Decadal hydrological impact assessment of evolving land use and land cover in an Indian river basin: a multi-model approach

River basins worldwide undergo continuous land use and land cover (LULC) changes, significantly influencing their hydrological behavior (Tripathi et al. 2004; Chauhan et al. 2020; Roy & Chintalacheruvu 2024). These alterations, driven by urbanization, deforestation, agriculture, and climate variability, result in complex changes in the hydrological processes governing streamflow, sediment transport, and nutrient loadings (Neitsch et al. 2011; Astuti et al. 2019). Understanding the dynamics of LULC changes and their impact on hydrology remains crucial for effective watershed management, water resource planning, and environmental preservation (Ali & Chandran 2022; Vuppati et al. 2023).

Recent studies have increasingly focused on the implications of LULC changes on hydrology. Rautela et al. (2023) investigated the impact of urban expansion on hydrological processes and found significant alterations in runoff and infiltration patterns. Similarly, Sofi et al. (2024) analyzed the effects of deforestation on river basins and highlighted the increased risk of flooding and sedimentation. Moreover, Rautela et al. (2022, 2024) explored the role of agricultural intensification in modifying groundwater recharge rates, emphasizing the need for sustainable agricultural practices.

Extensive research has explored the intricate relationship between LULC changes and hydrological responses in various global river basins (Halwatura & Najim 2013; Koneti et al. 2018). Diverse methodologies, including historical data analysis, field investigations, remote sensing, and hydrological modeling, have been employed to analyze the multifaceted impacts of LULC changes (Shi et al. 2007; Chakravarti et al. 2015). These investigations revealed that hydrological components of a basin like runoff, infiltration, groundwater recharge, and base flow are significantly impacted due to changes in land use (Bronstert et al. 2002; Sajikumar & Remya 2015).

Lately, hydrological modeling has come out to be a potent tool to quantify the impacts of changes in LULC over watershed hydrology. Semi-distributed models like the Soil and Water Assessment Tool (SWAT) are widely used to simulate hydrological processes for watersheds. These models are also capable of simulating the impacts of LULC changes on a watershed's hydrology (Beasley et al. 1980; Letha et al. 2011; Luo et al. 2018; Dixit & Patil 2019). These models are widely accepted for incorporating any spatial variability of LULC and physical processes, which helps in a comprehensive evaluation of water resources. Studies in various river basins showcased that there has been a rapid increase in impermeable land surfaces due to rapid urbanization, resulting in higher runoff coefficients, the magnitude of streamflow, and significant changes in the timings of streamflow (Nie et al. 2011). Similarly, agricultural intensification and deforestation have been associated with alterations in infiltration rates, flood peaks, and groundwater recharge (DeFries & Eshleman 2004; Eshtawi et al. 2016).

Several researchers have recently tried to study the impacts of changes in LULC on the hydrology of a catchment, utilizing the SWAT model (Qiu & Wang 2014; Lin et al. 2015; Sajikumar & Sobhana, 2015). These investigations have illustrated the impact of land cover changes across various river basins and at different temporal scales. However, certain limitations exist due to data and technological constraints. For instance, some studies have focused on calibrating and validating the SWAT model based on a single land cover map, and then applying the same calibrated model to different land cover maps. To address this issue, it is crucial that calibration and validation of the SWAT model for each specific land cover map must be performed, given that basin parameters are likely to vary. Additionally, the duration of the study should be sufficiently extended and well-segmented to comprehend the change patterns at each time step. This approach is essential for a comprehensive understanding of the effects of land cover changes on hydrological processes.

The present study extends the extensive body of prior research by leveraging the computational capabilities of Google Earth Engine (GEE) and SWAT for the Cauvery River basin in India. The Cauvery River basin, like many others, faces the challenges of rapid LULC changes driven by urban expansion, agricultural intensification, and other human activities. However, this study's approach extends beyond mere examination. A two-tiered modeling strategy, utilizing SWAT models, is employed, where the first set of SWAT models (Set ‘A’) focuses on calibration and validation, establishing the model's performance under four different LULC conditions corresponding to each decade from 1980 to 2020. Subsequently, the second set of SWAT models (Set ‘B’) explores the impact of LULC changes on hydrological responses, holding climatological data constant across all models. The primary objective is to assess the historical influence of LULC changes on hydrological dynamics within the Cauvery River basin, providing valuable insights for sustainable water resource management. In summary, this study bridges the gap between past research findings and practical implications for water resource management by delving into the specific context of the Cauvery River basin. A comprehensive analysis of LULC changes and their hydrological consequences aims to enhance the understanding of the complex interactions shaping the future of river basins worldwide.

