Abstract
This paper focuses on exploring the potential of Climate resilient agriculture (CRA) for river basin-scale management. Our analysis is based on long-term historical and future climate and hydrological datasets within a GIS environment, focusing on the Ajoy River basin in West Bengal, Eastern India. The standardized anomaly index (SAI) and slope of the linear regression (SLR) methods were employed to analyse the spatial pattern of the climate variables (precipitation, Tmax and Tmin) and hydrological variables (actual evapotranspiration (AET), runoff (Q), vapor pressure deficit (VPD), potential evapotranspiration (PET), and climate water deficit (DEF)) using the TerraClimate dataset spanning from 1958 to 2020. Future climate trend analysis spanning 2021 to 2050 was conducted using the CMIP6 based GCMs (MIROC6 and EC-Earth3) dataset under shared socio-economic pathway (SSP2-4.5, SSP5-8.5 and historical). For spatiotemporal water storage analysis, we relied on Gravity Recovery and Climate Experiment (GRACE) from the Center for Space Research (CSR) and the Jet Propulsion Laboratory (JPL) data, covering the period from 2002 to 2021. Validation was performed using regional groundwater level data, employing various machine learning classification models. Our findings revealed a negative precipitation trend (approximately −0.04 mm/year) in the southern part, whereas the northern part exhibited a positive trend (approximately 0.10 mm/year).
HIGHLIGHTS
The slope of the linear regression method was employed for the spatial distribution of the climatic and hydrological conditions (1958–2020).
Future climate trend analysis (2021–2100) has been executed through the CMIP6 (MIROC6 and EC-Earth3) SSP245, SSP585 and historical dataset.
A novel ensemble boosting machine learning algorithm was used for the validation of groundwater level.
INTRODUCTION
Climate change is a major global concern for agricultural water management, it significantly threatening agricultural productivity (Guptha et al. 2022). The Intergovernmental Panel on Climate Change (IPCC) reported that climate change refers to long-term change in climatic patterns. These changes are usually observed in the annual or seasonal variations of the climate (IPCC 2013). Moreover, the IPCC's fifth assessment report (AR5) also identified different climate-resilient pathways, e.g., mitigation and adaptation strategies (Werners et al. 2021). These pathways aim to provide a comprehensive view of the diverse impacts of climate change on both natural and anthropogenic systems, including rural poverty, the environment, and livelihoods (Sam et al. 2020; Pushpanjali et al. 2021). It is estimated that, around half of the world's population, around 8 billion people, may experience severe water scarcity at some point during the year due to a combination of non-climatic and climatic factors (IPCC 2019). Climate change is contributing to the likelihood of severe and prolonged droughts and floods in various regions with high level of confidence (Pekel et al. 2016; Singha et al. 2022). The Fifth AR5 emphasizes that natural disasters such as floods and droughts pose a significant threat to agricultural production, highlighting climate-resilient strategies as a top priority for ensuring food security (Caretta et al. 2022). The over-exploitation of groundwater for agricultural, domestic, and industrial purposes has accelerated the depletion of water storage worldwide (Arneth et al. 2019).
In the face of these challenges, both private and public organizations recognize adaptation a crucial measure to combat with the effects of climate change on agriculture and water security-related issues (Rusia et al. 2018). Neglecting the need for measures addressing climate change's effects on the water cycle could lead to a projected decrease in global gross domestic product (GDP) by 2050, particularly in low- and middle-income countries (Acevedo et al. 2020). Thusway, climate-resilient agriculture (CRA) practices help farmers to manage natural hazards and mitigate the associated effects of climate change. This domain of the CRA approach involves developing new strategies and practices to enhance the productivity of the farm and land, reduce greenhouse gas emissions (GHGs) from agriculture, and increase farmers' incomes (Singha et al. 2020). These practices also contribute to climate adaptability, poverty reduction, and food security amid changing climate scenarios with the Sustainable Development Goals (SDGs) specially with numbers 1, 2, and 13 (Singh 2020). Recent studies showed that, climate change's impact on hydrology at the river basin scale has been assessed using various hydrological models under CMIP5 and CMIP6 scenarios within the CRA domain (Karan et al. 2022). Amiri & Gocic (2023); Milan & Amiri (2023) showed the long-term (1946–2019) water balance scenario through the various rainfall indices namely the standardized anomaly index (SAI), the standardized precipitation index (SPI), the rainfall anomaly index (RAI), the China Z index (CZI), the percent of normal precipitation (PNP), and the modified China Z index (MCZI) over Serbia. Another study of Amiri & Gocic (2021) showed the precipitation concentration degree (PCD), the precipitation concentration index (PCI), the precipitation concentration period (PCP) and the seasonal PCI (SPCI), were analyzed using long-term precipitation data (1946–2019) for climate change applicability in Serbia. Topographical variation was directly correlated with the precipitation variability. The maximum correlations are also found between the for latitude, and PCD and longitude and the summer PCI at 0.73, and 0.75, respectively. Milan & Amiri (2023) estimated the precipitation trend analysis through the polynomial fit equation in northwest Iran. The outcome showed that southwest part had incremental trend in the study period (1991–2010).
