (1990–2014) and projected (2031–2040, 2051–2060, and 2071–2080) periods for the three SSP scenarios (SSP126, SSP245, and SSP585) are used in the present study.

From the bias-corrected dataset, five GCMs are selected using a multicriterion decision-making method, the Preference Ranking Organisation Method of Enrichment Evaluation (PROMETHEE-2). The projected precipitation, maximum temperature, and minimum temperature GCM data are compared with the observed data of the baseline period (1990–2017) for this purpose. These projected climate data are input to the hydrological model to study the change in water availability in the Idamalayar basin. Finally, the output is used for frequency analysis to estimate streamflows for various return periods.

The US Army Corps of Engineers (USACE) developed the HEC-HMS model to route a runoff hydrograph through a stream network and calculate watershed discharge over time (USACE 2018). It is widely used for event as well as continuous simulations. A control specification, a meteorological model, and a basin model make up a typical HEC-HMS project. The time constraints of the model are specified in the control specification. The basin's rainfall is given using the meteorological model. The basin model represents the physical watershed with the help of hydrologically related stream reaches, sub-basins, and junctions.

In the present study, the deficit and constant loss method is used to model the infiltration losses, and the Clark unit hydrograph method is used to compute direct runoff. The recession model is employed to model baseflow, and the Muskingum routing model is used to simulate the streamflow in the reaches.

These two models are used for simulating the daily streamflow from the Idamalayar River Basin. In the SWAT model, the simulation is carried out from 1987 through 2017. In HEC-HMS, the simulation is carried out from 1990 to 2017. The change in the start date of simulation in the two models is due to the application of a warm-up period of 3 years in SWAT.

Model calibration is the process of adjusting the model parameters so that the simulated output matches the observed data. For the automatic calibration and validation of SWAT, a standalone tool named SWAT-Calibration and Uncertainty Program (SWAT-CUP) is used. SWAT-CUP provides different options from which the Sequential Uncertainty Fitting Algorithm (SUFI-2) is employed in the current study. Along with auto-calibration, the SUFI-2 algorithm ranks the significant parameters with respect to their sensitivity (Bhatta et al. 2019). Detailed information regarding SWAT-CUP and SUFI-2 algorithm is available in the study by Abbaspour (2012). The HEC-HMS model supports manual calibration in addition to its auto-calibration system. In the present study, the models were calibrated for a daily time scale from 1990 through 2007 and validated for 2008–2017 using the respective methods.

The efficacy of the models was evaluated using four goodness-of-fit measures, i.e., R², NSE, PBIAS, and RSR. R² ranges from 0 to 1, with 1 denoting the ideal value. Another performance measure is NSE, which describes the relative magnitude of the residual variance compared to the variance of the observed data as a range between −∞ and 1.0, with 1 denoting the ideal value. An NSE greater than 0.50 is considered to be satisfactory for a daily time scale model, according to Moriasi et al. (2007). PBIAS measures the simulated data's average tendency to be larger or smaller than their observed counterparts (Moriasi et al. 2007). The optimal value of PBIAS is zero. Positive values of PBIAS indicate model underestimation bias and a negative value indicates an overestimation bias. Singh et al. (2005) have recommended another model evaluation statistic, i.e., RSR. The lower the RSR, the better the model's performance.

The best five GCMs are selected based on several performance measures from the bias-corrected dataset using multicriterion analysis, PROMETHEE-2. Raju & Nagesh Kumar (2020) describe many performance metrics used to rate GCMs in various studies conducted worldwide. The current study chose four performance indices: R, NSE, normalized root-mean-square error (NRMSE), and the absolute normalized mean bias error (ANMBE) to evaluate the GCM and rank them. The value of R shows the linear correlation between observed and projected data, ranging from −1 to 1. A positive R indicates a strong correlation, and a negative R implies a weak correlation. NRMSE determines the similarity of the two time series by considering mean and standard deviation. It is defined as the root-mean-square error divided by the corresponding standard deviation of the observed field. Smaller values of NRMSE imply better model performance. Ideally, a value of 0 is preferred. ANMBE is the ratio of the mean difference between the observed and projected values to the mean of observed values. A small value of ANMBE is preferred.

The entropy method, which has the advantage of reducing the bias to any indicator, was used to determine the weight of the indicators for each climate variable (Raju & Kumar 2014). Then the PROMETHEE-2 method was applied to find the net ranking of each GCM based on each climate variable. The overall ranking of the GCMs is found by adding the net ranking indices by giving weight to each climate variable (0.5 for precipitation and 0.25 each for maximum temperature and minimum temperatures).

