Assessing the impacts of land use and climate change on streamflow generation in the Nowrangpur catchment based on the SWAT–land-use update tool

SWAT

Soil and Water Assessment Tool

LUT

land-use update tool

LULC

land use/land cover

CA-ANN

cellular automata–artificial neural network

GCM

general circulation model

CMIP

Coupled Model Intercomparison Project

SSPs

shared socioeconomic pathways

SRTM

Shuttle Radar Topography Mission

DEM

digital elevation model

HRU

hydrologic response unit

NSE

Nash–Sutcliffe efficiency

PBIAS

percent bias

RSR

ratio of root mean square error to standard deviation

MOLUSCE

modules for land-use change evaluation

RMSE

root mean square error

nRMSE

normalized root mean square error

Hydrologists all over the world are skeptical about the impact of anthropogenic activities on the natural water balance in several river basins. Changing climate and land use/land cover (LULC) are two prominent aspects that are responsible for the alteration of the hydrologic responses of river basins (Tamm et al. 2018). The rapid increase in greenhouse gas emissions all around the world is a major cause of rising global temperatures and disturbances in the precipitation patterns that ultimately result in natural disasters such as floods and droughts. Changes in LULC are driven by natural phenomena as well as human activities such as deforestation, plantations, and construction of roads and buildings.

The Soil and Water Assessment Tool (SWAT) is a semi-distributed, continuous, and physical hydrological model built by the United States Department of Agriculture to assess the impacts of changing land-use and management practices as well as climate on the hydrology of a river basin (Arnold et al. 2012). It has been widely accepted as a powerful hydrodynamic tool to simulate local hydrological processes and gives reliable results pertaining to water flow as well as transport of water quality loads (sediment, pesticide, and nutrient concentration) by incorporating various spatial inputs such as LULC, soil type, topography, and weather data (e.g., rainfall, temperature, wind speed, relative humidity, and radiation). Among these inputs, soil-type and topography-related inputs tend to remain unchanged over time; however, the inputs pertaining to LULC and weather data are dynamic in nature (Tram et al. 2021). Changes in the LULC significantly influence the water balance components of a catchment (e.g., evapotranspiration (ET), surface runoff, and groundwater), making it essential to accurately represent these changes while modeling streamflow generation (Tamm et al. 2018; Tan et al. 2021). To account for the dynamic character of LULC, SWAT-LUT, a stand-alone graphical interface, was developed by Moriasi et al. (2019). Tan et al. (2021) studied spatiotemporal LULC changes (using SWAT-LUT) and their impact on water balance in the Muda River Basin, situated in Malaysia. Tram et al. (2021), in their study at Dakbla watershed of Vietnam, compared the performance of the SWAT model for two cases, namely, (i) static LULC throughout the study period from 2000 to 2018, and (ii) dynamic LULC implemented via SWAT-LUT using four different maps, for 2005, 2010, 2015, and 2018. They concluded that the second case improved the calibration of the SWAT model (indicated in terms of improved R², NSE, and PBIAS) highlighting the efficacy of the SWAT-LUT. Rashid et al. (2021) used SWAT-LUT to explore the separate and combined impact of land-use change and climate change on ET, surface runoff (SURQ), and sediment yield (SYLD) in the Minjiang River Watershed, China. There are very few Indian studies that have implemented SWAT-LUT to model dynamic LULC changes in hydrological modeling. Surinaidu (2022) investigated the combined impact of dynamic LULC changes and climate change on the total water budget of Nagavali River Basin in India, using SWAT-LUT and MODFLOW.

The assessment of the dynamic LULC impact on streamflow generation for future time-periods requires projecting the LULC information into the future. Typically, the LULC transition models comprise spatiotemporal prediction models such as the Markov chain model and the cellular automata (CA) model to predict LULC for future time-periods (Kafy et al. 2021; Dehingia et al. 2022; Kamaraj & Rangarajan 2022). CA has been widely used to model non-linear and complex phenomena such as LULC changes. The CA is composed of a lattice of homogeneous cells, each of which can exist in one of a finite number of states. All cells evolve over time simultaneously by either retaining or modifying their states, based on a set of rules. The state of each cell at time t is a function of its state at time t − 1, the states of its neighboring cells, and rules that define the interaction of the neighboring cells with the cell under consideration.