The use of geospatial technology has both potential advantages and indirect implications for the establishment of a long-term CRA framework. Various studies have used different multisensory remote sensing (RS) data sources, including the Global Land Data Assimilation System (GLDAS), TerraClimate, and Gravity Recovery and Climate Experiment (GRACE), to assess the resiliency, reliability, vulnerability, and sustainability of river basin water management (Koudahe et al. 2017; Bherea & Reddy 2022; Salehie et al. 2022; Sharma et al. 2022). Climate change, in particular, has implications for how water resources are stored and managed. These problems are solved by implementing a CRA system, but changes in the weather patterns resulting from climate change can raise questions about its operations. Despite the increasing number of studies focused on improving the resilience of the agriculture sector, many questions still need to be answered regarding the effects of climate change on this sector. Moreover, understanding the temporal and spatial distribution of rainfall is also important in assessing the risks associated with various hydrological change events (Yang et al. 2020a, 2020b).
The lack of reliable climate data in various regions globally has hindered the development and implementation of climate simulation models and statistical analyses related to water resources, hydrology, and ecology. Thus, our research aims to provide insights into CRA system and water security goals over a 90-year period, encompassing both historical (1958–2020) and future (2021–2050) periods. Therefore, this study enquiring the various hydrometeorological conditioning factors and effects of climate change on current and future performance for the suitable CRA system. Climate uncertainty provides an opportunity to develop the CRA system under this framework. This type of initiatives can be applied to small- and medium-sized enterprises to enhance entrepreneurship and livelihood strategies for owners, as well as cultivate the attitudinal traits of successful proprietors. The United Nations is implementing the 2030 framework, consisting of 17 Sustainable SDGs, which serve as an urgent call to action for all countries worldwide. This approach represents a novel contribution to the field of Eastern Indian river basin management. What sets this research apart is its ability to provide precise and comprehensive methods for identifying climate-resilient groundwater management strategies. The study's primary focus is the development of a reliable Climate Resilience Assessment approach, specifically tailored to address the challenges posed by climate change in real-world scenarios. The study aims to identify groundwater vulnerability zones, which will play a crucial role in alleviating water stress and preserving the hydrological ecosystem services and functions of the basin. Additionally, it will enhance the effectiveness of CRA management strategies in the region. This approach apart is its practicality – it not only advances theoretical understanding but also offers tangible solutions. Furthermore, it underscores ways to mitigate groundwater vulnerability, providing valuable insights for local practitioners. There have been limited studies on the changes in the hydrometeorological patterns in Ajay river basins. These studies also filing that gap to building the CRA system that influence these changes. In addition, few research on assessing the CRA system, only applied precipitation data and conventional statistical techniques, not apply the multisensor datasets for areas lacking the measured through the novel machine learning (ML) techniques. The study's findings were able to quickly do real-time assessment of direct impact of climate uncertainty on the vulnerability of agricultural productivity within the region. The novelty of this research, as well as the research gap filing it addresses, make significant contributions to the fields of groundwater conservation and climate change management. Various space-based multisensory datasets namely: TerraClimate, CMIP6, GLDAS, and GRACE, are employed for the measuring of water sustainability conditions in the Ajay River Basin region. In addition, this study envisions quantifying groundwater storage (GWS) resilience to analyze the future climate change resilience and its impact on achieving the SDGs at a river basin scale. The outcomes of this research will aid in identifying water stress zones and the necessary adaptation strategies required in any river basin.
