Impact evaluation and analysis at a river basin scale under projected climate and land-use change

Studies in hydrology and water resources nowadays are focusing on the prediction of hydrological processes for the sustainable use of water resources. Water management issues will become more complex in the future, depending on the magnitude, rate, and regional characteristics of future climate change, and will be extremely troublesome for places with inadequate water resources (Hung et al. 2020). The freshwater ecosystems will be adversely affected by the variation in streamflow magnitude, timing and associated extremes. Anthropogenic climate change has caused recent changes in the balance of precipitation, runoff, and evapotranspiration (Raihan et al. 2021). The water scarcity and scuffles are symptoms of a growing disparity between water demand and supply. This is exacerbated by the unpredictability of available water due to climate change. Therefore, understanding the hydrological impact of climate change is critical for implementing effective climate change adaptation measures. The two main key factors affecting hydrological systems are climate change and land use/land cover change (LULC). In the previous few decades, changes in landuse/cover and climate have had a net negative influence on water resources in almost every region of the world, and this impact is anticipated to worsen in the future (Chawla & Mujumdar 2015).

Several prior studies have assessed the influence of climate change on streamflow in different watersheds with various climatic types around the world for current and future conditions; however, the majority of these studies did not include future land use changes in their studies (Zhang et al. 2007; de Oliveira et al. 2017; Dlamini et al. 2017; Luo et al. 2017; Su et al. 2017; Neves et al. 2020; Ndhlovu & Woyessa 2021; Quansah et al. 2021). Although few studies have attempted to incorporate climate change and LULC on streamflow prediction (Tarigan & Faqih 2019; Farinosi et al. 2019; Sinha et al. 2020; Raihan et al. 2021), investigating the relative impacts of climate change and LULC on regional hydrological processes under various scenarios is critical for establishing effective adaptation measures and assisting decision makers in achieving watershed stability (Haleem et al. 2022). Several studies have considered both climate and LULC for streamflow prediction (Ahiablame et al. 2017; Tarigan & Faqih 2019; Farinosi et al. 2019; Hung et al. 2020; Sinha et al. 2020; Haleem et al. 2022; Raihan et al. 2021), but none have considered the importance of region-specific climate model selection and bias correction methods when using the climate model simulations for hydrological evaluation.

‘Regional climate models (RCMs) are preferred over GCMs (general circulation models for regional/local impact studies because GCMs have a coarser resolution and cannot provide reliable outputs at the hydrological scale’ (Fang et al. 2015). However, RCM outputs still necessitate post-processing known as bias correction (BC) to remove systematic biases in simulated data (Maraun et al. 2010; Diallo et al. 2012; Fang et al. 2015; Mendez et al. 2020). The Coupled Model Intercomparison Project-5 (CMIP5) multi-model mean outperforms the CMIP3 multi-model mean for climate change impact studies and is more skillful at representing precipitation patterns (Annamalai et al. 2007; Sperber et al. 2013; Farinosi et al. 2019). Many research studies have proved the necessity for region specific selection of climate models (Raju & Kumar 2015; Ruan et al. 2018; Yang et al. 2020) and bias correction of RCMs (Fang et al. 2015; Luo et al. 2017; Mudbhatkal & Mahesha 2018) for reliable simulation of hydrologic processes.

Climate models always contain inherent uncertainties which may reflect significantly at local scale when evaluating hydrological processes. These uncertainties cannot be completely eliminated, but they are reduced in this study by selecting simulations from a climate model ensemble whose satisfactory performance over the study area has already been reported, as well as by selecting a suitable bias correction method by evaluating the performance of bias correction methods over the study area. This study focuses on the Meenachil River Basin, a major river in the Kottayam district of Kerala, India. This river can overflow or flood low-lying areas along the river during monsoon season, and it was a key contributor to the flood in Kerala, India in 2018. Saranya & Vinish (2021) identified multimodel ensemble of CMIP5 GCM-RCM pairs that perform well for simulating precipitation (Pr), maximum temperature (Tmax), and minimum temperature (Tmin) across the Meenachil River Basin. These climate models are utilised in the present study and taken into account for predicting future simulations. There has been no previous study in this river basin that quantitatively distinguishes and forecasts the effects of climate change and LULC change on streamflow. To have proper water resource management actions over the Meenachil River Basin, it is critical to identify and understand the transformation of climate and LULC in the Meenachil River and their impacts on fluctuation in streamflow.