Global climate change has altered the intensity and frequency of hydroclimatic variables (Piras et al. 2016). India, being one of the focal points of global warming (Krishnan et al. 2020), has witnessed heightened frequency of floods and droughts, particularly since the 1970s. The Coupled Model Intercomparison Project (CMIP) is an initiative by the World Climate Research Programme, which aims to understand and quantify past, present, and future climate changes due to variations in radiation levels that directly determine global temperatures and hydrology at every level (Eyring et al. 2016). CMIP6, the latest phase of CMIP, uses the novel shared socioeconomic pathways (SSPs) that represent the various possible magnitudes of greenhouse gas emissions and LULC changes that are expected to occur in the future (O'Neill et al. 2016). Ullah et al. (2023) reported an expected increase in future streamflow in the SWAT catchment in Pakistan using ten general circulation models (GCMs) of CMIP6 under SSP 245 and SSP 585. Ma et al. (2024) predicted increased mean annual precipitation and streamflow during 2071–2100 within the Upper Mekong River Basin in China using the SWAT model driven by the projections obtained from several CMIP5 and CMIP6 models. In the Indian context, Shrestha et al. (2020) used seven CMIP6 models and three SSP scenarios to assess drought conditions in three different regions of India. Dixit et al. (2022) investigated the possibility and magnitude of drought hazard in the Wardha River Basin, Madhya Pradesh (India), under changing climate-change scenarios. A study by Verma et al. (2023) in the Mahanadi Reservoir Project complex, Chhattisgarh, India, compared CMIP6 and CMIP5 projections and concluded that the former is superior while simulating the precipitation and temperature. Balu et al. (2023) used 13 GCMs to investigate the impact of future climate change on the hydrology of the Ponnaiyar River Basin, Tamil Nadu, under SSP 245 and SSP 585. Reddy et al. (2023) used the EC-Earth3 data to study climate-change impacts on the Godavari River Basin, and concluded that streamflow shall increase in all four SSP scenarios. Rudraswamy et al. (2023) examined the impact of climate change on hydrological processes within the Tungabhadra River Basin in India by considering 13 GCMs and four SSPs (SSP 126, SSP 245, SSP 370, and SSP 585) for the future period of 2015–2100.

Mishra & Lilhare (2016) pointed out that climate change could have significant implications for hydrological processes in river basins across the Indian subcontinent. Furthermore, the severity of the impact of climate change on water resources among various river basins could differ depending on the regional hydroclimatological conditions (Abeysingha et al. 2020). Therefore, basin-specific studies are required to be performed in order to understand and predict the repercussions of climate change on water resources within individual river basins. The Godavari River Basin, the second largest river basin in India, has been termed as extremely fragmented and is vulnerable to adverse climate-change impacts (Jain & Kumar 2014; Singh et al. 2022). Hence, it is imperative to perform a detailed analysis of LULC changes and climate change impacts on sub-catchments within the basin under the latest CMIP6 climate models and multiple SSPs. In this study, we have attempted to investigate the impacts of the changes in LULC and future climate on streamflow generation in the Nowrangpur catchment located within the Godavari River Basin in India by considering SWAT-LUT. The motivation behind the selection of the Nowrangpur catchment is the presence of the Indravati Dam located upstream of the Nowrangpur gauge, which forms a reservoir extending over an area of 110 km². This dam was constructed with the strategic aim of inter-basin transfer of water from the Godavari Basin to the Mahanadi Basin for the purpose of irrigation and electricity generation (Choudhury et al. 2012). The project is of immense value to the stakeholder states of India; hence, its optimal management and operation are crucial. Per the authors' knowledge, no prior study has been performed to investigate the impacts of the changes in LULC and future climate (under SSP 126, SSP 245, SSP 370, and SSP 585 climate-change scenarios based on CMIP6 experiments) on streamflow generation in the study area. The main objectives of this study are as follows: (i) investigating the efficacy of SWAT-LUT to model dynamic LULC changes over a static version of SWAT; (ii) developing a calibrated SWAT model for the Nowrangpur catchment considering dynamic LULC changes that have occurred in the past (using SWAT–LUT); (iii) predicting LULC maps for 2015, 2025, 2035, and 2045 using the CA-ANN model; (iv) downscaling the daily precipitation and daily maximum and minimum temperatures over the catchment corresponding to SSP 126, SSP 245, SSP 370, and SSP 585 climate-change scenarios for near-future (2020–2045), mid-future (2046–2070), and far-future (2071–2100) periods (based on ten GCMs from CMIP6 experiments); and finally (v) predicting the streamflow at the outlet of the Nowrangpur catchment under climate change and land-use change based on the calibrated SWAT model.