MATERIALS AND METHODS
Study area
Data
The historical monthly climatic data collected during the study were analyzed through the TerraClimate platform (1958–2020) with a spatial resolution of 4,638.3 m. The TerraClimate dataset has been previously evaluated over surrounding regions and countries and had very good accuracy (Filgueiras et al. 2022; Araghi et al. 2023). Terrestrial water storage (TWS) data were acquired from the NASAs GRACE (2002–2021) mission with a spatial resolution of 111,320 m. The GLDAS (2003–2021) datasets were analyzed in the Google Earth Engine (GEE) cloud API. The spatial resolution of the GLDAS v2.2 datasets were 27,830 m. The historical and future climatic condition was assessed under 245, and SSP585 scenarios through CMIP6 GCMs (i.e. MIROC6 and EC-Earth3) (2021–2050) with a spatial resolution of 0.25° × 0.25°. Additionally, Our analysis encompasses a total of 90 years of model outputs, including temperature and precipitation, spanning both the historical period (1958–2020) and the future period (2021–2050) for climate change analysis.
Groundwater level data
In total, 22 district-wise pre-monsoon groundwater level (mbgl) data (2018–2019) were collected from the Central Ground Water Board (CGWB, India) (URL: http://cgwb.gov.in/GW-data-access.html). Inverse distance weighting (IDW) spatial interpolation techniques were used for the spatial distribution of the groundwater level inventory mapping within the river basin. Furthermore, this map was used for validation and sensitivity analysis in our study of CRA and water sustainability in this region.
Methodology
Statistical trend analysis and anomaly index








Climatological and hydrological trend analysis and anomaly indices are interpreted as follows: negative values indicate a declining trend, while positive values suggest an increasing trend. SAI was analyzed for the TerraClimate parameters with 10-year intervals (i.e., 1960, 1970, 1980, 1990, 2000, 2010, and 2020), while GLDAS analysis covered the years 2005, 2010, 2015, and 2020. Monthly GRACE solution data (CSR and JPL) were averaged over the study area to provide the anomaly of ETW in centimeters during the period from 2012 to 2021. GRACE-based ETW heat maps were used to visualize the regional water balance patterns. Additionally, the Pearson correlation coefficient was employed to quantify the relationships among all hydrometeorological variables. These visualizations and assessments were generated using the Matplotlib library in Anaconda Python 3 software.




Validation and sensitivity analysis
The current study employed ten ML classification models, i.e. random forest (RF), support vector machine (SVM), Naïve Bayes (NB), logistic regression (LR), neural network (NN), CatBoost (CAB), extreme gradient boosting (XGB), AdaBoost (ADB), decision tree (DT), quadratic discriminant analysis (QDA), and k-nearest neighbors (KNN), for the final GWS validation analysis.
RF
RF model used for both classification and regression problem. RF combines the training of a regression tree with the boosting technique to produce a large number of efficient tree models (Breiman 2001). Each of the training trees is analyzed using a randomly chosen set of variables by the voting classifier. The last predictor of the model is calculated by considering the results of the other trained trees.
SVM
The non-linear problems are mapped into a higher dimensional space in SVM, which makes them easier to solve with the help of the kernel functions (i.e. linear, radial, polynomial, and sigmoid) (Cortes & Vapnik, 1995). The selection of the appropriate kernel function is very important in order to develop prediction models that are derived from this framework.
NB
The NB classifier is a part of the group of posterior probabilistic classification systems that are based on Bayes' Theory (Mushtaq & Mellouk 2017). It assumes that all the variables are independent and can be studied separately. This allows the system to perform faster and simpler analysis through the ‘parent node’ and ‘child’ node system.
LR
The LR model can also be utilized to analyze the link between a given variable and several independent ones (Kim et al. 2019). Due to the nature of a binary dependent variable, it can lead to various issues, like predicted values and errors in which the ranges from 0 to 1. LR is a better alternative to performing a comprehensive analysis of such variable.
NN
Hidden layer-based NN used to take complex decision. The NN topology might be developed through a trial-and-error approach to address the hidden layer's number. The NN framework assigns varying weights to the inputs in order to perform prediction calculations (Kumar et al. 2002). An error minimization method is then used to finetune the predictions in subsequent iterations.
CAB
The CatBoost algorithm was utilized to perform with gradient-enhanced DT algorithm (Prokhorenkova et al. 2018). It will extracted variables as categorical type. The collected data is then randomly generated and presented with various random sequence of feature combinations method.
XGB
One of the most popular implementations of the gradient boosting machine (GBM) algorithm is Xgboost (Chen & Guestrin 2016). It is regarded as a superior performer in supervised learning. It can be applied for both classification and regression problems.