The primary objective of this study is to evaluate the relative contributions of climate and land use change to streamflow as well as to quantify the integrated and individual impact of climate and LULC on future streamflow for the medium term (2025–2055) and long term (2056–2086) by hydrological modelling in Soil and Water Assessment Tool (SWAT). Climate data projected using GCM-RCM pairs from the Coordinated Regional Downscaling Experiment South Asia (CORDEX-SA) and LULC maps simulated using TerrSet LCM were used to drive the SWAT model for the future period. The annual and seasonal variation of streamflow was quantified for the medium and long term.

The hydrological model SWAT requires meteorological, topographic, soil and land use data as input for simulation. The meteorological parameters required for SWAT model simulation such as precipitation (Pr), maximum temperature (Tmax), minimum temperature (Tmin), relative humidity (RH), wind speed (WS) and solar radiation (SR) at a daily timescale were collected from India Meteorological Department (IMD) corresponding to the baseline period (1980–2010). In addition, daily precipitation data for the station Erattupetta within the study area was obtained from the Irrigation Design and Research Board (IDRB) Trivandrum. The daily discharge data corresponding to Kidangoor stream gauge station (1987–2017) was received from Central Water Commission (CWC) through Water Resources Information System of India (WRIS) (https://indiawris.gov.in/wris/#/). The Digital Elevation Model (DEM) at 30 m resolution corresponding to the study area was downloaded from https://search.earthdata.nasa.gov/search. The digital soil map of the world at 1:5,000,000 scale was collected from Food and Agricultural Organisation (FAO) of United Nations. The LULC maps for 1992, 2000, and 2008 were generated from Landsat imagery downloaded from https://earthexplorer.usgs.gov/. Climate variables such as Pr, Tmax, and Tmin simulated by the RCMs of the CORDEX-SA framework were obtained at 0.5° × 0.5° resolution from the climate data portal of Indian Institute of Tropical Meteorology (IITM), Pune (http://cccr.tropmet.res.in/home/data cccrdx.jsp) for the control period 1980–2005 and future periods (2025–2055 (medium term) and 2056–2086 (long term)). The above mentioned future projections corresponding to the high baseline emission scenario RCP 8.5 and medium stabilization scenario RCP 4.5 (Sanford et al. 2014) were chosen for this study.

In this study, the hydrological impact evaluation in terms of streamflow simulation was carried out using the semi-distributed hydrological model SWAT. The SWAT hydrological model is based on the water balance equation and simulates surface streamflow using the SCS curve number (CN) method (Arnold et al. 1998). This model has been proven to be effective for simulating at river basin scale and has been successfully used over the study area in previous studies (George & James 2013; Saranya & Vinish 2021). An initial simulation was run at daily time scale for the period 1980–2005 with observed climate data and all other input parameters to understand the performance of the model. Then the model was calibrated in SWAT-CUP software using the automatic calibration algorithm sequential uncertainty fitting version 2 (SUFI-2, Abbaspour 2015). Calibration was performed using monthly streamflow records from the Kidangoor station for the period 1987–2004. Sensitivity analysis was performed to determine the most influential parameters for streamflow simulation, which were then adjusted during calibration (Abbaspour 2015). The calibrated parameters were entered into the SWAT model, and the model was validated for the years 2005 to 2010. The land use map for the year 1992 and 2008 were respectively used for the calibration and validation period. The model performance during calibration and validation was evaluated in terms of performance measures recommended by Moriasi et al. (2007) such as Nash-Sutcliffe efficiency (NSE), root mean square error to the standard deviation of observed data (RSR), and percentage bias (PBIAS). The validated model was used to assess the performance of bias correction methods for simulating streamflow and projecting streamflow for the future.

The previous study (Saranya & Vinish 2021) over the Meenachil River Basin assessed the performance of climate models in simulating precipitation and temperature, as well as the hydrological impact of climate model combinations, and found the top performing climate model combination over the same area. Hence, we chose the ensemble average of the GCM-RCM pairs mentioned in their study for precipitation and temperature (Tmax and Tmin) projection, as well as streamflow projection. Table 1 summarises the five GCM-RCM pairs used for precipitation simulation and the five GCM-RCM pairs used for temperature simulation. Since the direct use of climate variables simulated by RCM is not recommended, bias correction is required for both precipitation and temperature. Separate BC methods for precipitation and temperature are available, ranging from simple to complex, but their performance varies by region (Mudbhatkal & Mahesha 2018; Mendez et al. 2020).