Datasets primarily required for rainfall-runoff modeling using SWAT basically comprise a set of spatial and temporal datasets. Spatial data, such as a digital elevation model (DEM), soil type information, and LULC data, are required. Weather data, which covers temperature, moisture, rainfall, wind velocity, and solar radiation, are other required inputs.

According to the elevation report for the basin (Figure 1), the research location exhibits a wide range of elevations, ranging from 0 to 2,492 m with a mean elevation of 615.42 m. This diverse topography indicates varied terrain and landscape characteristics. The DEM used for modeling was obtained from the NASA 90-meter Shuttle Radar Topographic Mission (SRTM) dataset, which is available for free download at https://srtm.csi.cgiar.org/srtmdata/.

The study area experiences four distinct periods: winter, summer, monsoon (southwest), and monsoon (northeast), each with distinct climatic traits that affect the hydrological patterns of the Cauvery basin. Statistics from the India Meteorological Department (IMD) spanning the years 1978–2020 were utilized. The analysis of the data indicated that the Cauvery basin experiences the highest rainfall in July (monthly average of 224.96 mm), while January records the lowest rainfall (monthly average of 5.57 mm) (refer to Supplementary Material S1). The mean annual rainfall for the considered period was approximately 1,221.07 mm. The basin's monthly average temperature varies from 22.85 to 28.42 °C. The average maximum temperature during the summer is 34.59 °C, making April the hottest month. The mean minimum temperature in January, the coldest month, is 16.82 °C. The Cauvery basin's average maximum temperature is 30.75 °C, while the average minimum temperature is 20.18 °C (refer to Supplementary Material S2). Table 1 shows the monthly mean data for the years 1978–2020.

Since the study required the use of multiple LULC maps, to maintain the homogeneity in data, it is important to keep the source of the data the same. LULC maps for different decades from a single source are not often available, thus we make use of GEE. GEE is a platform for geospatial analysis that brings along with it a collection of free open-source datasets, one of which is Landsat images. Landsat images from 1972 are available for preparing LULC maps (Roy & Rao 2023b, 2023c). The GEE data catalogue provides atmospherically corrected surface reflectance from Landsat 5 and Landsat 8, which can be directly utilized for the supervised classification of LULC categories. The Landsat 4 and 5 images were employed to create LULC maps for 1985, 1995, and 2005, while Landsat 8 images were used for the year 2015. The images were resampled to a resolution of 100 m for consistency in the analysis. For the LULC classification, we used the classification and regression trees (CART) supervised classification algorithm, known for its robustness and effectiveness in handling various types of data (Roy & Rao 2023a; Roy et al. 2024). Table 1 provides information on the areal extent of each LULC class in the Cauvery River basin.

The study utilized river streamflow data from the Central Water Commission (CWC) which is made available through the India-WRIS (Water Resources Information System) data portal (https://indiawris.gov.in/wris/#/RiverMonitoring, accessed 22 June 2024). Spanning from the year 1981 to 2017, the dataset was originally provided on a daily timescale and was converted to a monthly timescale to perform calibration and validation of SWAT models. Table 2 provides a detailed summary of all data and their sources, respectively.

The present research made use of SWAT, which is a semi-distributed scale model used for simulation of hydrological processes over any area. It is developed by the USDA Agricultural Research Service (Arnold et al. n.d.). SWAT provides the capability to simulate either an individual watershed or a network of hydrologically interconnected watersheds. While setting up the model, the initial step involved delineation of the watershed using the DEM. Wherein a series of operations were performed, i.e., determination of flow direction, accumulation of flow, and generation of a stream network. This is followed by the selection of the basin outlet (Bera & Maiti 2021).

A dedicated program for addressing uncertainties in SWAT model calibrations, which is known as SWAT-Calibration and Uncertainty Program (CUP) was utilized to predict the uncertainties. This free program links the developed SWAT model to various algorithms. Sequential Uncertainty Fitting Version 2 (SUFI-2) is one such algorithm, within SWAT-CUP, which effectively handles models with numerous parameters and complicated interactions between them. It achieves this by sampling from the parameter space using Latin hypercube sampling, ensuring a more uniform distribution and better coverage of the full parameter space.