ADB
ADB model was implemented with the boosted decision trees that are dealing with binary classification issues (Freund & Schapire 1995). It is perfect utilized with weak learners for the classification problems.
DT
The supervised MLas DT method is commonly used to split a set of factors into multiple decision trees. Generally, the DT prediction tree structure used for classification regression and classification problem (Kadavi et al. 2019). The three nodes that make up DT are the root node, internal node, and leaf node. The outcome is then displayed as root node, internal node and leaf nodes in the dataset. The DT algorithm is utilized to divide the data into smaller classes.
QDA
The QDA is a flexible and classical classification method that allows groups to be distinguished according to covariance matrix and mean vectors (Qin 2018). In the classical QDA framework, the sample covariance matrix is high-dimensional, its singularity can occur.
K-nearest neighbor (KNN)
The kNN algorithm is widely used in supervised learning. It can predict new data points with accuracy of approximating their proximity (Zhang & Zhou 2007). The assigned values for these predictions are based on the training set's points. The KNN algorithm run with the following steps, namely first estimated the nearest neighbors of the k value, next classify datapoints through the Euclidean distance measure and then sort the training sample and finally majority of the nearest neighbors decided the final prediction.
Boruta feature selection techniques were employed to prioritize the sensitivity parameter of TWS sustainability. This analysis was carried out on the relevant dataset using the RF classifier with the BorutaPy library in Anaconda Python 3. This new approach allows us to identify the most important variables in the CRA system (FAO, 2020).
RESULTS
Trend analysis of historical climatic and hydrological parameters
Trend analysis of historical climatic and hydrological parameters: (a) precipitation, (b) Tmax, (c) Tmin, (d) AET, (e) PET,(f) VPD, (g) runoff, and (h) DEF, (i) TWS, and (j) GWS in the study area.
Trend analysis of historical climatic and hydrological parameters: (a) precipitation, (b) Tmax, (c) Tmin, (d) AET, (e) PET,(f) VPD, (g) runoff, and (h) DEF, (i) TWS, and (j) GWS in the study area.
Trend analysis of future climatic conditions
Spatial pattern maps of the future precipitation, Tmax, and Tmin trends (SLR) were generated using the CMIP6 SSP2-4.5, SSP5-8.5 and historical dataset (2021–2050) for the Ajay River Basin. All the scenarios are estimated through the two different type of general circulation models (GCMs) models i.e. MIROC6 and EC-Earth3 ((Tatebe et al. 2019; Sahoo & Govind 2023).
MIROC6 general circulation model (GCM)
Spatial pattern maps of the future: (a) Tmax, (b) Tmin, and (c) Precipitation, for the MIROC6 CMIP6 model (SSP2-4.5) in the study area (2021–2050).
Spatial pattern maps of the future: (a) Tmax, (b) Tmin, and (c) Precipitation, for the MIROC6 CMIP6 model (SSP2-4.5) in the study area (2021–2050).
Spatial pattern maps of the future: (a) Tmax, (b) Tmin, and (c) Precipitation, for the MIROC6 CMIP6 model (SSP5-8.5) in the study area (2021–2050).
Spatial pattern maps of the future: (a) Tmax, (b) Tmin, and (c) Precipitation, for the MIROC6 CMIP6 model (SSP5-8.5) in the study area (2021–2050).
Spatial pattern maps of the future: (a) Tmax, (b) Tmin, and (c) Precipitation, for the MIROC6 CMIP6 model (historical) in the study area (1958–2020).
Spatial pattern maps of the future: (a) Tmax, (b) Tmin, and (c) Precipitation, for the MIROC6 CMIP6 model (historical) in the study area (1958–2020).
EC-Earth3 general circulation model (GCM)
Spatial pattern maps of the future: (a) Tmax, (b) Tmin, and (c) Precipitation, for the EC-Earth3 CMIP6 model (SSP2-4.5) in the study area (2021–2050).
Spatial pattern maps of the future: (a) Tmax, (b) Tmin, and (c) Precipitation, for the EC-Earth3 CMIP6 model (SSP2-4.5) in the study area (2021–2050).
Spatial pattern maps of the future: (a) Tmax, (b) Tmin, and (c) Precipitation, for the EC-Earth3 CMIP6 model (SSP5-8.5) in the study area (2021–2050).
Spatial pattern maps of the future: (a) Tmax, (b) Tmin, and (c) Precipitation, for the EC-Earth3 CMIP6 model (SSP5-8.5) in the study area (2021–2050).