Saranya & Vinish (2021) evaluated the performance of bias correction methods such as linear scaling (LS), local intensity scaling (LOCI), modified power transformation (MPT), and distribution mapping (DM) for precipitation and LS, DM, and variance scaling (VS) for temperature in the Meenachil River Basin and found that all of them improve RCM outputs. However, the hydrological impact of the BC methods was not assessed in their study, which would be significant in reducing uncertainty in projecting future streamflow. As a result, the current study evaluated the hydrological impact of precipitation bias correction methods by simulating streamflow in SWAT with precipitation bias corrected by the LS, LOCI, MPT, and DM methods. Temperature (Tmax and Tmin) were bias corrected using the DM method only because the variability in temperature BC methods is very low when compared to precipitation (Chen et al. 2013; Fang et al. 2015).

The bias correction of the Pr, Tmax and Tmin was performed using the corresponding observed daily data available from IMD for the period 1980–2005 based on Teutschbein & Seibert (2012) and Smitha et al. (2018). The bias corrections were executed in CMhyd (Climate Model data for hydrologic modelling) tool for both precipitation and temperature (Rathjens et al. 2016). Thereafter, the performance of four bias-corrected precipitation (Pr)-temperature (Tmax and Tmin) combinations, such as LS-DM, LOCI-DM, MPT-DM, and DM-DM, in simulating streamflow was compared to streamflow simulation driven by observed precipitation and temperature. The performance evaluation of the simulated streamflow under the four combinations were done using time series based indices such as NSE, correlation coefficient (R), PBIAS, and mean absolute error (MAE), as well as flow quantiles. The bias corrected climate variables that generated satisfactory results for streamflow simulation during the baseline period were then used to project precipitation and temperature for the medium term (2025–2055) and long term (2056–2086) under both RCP 4.5 and 8.5 scenarios.

In order to understand the changes in LULC and to aid in projecting future changes in LULC, cloud free Landsat images for the years 1992, 2000 and 2008 were selected. The LULC maps for the respective years were then created by classifying the images into nine land use classes such as water, rubber plantation, mixed vegetation, urban area, paddy field, tea plantation, barren land, grass land, and forest using the maximum likelihood algorithm of image classification available in ArcGIS. Land change modeler was used to predict the future scenarios of LULC maps for the years 2030 and 2060. The TerrSet software from Clark Labs in the United States includes a land change modeler interface, which was used in this study. The three steps in land change prediction using LCM are change analysis, transition potential modelling, and change prediction (Clark Labs 2021). In the change analysis, the land cover changes from one map to the next were evaluated by comparing the LULC maps from 1992 to 2000. The potential of the land to transition was identified based on driver variables such as slope of basin, digital elevation model, distance to roads and distance to urban area for the transition potential modelling. The transition sub model was then modelled using a multilayer perception (MLP) neural network. Furthermore, the CA Markov model built in LCM was used to project the LULC map for 2008. This was then validated by comparing it to the LULC map created from Landsat image for the same year. Following validation, projected maps for the years 2030 and 2060 were created using LCM.

We used SWAT model for evaluating the individual as well as combined effect of climate and land use change over Meenachil River Basin by simulating streamflow for the baseline period (1980–2010) under different scenarios. Also, the projected variation in precipitation, temperature, and streamflow for the medium term (2025–2055) and long term (2056–2086) were assessed and discussed in the following sessions.

The most sensitive parameters for streamflow simulation found out by global sensitivity analysis using SWAT CUP software are shown in Table 2. The results indicate that streamflow is sensitive to both ground water and surface parameters. The best fitted values for these parameters (see Table 2) were given by the SUFI-2 algorithm during calibration (1987–2004).

The R² values (see Figure 2(c) and 2(d)) obtained for the calibration and validation periods were 0.8 and 0.9, respectively, which were also acceptable and confirmed the SWAT model's applicability for the study area and for future simulations.

The precipitation bias corrected by the LS, LOCI, MPT, and DM methods, and the temperature (Tmax and Tmin) bias corrected by the DM method, resulted in four combinations (LS-DM, LOCI-DM, MPT-DM, and DM-DM) for evaluating the performance of bias correction methods in simulating streamflow. Table 4 shows the performance of uncorrected and bias corrected climate variables for simulating streamflow on a daily and monthly scale using time series based indices. These indices were calculated by comparing the streamflow simulation driven by the four combinations to the streamflow simulated by observed precipitation and temperature for the period 1980 to 2005. Since the warm up period taken for SWAT simulation was 3-year, streamflow simulations were evaluated for the period 1983–2005. According to Table 4, streamflow simulated by raw meteorological outputs (uncorrected) from GCM-RCM pairs is biased with negative NSE values −0.25 and −0.29; PBIAS values 70.48 and 70.53 for daily and monthly streamflow. All bias correction combinations have significantly reduced the PBIAS, which ranges from −1.85 to −3.28. Also, NSE and R values were also significantly improved on a daily and monthly scale. The NSE value is widely used in hydrological models to assess how well simulations match observed data. At both the daily and monthly scales, LOCI-DM produced the highest NSE value of the four bias correction combinations, followed by DM-DM. The NSE value on a monthly scale for the LOCI-DM combination was 0.70 and 0.69 for the DM-DM combination.