Sensitivity analysis, involving the analytical evaluation of river basin parameters to aid in achieving better model accuracy (Khalid et al. 2016), is an essential aspect of model development. The sensitivity evaluation of basin parameters for the SWAT model of the Cauvery River basin employed 12 selected basin parameters (as listed in Table 2) sourced from various references. This examination was conducted using a global sensitivity analysis approach.

Calibration, often referred to as modifying model basin inputs to obtain the best possible simulation that aligns with the actual data, is a critical step. Analysis of uncertainty involves quantifying and addressing errors in model basin parameters during the calibration process. Calibration and uncertainty analysis are closely intertwined (Sharma et al. 2022). Following calibration, without altering the values of any input parameters, model validation tests the calibrated simulation against field observations (not used in calibration) and compares them with the model forecasts. The SWAT model's latest iteration, SWAT-SUFI-2, introduces various objective functions to assess the performance. Commonly used metrics include the Nash and Sutcliffe efficiency (NSE), coefficient of determination (R²), among others. NSE is a standardized metric frequently employed for assessing the ability of the model to explain the variance in measured data compared with residual variance, distinguishing between ‘information’ and ‘noise’ in hydrologic models (Guug et al. 2020). R² indicates the percentage of total deviation in the collected data, which can be accounted for by the model.

The study considered data spanning four decades and two years of warm-up period for each decade's model is considered (i.e., 1978 and 1979 for 1980–1990; 1988 and 1989 for 1990–2000; 1998 and 1999 for 2000–2010; 2008 and 2009 for 2010–2020). In a sequential set of models (referred to as the A-series) (see Table 2), respective climatic data and LULC information for each decade were employed in the SWAT simulations. Following the output generated by the SWAT model, the SUFI-2 algorithm incorporated inside SWAT-CUP was selected to calibrate and validate the results. Optimized values obtained were then integrated into the models. Furthermore, while maintaining constant climatic data (using 12 years of climate data from 2008 to 2020, where 2008 and 2009 were set as warm-up periods) and varying only the LULC (in the B-series of models) (refer to Table 2), surface runoff was observed to account for the influence of LULC changes on surface runoff.

The percentage of the watershed covered by vegetation decreased from 25.34% in 1985 to 21.32% in 2015 (Figure 5), indicating a decrease of 4.02%. Deforestation and land clearing for various purposes, such as agriculture, urbanization, and infrastructure development, may have contributed to the decline in vegetation cover.

The percentage of the watershed covered by agriculture decreased from 70.04% in 1985 to 70.3% in 2015 (Figure 5), indicating a net decrease of 0.74%. Changes in agricultural practices, such as the expansion of farmland or changes in crop types, may have influenced the percentage of watershed covered by agriculture.

The percentage of the watershed covered by barren land decreased from 1.13% in 1985 to 0.32% in 2015 (Figure 5), indicating a decrease of 0.81%. Conversion of barren land into other land uses, such as agriculture or urbanization, could explain the decrease in barren land coverage.

The percentage of the watershed covered by urban areas increased from 0.87% in 1985 to 5.54% in 2015 (Figure 5), indicating a significant expansion of urbanization. Factors such as economic development, industrialization, and improved infrastructure may have attracted people to settle in urban areas, leading to their expansion. In 1985, water bodies covered 2.62% of the watershed, which decreased to 2.52% by 2015 (Figure 5). Human activities or land use changes affecting water flow and drainage patterns might have impacted the coverage of water bodies.

Among the 12 selected input parameters (refer to Supplementary Material S4) obtained from various sources, the analysis identified the five most sensitive parameters. The sensitivity of any parameter is assessed using the t-statistic and p-value, and ranks are assigned based on the level of responsiveness. Parameters exhibiting higher absolute values of the t-statistic indicate a heightened degree of sensitivity, while p-values of zero indicate significant statistical significance. Supplementary Material S5 presents a comprehensive summary of the parameters deemed sensitive and their corresponding rankings in terms of sensitivity.

The final fitted values of basin parameters were incorporated into the A-series models, and climate data was made constant to obtain the B-series of the model, wherein any changes in simulated streamflow will be due to the changes in LULC maps across all the decades.