For this study, we utilized data from ten GCMs (downloaded from https://aims2.llnl.gov/search/cmip6/, accessed on 2 May 2023; refer to Table S1), initialized with r1i1p1f1 conditions, under the five following scenarios: Historical, SSP126, SSP245, SSP370, and SSP585. Various quantile mapping techniques have been employed to downscale climate variables from the global spatial scale to the catchment scale. These techniques include the linear scaling approach, quantile–quantile mapping, quantile delta mapping, empirical quantile mapping, and the change factor method (e.g., Piani et al. (2010), Anandhi et al. (2011), Choudhary & Dimri (2019), Rajulapati & Papalexiou (2023) and Swain et al. (2024)). In the current study, we employed the multiple change factor method to project future precipitation and temperature under different SSP scenarios. This method considered gridded historical data of variables from the India Meteorological Department over the catchment area for the baseline period (1984–2014).

The motivation behind using SWAT lies in its user-friendly interface, ease of calibration, and ability to incorporate dynamic LULC changes via SWAT–LUT. SWAT has been applied by several researchers in eastern Indian river basins, namely, the Godavari River Basin (e.g., Jothiprakash & Praveenkumar 2024; Saraf & Regulwar 2024), the Mahanadi river basin (e.g., Gunjan et al. 2023), the Subarnarekha River Basin (e.g., Kumari et al. 2024), and the Brahmani and Baitarani river basins (e.g., Swain et al. 2020). The SWAT model for the Nowrangpur catchment was set up in ArcSWAT (compatible with ArcGIS 10.3) using the Universal Transverse Mercator 44 projected coordinate system. The entire catchment was divided into three sub-basins corresponding to a default threshold area of 721.5 km² (refer to Figure S2). The Indravati Dam reservoir was added to sub-basin 1 while building the SWAT model. The reservoir became operational in the year 1999, which is evident from the increased area of the water body in the LULC map for the years 2005 and 2015 (refer to Figure 2). To achieve our first objective, we prepared two models, namely, static LULC (using the 1985 LULC map) as a baseline scenario and dynamic LULC (using the four LULC maps for 1985, 1995, 2005, and 2015). The two models were built using exactly the same data (i.e., DEM, soil, and weather inputs mentioned in Section 2.1) except for the LULC data. The aim is to isolate the impact of LULC changes alone while keeping all the other factors the same. In the former case, a total of 141 hydrologic response units (HRUs) were generated, while the latter model considered 188 HRUs. The SWAT model was run from 1982 to 2014 on a monthly scale, with the first three years as the warm-up period. After one run of the model for the baseline scenario, the next step was to apply the LUT tool for incorporating dynamic land use. The maps for the years 1985, 1995, and 2005 were obtained from Roy et al. (2015), while the 2015 LULC map was prepared by supervised classification of Landsat 8 images in Google Earth Engine.

Many long-term studies examining various watersheds often assume that land-use characteristics remain static over multiple years, which fails to capture the dynamic nature of land use in real-world scenarios. The SWAT-LUT is an innovative tool that facilitates the incorporation of dynamic land-use patterns by accepting diverse LULC maps in raster format, each linked to a specific date. We considered the yearly time-step for LULC updates. Intermediate LULC maps were created through linear interpolation. Typically, the modeling approach involves either calibration of a static SWAT model followed by integration of LUT or vice versa. In the present study, we embraced the latter approach following Moriasi et al. (2019). Comparing model outputs based on constant land use since 1985 with those considering dynamic land-use patterns revealed notable variations in sub-basin-level processes related to hydrology and sediment transport, offering a more authentic depiction of real-world scenarios.