Spatial pattern maps of the future: (a) Tmax, (b) Tmin, and (c) Precipitation, for the EC-Earth3 CMIP6 model (historical) in the study area (1958–2020).
Spatial pattern maps of the future: (a) Tmax, (b) Tmin, and (c) Precipitation, for the EC-Earth3 CMIP6 model (historical) in the study area (1958–2020).
SAI analysis of climatic and hydrological parameters
Supplementary material, Figures S1–S10 show the spatial distribution of long-term SAI maps for the Ajay River Basin. The precipitation showed a similar pattern to SAI: high in the southern part and low in the northern part except for the years 1980, 2010, and 2020. Negative SAIs were observed for the years 1980, 2000, and 2010, while positive SAIs were noted for 1990 and 2020 (Supplementary material, Figure S1). The spatial pattern of the Tmax is very low in 1990 (−0.105 to −0.087) compared to other years (Supplementary material, Figure S2(d)). The SAI of Tmin over the study domain specified that southern part has higher Tmin value (i.e., 1960, 1970, 1980, 1990 and 2020) compared to other years (Supplementary material, Figure S3). Supplementary material, Figure S4 shows the positive SAI of AET for the years, e.g., 1990, 2000, and 2020, while that only negative SAI recorded in 1960. SAI of PET displayed a relatively consistent pattern across the region, with maximum PET identified over the southern part in 1980, 1990, 2010, and 2020, and minimum PET found in the northern part in 1960 and 1970 (Supplementary material, Figure S5). Negative runoff was observed in 1980, 2000, and 2010, while positive values were recorded for 1970, 1990, and 2020 (See Figure 6). The positive SAI of DEF value was observed in 1960 and 2010, while more negative values were found during 1990, 2000, and 2020, respectively. Negative DEF (>0.005) was recorded during 1970–1980 in the northern part of the basin, while positive DEF values (below −0.02) were seen in the southern part. This indicates temporal and spatial variation in water suitability (Supplementary material, Figure S7).
In Supplementary material, Figure S8, the majority of the basin had SAI values greater than 0.005 for VPD in 1960 and 2010, with mostly negative SAIs occurring from 1970 to 1990. In 2020, the SAI of VPD was highest in the lower part and lowest in the upper part of the basin. Supplementary material, Figure S9(a)–(f) display the SAI of GWS, with maximum values (>0.3) observed over the northern part during 2005, 2015, and 2020. Completely negative SAI of GWS was seen only in 2010 (Supplementary material, Figure S9(b)). Similarly, positive SAI of TWS was observed during 2005, with higher positive values, and during 2020, it had the highest positive value (around 1) in the southern part of the study domain (Supplementary material, Figure S10(d)). Based on SAI maps of GWS and TWS, the spatial pattern of water availability exhibited a declining trend from 2015 to 2020 in the southern part, except in 2015. In 2020, the earlier years showed very low precipitation, AET, runoff, TWS, and GWS values in the southern part, whereas Tmax, Tmin, PET, VPD, and DEF exhibited the opposite conditions (Supplementary material, Figure S11).
The annual precipitation, Tmax, and Tmin time series were analyzed using the MK trend test of the study area domain (1958–2020). It was used to identify potential trends in the data. Table 1 shows MK test statistic (S), calculated ZS values, p value and tau conditions. Based on the S score, precipitation had 47, Tmax had 229.01 and Tmin had 713.05. According to the significant analysis (p < 0.05), all the three climatic parameters are significant. The results of trend agreement with the critical Z and tau values for the precipitation, Tmax and Tmin were recognized i.e. 2.216, −9.71, 4.22 and 0.013, −0.054, 0.365, respectively.
Mann–Kendall trend test for the climatic data (1958–2020)
. | z . | S . | p . | tau . |
---|---|---|---|---|
Precipitation | 2.216 | 47 | 0.027 | 0.013 |
Tmax | −9.714 | 229.01 | 0.000 | −0.054 |
Tmin | 4.223 | 713.05 | 0.000 | 0.365 |
. | z . | S . | p . | tau . |
---|---|---|---|---|
Precipitation | 2.216 | 47 | 0.027 | 0.013 |
Tmax | −9.714 | 229.01 | 0.000 | −0.054 |
Tmin | 4.223 | 713.05 | 0.000 | 0.365 |
Temporal distribution of current water storage conditions
Monthly equivalent water thickness (EWT) variation from 2002 to 2021 obtained by GRACE datasets: (a) Center for Space Research (CSR) and (b) Jet Propulsion Laboratory (JPL) in the Ajay River Basin.