The best five GCMs performed well in describing the historical precipitation and temperature are identified. The climatic variables for the future period are obtained using these five GCMs. The changes in climatic variables for three future periods, say, NF, MF, and FF, are analysed against the baseline scenario (1990–2017). The analysis is carried out for three SSP scenarios, i.e., SSP126, SSP245, and SSP585.

Using the projected climatic variables for the five GCMs, under three scenarios, the streamflow from the river basin is simulated for three future periods. This simulated output is then compared to the streamflow for the baseline scenario to understand the impact of climate change on water availability in the river basin.

Generalized extreme value (GEV) distribution is utilized for flow frequency analysis. GEV is a combination of three distributions, i.e., Gumbel, Frechet, and Weibul distributions, and is widely used for flow frequency analysis (Mandal & Simonovic 2017). The flow frequency analysis is conducted utilizing the ismev package in RStudio (Coles 2001).

Figure 3(a) and 3(b) shows that both models captured the low and peak flows satisfactorily. In addition, both models perform well in capturing the flow pattern. But while comparing the performance of the models, it was observed that the values of evaluation indices were higher for the SWAT model, which reflects the suitability of the SWAT model over the HEC-HMS model in reproducing the observed streamflow from the Idamalayar catchment. Hence, the SWAT model was used for simulating the streamflow for future periods under climate change conditions.

The results of the ranking of 13 GCMs are presented in Table 3. Based on the ranking, the top five GCMs, namely, EC-EARTH3, EC-EARTH3 VEG, CAN-ESM5, Nor ESM 2 LM, and MPI ESM 1-2 LR, were used in the study for further analysis. These will be called GCM1, GCM2, GCM3, GCM4, and GCM5. The future climatic data, including precipitation, maximum temperature, and minimum temperature from these five GCMs, were analysed and compared to those in the baseline scenario (1990–2017). The calibrated hydrologic model used outputs from these five GCMs to simulate future Idamalayar streamflow.

Regarding the NF period of SSP245, the change in precipitation from the baseline period ranges from −17.5 to 6.8%, with the median lying at 2.8%. For the MF period under SSP585, the median lies at 2.3%. For the FF period, the median lies at 2.2%, with variations ranging from −20.3 to 7.3%. For all three future periods, the median lies around 2%, as shown in Figure 4(b).

While observing the box plots for the NF period under SSP585, it can be seen that the change in precipitation compared to the baseline period varies from −6.8 to 22.1%, with the median at 2.2%. For the MF period, the median lies at 5.2%. In addition, the range of variation is higher, which extends up to a 47% increase. The same pattern can be observed in the case of the FF period under the same scenario, with the median lying at 15% and the highest variation ranging up to 75%.

As shown in Figure 4(d), the decrease in maximum temperature in the NF period under SSP126 ranges from 2.04 to 2.95 °C with the median at 2.81 °C. Similarly, for the MF period, the median lies at 2.55 °C, and for the FF period, it lies at 2.52 °C. Under SSP245, in the NF period, the decrease ranges from 2.11 to 3.09 °C with the median at 2.76 °C. For the MF period, the range lies between 1.57 and 2.65 °C, with the median at 2.42 °C. For the FF period, median is at 2.16 °C. It is observed that the average decline across GCMs is higher in the NF period, which decreases in the MF period and further decreases in the FF period. A similar pattern can be observed under the SSP585 scenario. The median decreases across GCMs are 2.72, 2.19, and 1.78 °C for NF, MF, and FF periods, respectively. While observing the changes in monthly maximum temperature, it was found that generally, the decrease in temperature is higher during monsoon months (June–September) and lesser during summer months (February–May).

When analysing the changes in the average annual minimum temperature, patterns resembling those of the annual maximum average temperature were seen (Figure 4(g)–4(i)). The decrease from the baseline period tends to reduce in MF and FF periods compared to that in the NF period. This observation is also similar to the case of average annual maximum temperature. Similar tendencies to those of the monthly average maximum temperature were seen while taking the change in monthly average minimum temperature into account. While observing the monthly minimum temperature, the same pattern was seen as that of the monthly maximum temperature.

Comparing the projected climate variables to the baseline scenario gave essential insights regarding the future climate of the Idamalayar catchment. A slight increase in average annual precipitation is observed in some instances. A shift in the peak rainfall from July to October is also observed in most cases. It was also found that the average annual maximum and minimum temperatures are decreasing except in a few instances. The monthly variation of all these variables varies from GCM to GCM and in various SSP scenarios. These variations in climate variables will be reflected in the water availability in the river basin discussed in the following section.