Normalized RMSE (nRMSE) is calculated as the ratio of the RMSE to the mean of the dependent variable (Gupta et al. 2019). We conducted calibration using data from 1999 to 2013 and validation by updating parameter values from the final iteration from SWAT-CUP in the original SWAT model constructed earlier. Streamflow data from 1985 to 1993 were used for validation.

In order to evaluate the impact of future changes in LULC on streamflow generation in the catchment, we obtained the LULC maps in raster format extending up to the year 2045 by utilizing the CA-ANN model integrated into Modules for Land Use Change Evaluation (MOLUSCE) version 3.0.13, a plugin of Quantum Geographic Information System (QGIS). The model was trained based on past LULC maps, and on geophysical and sociodemographic characteristics in order to simulate future LULC maps. Five hyperparameters are required to be specified for the training process. Several combinations were tested to arrive at the final values of each. These are explained briefly as follows:

• (i) Neighborhood – The number of pixels surrounding a single pixel under consideration that play a role in changing its state. We have specified a one-pixel neighborhood.

• (ii) Learning rate – The step size in each iteration while moving toward the minimum of the loss function. Here, we used a value of 0.001 to ensure stable learning.

• (iii) Momentum – Derived from the change in parameters in the preceding iteration, momentum builds an inertia toward the minimum of the loss function and helps to get past noise in the gradients. It helps in reaching the optimal solution quickly without getting trapped in local minima. It was set to a value of 0.06.

• (iv) Hidden layers – The number of hidden layers in the neural network and the number of neurons in each layer must be specified in the form of a string input: N1 N2 N3… Nk, where Nk is the number of neurons in the kth layer and k is the number of hidden layers. We built a neural network with one layer comprising ten neurons, as further increases in number did not improve the validation results significantly.

• (v) Number of maximum iterations – This determines the stopping criteria for the training process. A value of 1,000 was given as input. Stabilization in error was observed at the end, implying that there was no need to increase this number.

The LULC maps of 1985 and 1995 (used in setting up the SWAT model) were used as inputs to the CA-ANN model along with DEM, slope, aspect map, distance to roads, and distance to built-up area maps to predict the 2005 LULC map. The ANN learning curve is shown in Figure S11. The algorithm stores the most robust neural network configuration for further use. Validation was performed using the actual 2005 LULC map. Based on the validated CA-ANN model, the geometrically consistent maps for 2015, 2025, 2035, and 2045 were prepared for further analysis.

In this study, the multiplicative and additive change factors were considered for obtaining the local-scale future projections of daily precipitation and temperature, respectively, which were then provided as inputs to the calibrated SWAT model.

We conducted a comparative analysis of two distinct uncalibrated variants of SWAT model to evaluate the impact of land-use change. The first model operated without LUT, representing static LULC, while the second model integrated LUT, representing dynamic LULC, as demonstrated by Moriasi et al. (2019). Both models were fed with identical climatic data, including precipitation and temperature series, which allowed us to isolate the effects of LULC changes on hydrological variables such as potential evapotranspiration (PET), actual ET, soil water (SW) content, percolation (PERC), SURQ, lateral flow (LATQ), baseflow (GWQ), water yield (WYLD), and SYLD. Table 1 showcases the percentage decadal changes relative to the static model across nine hydrological variables in all three sub-basins, along with the average values for the entire catchment. Notable changes in LULC include forested areas, especially in Sub-basin 1, being converted to agricultural land, alongside a slight increase in urban built-up areas, as evident from Figure 2. Minimal alterations were observed in PET, whereas there was an overall rise in ET due to changes in vegetation characteristics resulting from deforestation and increased agricultural activities. Similar findings were reported by Rashid et al. (2021) in the Minjiang River watershed in China. Furthermore, irrigation practices might have played a role in maintaining optimal soil moisture levels, ensuring a continuous water supply for crops, potentially explaining the slightly higher soil moisture content in the case of LUT-based models. In addition, tillage of the soil in agricultural land and expansion of built-up areas might have reduced the infiltration, leading to increased surface runoff (Moriasi et al. 2019). Overall, the comparison between static and LUT-based SWAT models highlighted the significant differences in sub-basin-level hydrological responses, with the latter yielding more consistent representation of real-world conditions.