In order to understand the temporal variability of run off and to aid in accurately predicting the future water availability, dependable flows or quantile flows were evaluated under each bias correction method. Flow rates are usually denoted with letter Q and a subscript number (eg. Q₁₀, Q₂₀, Q₃₀, etc.). The subscript number represents the percent of time for which the flow is equalled or exceeded in the river. Flow duration curves (FDC) typically provide a brief description of the magnitudes of flow rates in a basin. Based on FDC, dependable flows such as Q₁₀, Q₅₀ and Q₉₀ corresponding to the streamflow simulated by four combinations of bias correction methods were estimated and compared to the streamflow simulated by observed climate data (Table 5).

The aforementioned flow rates were evaluated in order to determine how well these bias correction methods could simulate high and low flow rates. It can be seen from the results when streamflow is simulated using climate variables biased corrected by DM-DM combination, the extreme flow rates are comparatively better represented than the other three bias correction combinations. Since accurate prediction of high and low flow rates would assist the government and water management authorities in effectively managing and utilising water resources, streamflow prediction for the future period was done using climate variables (precipitation, maximum temperature, and minimum temperature) bias corrected by DM method.

The effect of climate change and land use change on streamflow was distinguished by simulating annual streamflows in SWAT under scenarios S1, S2, S3, and S4, the findings of which are shown in Table 6. The individual and combined effect of climate and land use change on streamflow was evaluated for the period 1980–2010. It was discovered that the total variation in streamflow was 40.07 m³/s. The land use change was responsible for 31.92% of the change in streamflow, whereas climate change was responsible for 68.07%.

Table 6

Change in streamflow under different scenarios of climate and land use change

Scenario .	Time period .	LULC map used .	Mean annual streamflow (m3/s) .	Variation in streamflow (m3/s) .	Streamflow contributed by LULC change (%) .	Streamflow contributed by climate change (%) .
S1	1983–1996	1992 (L1)	37.61 (⁠⁠)	–	31.92	68.07
S2	1983–1996	2008 (L2)	43.95 (⁠⁠)	6.34 (⁠⁠)
S3	1997–2010	1992 (L1)	51.13 (⁠⁠)	13.52 (⁠⁠)
S4	1997–2010	2008 (L2)	57.82 (⁠⁠)	20.21 (⁠⁠)

Scenario .	Time period .	LULC map used .	Mean annual streamflow (m3/s) .	Variation in streamflow (m3/s) .	Streamflow contributed by LULC change (%) .	Streamflow contributed by climate change (%) .
S1	1983–1996	1992 (L1)	37.61 (⁠⁠)	–	31.92	68.07
S2	1983–1996	2008 (L2)	43.95 (⁠⁠)	6.34 (⁠⁠)
S3	1997–2010	1992 (L1)	51.13 (⁠⁠)	13.52 (⁠⁠)
S4	1997–2010	2008 (L2)	57.82 (⁠⁠)	20.21 (⁠⁠)

View Large

Despite the fact that climate change is the most significant driver to streamflow variation, the change in LULC is equally substantial, and both parameters should be considered when estimating streamflow variation for the future period. In comparison to larger catchments, the effects of land use on runoff were proportionally more pronounced in smaller catchments (Hung et al. 2020). Thus, for a smaller catchment like Meenachil, considering land use scenarios is critical for accurate hydrological prediction.