In the case of the SSP245 scenario, for the NF period, the variation ranges between −21.91 and −44.35%, with the median at 27.58%. The median changes for the MF and FF periods lie at −29.64 and −28.72%. For the SSP585 scenario, in the NF period, the variation of streamflow from the baseline period ranges between −17.46 and −42.07%, with the median lying at −36.58%. At the same time, for the MF period, the range of variation extends between −16.75 and −44.11%. The median of the same lies at −30.59%. In the case of the FF period, the median lies at −21.90%. Figure 6(a)–6(c) shows that the extent of variation of streamflow from the baseline scenario across GCMs gets higher in the SSP585 scenario. The variation was lesser for the SSP126 scenario. Generally, the average annual streamflow is found to be decreasing. Similar results were also reported by Sadhwani et al. (2023), stating that there are more possibilities of a decrease in the average annual streamflow.

While comparing the average change in annual streamflow for the three SSP scenarios, it was observed that the decrease is highest under the SSP126 scenario and least under the SSP585 scenario. The comparison of average annual flows (historical and projected) is shown in Table 4. The results indicate a decline in the projected average annual flow. In addition, the comparison of average historical and future seasonal flows is also shown in Table 4. It shows that the flow during the south-west monsoon decreases by 48.57%, whereas the flow during the north-east monsoon increases by 25.28% in the NF period under the SSP126 scenario. The winter (December–February) flow increases by 108.40%, and the summer (March–May) flow is expected to decrease by 62.11%. This pattern of a decrease in flow during the south-west monsoon and summer and an increase during the north-east monsoon and winter seasons was observed for different future periods and SSP scenarios.

This study investigated the impact of climate change on the streamflow from the Idamalayar catchment. The efficiencies of two hydrologic models, i.e., SWAT and HEC-HMS, were assessed using four performance indicators. SWAT, which outperformed HEC-HMS in simulating the Idamalayar streamflow, was selected for further simulations. A multicriterion analysis (PROMETHEE-2) was used for ranking the GCMs. The changes in climatic variables such as precipitation, maximum temperature, and minimum temperature were analysed for three future periods, say, NF, MF, and FF, under three SSP scenarios, i.e., SSP126, SSP245, and SSP585.

The precipitation projections for the three SSP scenarios and future periods were observed to be different for all five GCMs. Under SSSP126, the average percentage change in annual precipitation for the five GCMs is positive (3.52%) in the NF period. It becomes negative in MF and FF periods (−1.16 and −3.57%, respectively). Under SSP245, the average change in precipitation is negative (−2.12%) in the NF period, while in the MF and FF, it is positive (16.06 and 11.09%, respectively). Under SSP585, precipitation is expected to increase in all future periods.

The maximum temperature shows a decreasing trend in the projections in all cases. The average change ranges from 2.38 to 2.68 °C under the SSP126 scenario. The decrease in maximum temperature under SSP245 scenarios ranges from 2.0 to 2.7 °C. In the case of the SSP585 scenario, this range lies between 1.13 and 2.71 °C. The smallest decline in maximum temperature happens under SSP585, while the highest decrease occurs under SSP126. Similar observations were found for minimum temperature. Generally, the minimum temperature decreases, with the highest and lowest declines projected under SSP126 and SSP585, respectively.

These outputs from the five GCMs were used to simulate the streamflow from the catchment. In general, it has been observed that the streamflow decreases in the future for different scenarios. An exception was observed for GCM3 under the SSP585 scenario. Under the SSP126 scenario, the average decline is around 26.6%, and the values are 21.94 and 9.24% for SSP245 and SSP585, respectively. Also, the peak flow shifts from July to October, i.e., from the south-west monsoon to the north-east monsoon, similar to precipitation. Frequency analysis showed that the variation of flows for different return periods is significant in the case of SSP245, in which outputs of GCM1 produced high flows.

In conclusion, the results indicate that the water availability in the Idamalayar basin will reduce in the future under the three SSP scenarios. There is a significant change in the flow pattern also. Both these necessitate the need for better management plans for the future. The analysis of the impact of climate change in the present study was limited to a particular region of Kerala. This can be extended to other river basins as well. Another vital area to be addressed is revisiting the rule curves of the Idamalayar reservoir, which will help to deal with the climate change impact on streamflow.

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

The authors declare there is no conflict.