In this study, we have incorporated the SWAT-LUT for developing a calibrated hydrological model for the Nowrangpur catchment. For this, ten out of 21 parameters were identified as sensitive parameters (after the first iteration of the SUFI2 algorithm) whose description and original and final values are given in Table S2. We used global sensitivity analysis to filter out the sensitive parameters by considering the t-stat and p-value. The LUT-based SWAT model was calibrated for the period of 1999–2013 and subsequently validated for the period 1985–1993. The performance measures for the calibration and validation are presented in Table S3. Per the criteria specified by Moriasi et al. (2007), the performance of the model is satisfactory. The plots for calibration and validation are shown in Figures S12 and S13. During calibration, RMSE was found to be 36.158, while the validation RMSE was 38.837. The value of nRMSE during calibration was 1.143, whereas the nRMSE for validation was 0.913.

We evaluated the efficacy of the CA-ANN model (integrated into the MOLUSCE plugin of QGIS) in the simulation of LULC maps up to the year 2045. The validation of the model, obtained by comparing simulated and reference maps for the year 2005, resulted in 89.52% correctness, kappa (histogram) of 0.947, kappa (location) of 0.876, and an overall kappa value of 0.829. The model performed satisfactorily (Okwuashi et al. 2012) and was utilized to generate future LULC maps. The observed and simulated LULC changes over the period 1985–2045 are presented in Table 2.

On comparing the area under various LULC classes in the observed and simulated maps for the year 2005, we noted that the percentage area under the dominant LULC classes, namely, forest, agriculture, and grasses/bushes, were in close agreement, contributing to a high overall kappa value. On the contrary, the area under the classes of barren land, urban, and water deviated from the observed values. However, as the extent of these LULC classes is quite small comparatively, the two maps matched well. Predictions suggested that historical trends in forest and agricultural land cover would persist in future periods. The decrease in urban area fraction from 1995 to 2005 contradicts the trend observed from 1985 to 1995, likely due to model inaccuracies. However, given that urban areas represent a small portion of regional land use, this anomaly is not projected to significantly affect model outcomes. Previous studies employing QGIS-MOLUSCE in developing towns, particularly metropolitan cities, have documented substantial changes in built-up areas (Kafy et al. 2021; Dehingia et al. 2022). Nonetheless, such dramatic shifts are not typical within natural catchments (Kamaraj & Rangarajan 2022). The notable increase in the area of water bodies by 2005 can be attributed to the operationalization of the Indravati Dam around 1999, resulting in the formation of a reservoir and subsequent expansion of water surface area. The model's inability to anticipate this sudden development based solely on provided inputs has been addressed by correcting all simulated maps to incorporate the reservoir in each raster file. Furthermore, a reservoir was integrated into the SWAT model during its setup. By including this additional component, changes in water distribution across the catchment stemming from reservoir construction and operation have been modeled. In 2005, total water bodies covered an area of 159.35 km² compared with 61.07 km² in 1995. We have attempted to account for the 98.283 km² increase through the formation of a reservoir covering an area of 110 km².

Even though the performance measures indicated satisfactory performance of the CA-ANN model, we noticed minimal changes in the future LULC maps that were produced by the model. For further analysis, we constructed separate models for various climate scenarios and GCMs, resulting in a total of 40 models. Each model was executed until the year 2045, first assuming a constant LULC pattern from 2015, and second, incorporating simulated future maps in conjunction with SWAT-LUT. Remarkably, the outcomes from both scenarios were nearly identical, indicating almost identical impacts of climate change alone versus when coupled with LULC changes. Therefore, we present the results focusing solely on climate change in the subsequent section, wherein the model was run under the assumption of constant LULC (as of 2015) until the year 2100.

Future time-series data corresponding to four SSP scenarios for precipitation and maximum and minimum daily temperatures were downscaled at a grid size of 1° latitude × 1° longitude through the CFM. We considered the ensemble of ten climate models (calculated as arithmetic mean) for the analysis to deal with the uncertainties associated with their predictions, following Li & Fang (2021), Mohseni et al. (2023), and Bhatta et al. (2019). The future trends in precipitation, temperature, and streamflow are elaborated upon in this section. The future period has been segmented into near future (2020–2045), mid future (2046–2070), and far future (2071–2100) for detailed analysis, following Li & Fang (2021). The period from 2000 to 2014 was used as a reference for climate change comparisons throughout this section.