The average annual rainfall across the four models was recorded as 992.9 mm, indicating a consistent rainfall pattern among the models, in line with our assumption for the B-series model. However, slight variations in CNs can be observed between the models. Generally, a higher CN suggests a higher potential for runoff and reduced infiltration capacity. As we progress from B1 to B4, a discernible trend emerges with an increase in CNs. B4 exhibits the highest potential for surface runoff compared with the other models, implying a variation in land characteristics among the models. This change could be attributed to urbanization, as indicated by the corresponding shift in LULC. When determining the percentage of precipitation that changes into surface runoff, the CN2 is shown to be the most responsive metric. It has been discovered that variations in CN values are correlated with changes in land usage. Table 3 displays the mean CN for Antecedent Moisture State II (CN2) for all models. The observed decrease in vegetation cover and increase in urban areas in the Cauvery River basin from 1985 to 2015 aligns with trends reported in other regions undergoing rapid urbanization and land use changes. For instance, Rautela et al. (2023) investigated the impact of urban expansion on hydrological processes and found significant alterations in runoff and infiltration patterns, similar to our findings where urban area increased from 0.87% in 1985 to 5.54% in 2015. Sofi et al. (2024) analyzed the effects of deforestation on river basins and highlighted the increased risk of flooding and sedimentation, which parallels our observation of a 4.02% decrease in vegetation cover due to deforestation and land clearing for agriculture and infrastructure development. Moreover, studies by Halwatura & Najim (2013) and Koneti et al. (2018) have demonstrated that changes in LULC significantly affect hydrological components such as runoff, infiltration, and groundwater recharge. These studies corroborate our findings where changes in CNs and average annual surface runoff were evident due to LULC changes. The impact of agricultural intensification and urbanization on hydrological responses was also explored by Lin et al. (2015) and Qiu & Wang (2014), who used the SWAT model to illustrate the influence of land cover changes on various river basins at different temporal scales.

This comprehensive study delves into the intricate relationship between LULC changes and hydrological parameters within the Cauvery River basin in peninsular India over four decades, from 1980 to 2020. The key findings from the study can be summarized as:

• 1. Over the study period, the percentage of the watershed covered by vegetation decreased from 25.34% in 1985 to 21.32% in 2015, indicating a substantial decline of 4.02%. This decline was primarily driven by deforestation and land clearance for various purposes, including agriculture, urbanization, and infrastructure development.

• 2. Urbanization emerged as a dominant driver of change, with the percentage of the watershed covered by urban areas increasing significantly from 0.87% in 1985 to 5.54% in 2015. Economic development, industrialization, and improved infrastructure played key roles in this expansion.

• 3. The loss of forest cover contributed to a reduction in vegetation cover. The percentage of barren land also decreased from 1.13 in 1985 to 0.32% in 2015, indicating a decline of 0.81%. Conversion of barren land into other land uses, such as agriculture or urbanization, explained this decrease.

• 4. Despite urbanization trends, agriculture remained a dominant economic activity within the basin, with a net decrease in agriculture land cover of 0.74% from 1985 to 2015. This decrease of 0.74% can be accounted to be due to changes in agricultural practices, such as shifts in crop types and expansion in farmlands.

• 5. Water resources within the basin are observed to have faced increasing stress. Water bodies that used to cover 2.62% of the watershed in 1985 came down to 2.52% by 2015. This decline was attributed to human activities and LULC changes affecting water flow and drainage patterns.

• 6. The SWAT hydrological model provided valuable insights. Sensitivity analyses identified the top five sensitive parameters while calibrating and validating and confirming the model performance. Variations in CNs associated with changing LULC were observed, with CN2 demonstrating a heightened response to LULC shifts.

• 7. Surface runoff potential was found to be responsive to LULC shifts. The mean CN for Antecedent Moisture State II (CN2) varied among the models, indicating changes in surface runoff potential. This suggests that LULC changes significantly influence surface runoff behavior.

• 1. These findings offer a robust foundation for evidence-based decision-making, watershed management, and planning for sustainable development within the Cauvery River basin.

• 2. Furthermore, these insights serve as valuable references for similar studies in other regions, aiding in the nuanced understanding of complex hydrological interactions in diverse landscapes, and providing critical data for informed policy and planning endeavors.

All authors contributed to the study's conception and design. S.D. and D.S. performed material preparation, data collection, and analysis. The first draft of the manuscript was written by S.D. and K.K.P. reviewed previous versions of the manuscript and made necessary corrections. All authors have read and approved the final manuscript.

The authors have no relevant financial or non-financial interests to disclose.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.