The change in average annual and seasonal rainfall and temperature (Tmax and Tmin) over the Meenachil River Basin was projected for the medium term (2025–2055) and long term (2056–2086) relative to the baseline period (1980–2010) under RCP 4.5 and 8.5 scenarios. For the baseline period, the annual average rainfall, maximum temperature, and minimum temperature in the study area were 2,850 mm, 31.70 °C, and 23.60 °C, respectively. The average annual rainfall was projected to decrease by −4.7% in the medium term and increase by +7.5% in the long term under RCP 4.5, whereas under RCP 8.5 it was projected to increase by +4.1% in the medium term and +17.2% in the long term. However, the temperature is showing an increasing trend under both the scenarios. In the medium and long term, the maximum temperature would rise by 1 °C and 1.5 °C under RCP 4.5 and 1.25 °C and 2.35 °C under RCP 8.5, respectively. Similarly, the minimum temperature would rise by 1.2 °C and 1.4 °C under RCP 4.5 and 2.1 °C and 2.7 °C under RCP 8.5. Overall, for the entire future period (2025–2086) under consideration, annual average maximum temperature rises by 1.25 °C and 1.81 °C, respectively, and annual average minimum temperature rises by 1.31 °C and 2.38 °C under the RCP 4.5 and 8.5 scenarios.

The projected variation in Pr, Tmax and Tmin during monsoon (June-November) and non-monsoon (December–May) season is shown in Table 7. With reference to the baseline period, precipitation would decrease (1.5–14.9%) during the non-monsoon season in both the medium and long term, with the decrease being greater in the medium term under both the RCP 4.5 and RCP 8.5 scenarios. Despite the fact that the projected changes in Tmax and Tmin show an increasing trend, the amount of increase for Tmin was slightly higher than for Tmax, and the increase was +3 °C under higher emission pathway in the long term.

Another maximum peak streamflow of 824.3 m³/s was also observed under RCP 8.5 in September 2085. The findings showed that the combined influence of land use change and climate change increased annual average, monthly average, and maximum peak streamflow more than the impact of climate change alone. The average annual basin streamflow rises with an increase in the coverage of impervious surfaces, which is correlated with an increase in the urban area (Sinha et al. 2020).

The mean annual streamflow rises as a result of rapid urbanisation and changes to agricultural areas. The decrease in vegetation coverage and subsequent increase in urban area over the study area had a negative impact on soil water storage capacity by significantly reducing infiltration and rising surface streamflow. Changes in landuse land cover pattern along the river's watershed region are responsible for a high degree of flow pattern deviation. The degrading state of the river and subsequent environmental and ecological stress in the watershed area can be linked to the high degree of hydrologic variation (Chellaiah & Eazon 2021). Municipal and household waste, illegal fishing, and river sand mining all have a negative impact on the environment of the Meenachil River. According to earlier research, the concentration of heavy metals like lead and iron in Meenachil is higher than the allowable limit. Variation in the hydrologic regime is also brought on by these human impacts.

The present study evaluated the impact of climate change and land use change on streamflow in the Meenachil River Basin for two future time periods: 2025–2055 (medium term) and 2056–2086 (long term) using the hydrological model SWAT. Streamflow predictions for the future period were made using projected climate model simulations as well as projected land use. The relative significance of climate change and land use change on streamflow was assessed by simulating annual streamflow for the period 1983–2010 under four different scenarios S1, S2, S3 and S4. The results of the relative significance assessment of the climate and land use parameters under various scenarios confirmed the need to take land use change into account for hydrological evaluation. The findings show that 31.92% of the variation in streamflow was influenced by land use change, while 68.07% was the result of climate change. Since most future hydrological evaluations ignored the effect of land use change, this study took into account projected land use scenarios along with projected climate while estimating streamflow for the future period.

The climate variables projected for the future period based on medium stabilization scenario (RCP 4.5) and high baseline emission scenario (RCP 8.5) were bias corrected by distribution mapping for future projection of streamflow. With reference to the baseline period 1980–2010, the annual average rainfall was showing an increasing trend (4.1–17.2%) under both the scenarios except for the medium term under RCP 4.5. For the period 2025–2086, annual average maximum temperature rises by 1.25 °C and 1.81 °C, respectively, and annual average minimum temperature rises by 1.31 °C and 2.38 °C under RCP 4.5 and 8.5 scenarios. The land change modeler module of the TerrSet software was used to predict the LULC maps for the future period. It is seen that the majority of the study area was covered with mixed vegetation and rubber plantation in 1992. The significant change in land use type from 1992 to 2060 was the increase in urban area from 3.8 to 13.6%.

The results of the study clearly demonstrate that land use change also is an important factor to be considered for prediction of future hydrological events. The effect of climate change alone, as well as the combined effect of climate and land use change, resulted in an increase in annual average streamflow in the future. Moreover, the combined effect of land use and climate change has substantially magnified the increase in streamflow in the river basin. These results may serve as a strong basis for the creation of an effective water management strategy and more efficient climatic adaptation techniques for the study area.

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

The authors declare there is no conflict.