Across all SSP scenarios, there is a notable uptrend in T_max across all months. T_max is anticipated to rise progressively over time, with mid-future values surpassing those of the near future, and far-future values surpassing those of the mid future, suggesting a consistent increase in T_max over the years. The most significant absolute increases in T_max are observed during the months of June and July across all SSPs. As anticipated, the rise in T_max is most pronounced in SSP 585, followed by SSP 370, SSP 245, and SSP 126. Historically, April and May are the hottest months in the region, and under SSP 585, they could experience an increase up to 3 °C in the average temperature in the far future. Similarly, there is a rise in T_min across all months under all SSPs. One can expect an absolute increase of up to 4.3 °C in T_max (in June) and 5.2 °C in T_min (December) in the far future relative to the baseline period. SSPs 370 and 585 indicate a higher increase in T_min compared with T_max. The most substantial increases occur during the months of November and December, while March and April exhibit the least increase. These findings suggest warmer winters for the study area.

In this study focused on the Nowrangpur catchment, we incorporated the impact of dynamic LULC changes on streamflow generation with the help of SWAT-LUT. The SWAT model showed satisfactory results during calibration and validation with NSE values exceeding 0.5 and RSR values lower than 0.7. During calibration of the SWAT model, there was a slight overestimation of peaks (PBIAS = −5.276), while during validation we noticed underestimation of peaks as indicated by the positive PBIAS value of 24.191. The nRMSE values were found to be 1.143 during calibration and 0.913 during validation. To predict future LULC changes, we used the CA-ANN model via QGIS-MOLUSCE; however, we observed the inability of the model to correctly follow the trend suggested by the input maps used in the process, for urban areas and water bodies. Furthermore, we noticed only slight changes in the future LULC maps that were produced by the model despite the good performance measures (kappa values). Next, we downscaled and bias-corrected climate data from ten CMIP6 models under the four SSPs and provided these as inputs to the calibrated SWAT model. We prepared an ensemble of all climate models for each SSP and presented the results related to the impacts of climate change on the precipitation, temperature, and streamflow in the study area.

Our findings indicating increased monsoon rainfall in the mid and far future in the study area are in accordance with Katzenberger et al. (2021), who investigated the variation of summer monsoon rainfall across all of India using 32 GCMs under CMIP6 for the scenarios SSP 126, SSP 245, SSP 370, and SSP 585. Most models predicted a substantial increase in rainfall from June to September under all SSPs. Similar findings regarding the increase in the monsoon precipitation for all climate-change scenarios corresponding to the near, mid, and far futures within the catchments of Godavari River Basin were reported by Mishra & Lilhare (2016) and Reddy et al. (2023). Das & Umamahesh (2016), in a study on the Godavari River Basin, used the second-generation Canadian Earth System Model (CanESM2) to obtain precipitation projections under the representative concentration pathways (RCPs) 2.6, 4.5, and 8.5. Their study predicted increased precipitation in the Indravati sub-basin (where the Nowrangpur catchment is located) during the far future (2070–2100) under RCP 8.5. Saraf & Regulwar (2018) noted a consistent increase in the annual precipitation within the Upper Godavari Basin, throughout the period from 2011 to 2099 per the CGCM3 and HadCM3 model projections. This is congruous with our results indicating an increasing trend in the annual precipitation series under all SSP scenarios. Mishra & Lilhare (2016) predicted that a majority of the Indian sub-continental river basins are proceeding toward warmer and wetter climate in the future, on the basis of CMIP5 model projections under RCPs 4.5 and 8.5. They reported an absolute increase in mean monsoon air temperature of approximately 3.3 °C in the far future (2070–2099) relative to a reference period of 1971–2000, in the Godavari River Basin, under RCP 8.5. Reddy et al. (2023) also reported an increase of 3.29 °C in the annual mean temperature in the Godavari River Basin in the far future under SSP 585, relative to the base period 1970–2015. Chaturvedi et al. (2012) predicted the mean (calculated by preparing an ensemble of CMIP5 models) temperature in the Godavari River Basin to increase by 4 °C (with reference to the 1880s) or more in the RCP 8.5 scenario. The difference in the reference period selected in separate studies renders it difficult to make precise comparisons; however, the increasing trend in temperature is common to all studies. Mohseni et al. (2023) utilized three bias-corrected CMIP6 GCMs (ACCESS-CM2, BCC-CSM2-MR, and CanESM5) to study the impact of climate change on precipitation, temperature, and streamflow within the Parvara Mula sub-basins of the Godavari River Basin. The results obtained from the mean ensemble of the three models predict an increase in in average annual precipitation, average annual temperature, and average annual streamflow under SSP 245, SSP 370, and SSP 585 for the near (2020–2040), mid (2041–2070), and far futures (2071–2100) relative to the baseline period (1990–2018). The increasing trend in temperature is similar to our findings. However, our study's results indicate a slight decrease in the annual streamflow for the near future followed by an increase of up to 41% in the mid- and far-future periods. It must be noted that the durations of the designated near-, mid-, and far-future periods differ in the aforementioned study. Our predictions of increased streamflow during monsoon agree with those of Mishra & Lilhare (2016) who also reported an increase in streamflow during the months from June to October in the Godavari River Basin under RCPs 4.5 and 8.5 for the near-, mid-, and far-future periods. Moreover, Reddy et al. (2023) also predicted escalated monsoon streamflow in Godavari Basin for the mid and far future. Saraf & Regulwar (2018) reported an expected surge in the mean annual flow of the upper Godavari River Basin under Scenarios A2 and A1B (considering CGCM3) and Scenarios A2 and B2 (considering HadCM3) during the years 2041–2099. Similar studies focusing on neighboring catchments in the peninsula have also identified similar trends. Anil & Raj (2024) conducted a study in the Krishna River Basin, India, to examine the impact of climate change on precipitation and streamflow using CMIP6. They concluded that the river discharge values would go up due to increased extreme precipitation, which may result in floods in the basin. Balu et al. (2023) predicted increased precipitation and streamflow in the future for the Ponnaiyar River Basin in Tamil Nadu, India.

The results of our study can aid the authorities and policymakers in efficient water resources planning and management in the study area. The inability of the CA-ANN model to predict significant changes in LULC for the future is a noticeable limitation of the current study. Future work could include the exploration of a different technique to simulate future LULC maps. Here, we used an ensemble of multiple GCMs; however, a more detailed uncertainty analysis could be performed to better understand the variability introduced by each model.

In this study, we investigated the impacts of the changes in LULC and future climate on the hydrologic response of the Nowrangpur catchment in India by incorporating dynamic LULC changes. We examined the efficacy of SWAT-LUT in producing more realistic outputs due to consideration of dynamic LULC changes. The results of the calibration and validation of the LUT-based SWAT model indicated satisfactory performance of the model. An ensemble of ten GCMs was used to study the impacts of climate change on the hydrological variables, namely, precipitation, maximum and minimum temperatures, and streamflow. Increased annual precipitation in the mid and far future is expected, the maximum increase being associated with SSP 585 toward the end of the century. The analysis of monthly precipitation indicated that, on average, the initial half of the year is projected to experience a decrease in precipitation, while subsequent months are expected to witness escalated precipitation in the future. Both maximum and minimum temperatures are predicted to increase with time. Across all SSP scenarios, there is an uptrend in maximum and minimum temperatures for all months, with an expected absolute increase of up to 4.3 °C in T_max (in June) and 5.2 °C in T_min (December) in the far future relative to the baseline period. The total annual streamflow is expected to decrease in the near future, followed by an increase (up to 41%) in the mid and far futures. The months August and September are predicted to experience the highest increase in streamflow (up to 150 cumec) under all the SSP scenarios, especially in the mid and far future. These results, predicting changes in hydrology due to climate change, serve as vital inputs for future water resources management of the Indravati Dam and conservation policies for the catchment of the dam. Additionally, the outcome from this research shall be beneficial for the design of future hydraulic projects within the study area.

The authors express their gratitude to the Indian Institute of Technology Ropar for facilitating this research. The authors are also thankful to the India Meteorological Department (IMD) and the Central Water Commission (CWC), for providing meteorological (e.g., precipitation, temperature) and hydrological (e.g., discharge at Nowrangpur) datasets, respectively.

No funds, grants, or other support were received for performing this research.

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

The authors declare there is no conflict.