Monthly equivalent water thickness (EWT) variation from 2002 to 2021 obtained by GRACE datasets: (a) Center for Space Research (CSR) and (b) Jet Propulsion Laboratory (JPL) in the Ajay River Basin.
Correlation of hydrometeorological parameters
The study further scrutinizes the association among different climatic and hydrological input parameters using the statistical Pearson correlation matrix, OLS, and Boruta feature importance technique. The GWS and TWS have a positive correlation with AET, runoff, and precipitation as well as an anticorrelation with the DEF, PET, Tmax, Tmin, and VPD parameters, respectively (Supplementary material, Figure S13). The correlation of AET is very strong with runoff and precipitation (above 0.85). Similarly, DEF showed strong positive correlations with the VPD (0.99), PET (0.97), Tmax (0.99), and Tmin (0.99) parameters, as well as an inverse relationship with AET (−0.94), runoff (−0.93) and precipitation (−0.96). Based on OLS evaluation, the GWS, runoff, precipitation, PET, DEF, and AET are statistically significant over >99% (p < 0.001), while TWS and Tmin are statistically significant over < 95% (p < 0.05) (Table 2). The coefficient range value varies from −3.91 to 3.56 while the values of R2 (0.93), F statistic (1311), and Durbin–Watson (1.396), Prob > χ2 (0.00) indicated a good concordance among the input parameters with satisfactory results.
OLS analysis results for hydrometeorological parameters
Parameters . | β . | ρ . | Std err. . |
---|---|---|---|
VPD | −0.4421 | 0.041 | 0.216 |
Tmax | 0.541 | 0.082 | 0.311 |
GWS | 3.4739 | 0.000*** | 0.987 |
TWS | −3.1759 | 0.001** | 0.95 |
Tmin | 0.8163 | 0.017* | 0.95 |
Runoff | 3.5603 | 0.000*** | 0.95 |
Precipitation | −2.1895 | 0.000*** | 0.95 |
PET | 2.0466 | 0.000*** | 0.217 |
DEF | −3.9187 | 0.000*** | 0.217 |
AET | −2.6534 | 0.000*** | 0.217 |
Prob > χ2 = 0.000 | F = 1311 | R2 = 0.933 | Durbin–Watson = 1.396 |
Parameters . | β . | ρ . | Std err. . |
---|---|---|---|
VPD | −0.4421 | 0.041 | 0.216 |
Tmax | 0.541 | 0.082 | 0.311 |
GWS | 3.4739 | 0.000*** | 0.987 |
TWS | −3.1759 | 0.001** | 0.95 |
Tmin | 0.8163 | 0.017* | 0.95 |
Runoff | 3.5603 | 0.000*** | 0.95 |
Precipitation | −2.1895 | 0.000*** | 0.95 |
PET | 2.0466 | 0.000*** | 0.217 |
DEF | −3.9187 | 0.000*** | 0.217 |
AET | −2.6534 | 0.000*** | 0.217 |
Prob > χ2 = 0.000 | F = 1311 | R2 = 0.933 | Durbin–Watson = 1.396 |
β, coefficient; std. err., standard error; robust standard errors; F, statistical; R2, linear regression.
*p < 0.05, **p < 0.01, and ***p < 0.001.
Validation and sensitivity analysis
The average depth to ground water level for the period of pre-monsoon, 2018–2019.
The average depth to ground water level for the period of pre-monsoon, 2018–2019.
The optimal scores obtained after the validation of the ten ML models are presented in Table 3. The performance of each ML model was evaluated using various statistical metrics (Table 2). Based on the AUROC results, the ADB and RF models reached the best score (0.999), followed by the LR (0.998), NN (0.998), SVM (0.997), CAB (0.995), QDA (0.995), KNN (0.995), NB (0.994), XGB (0.991), and DT (0.980) models, respectively (Supplementary material, Figure S14(a)). ADB and XGB model are the best performers with 99% train and test accuracy compared to other ML models. The combined precision and recall values were higher than 95% for the XGB, ADB, CAB, RF, and KNN models. Based on the F1 score (around 98%), the XGB, ADB, and KNN models were the top three performers.
Performances matrices of ML models
ML classifiers . | Train accuracy . | Test accuracy . | Precision . | Recall . | F1 Score . | AUROC . |
---|---|---|---|---|---|---|
SVM | 0.966 | 0.972 | 0.990 | 0.936 | 0.963 | 0.997 |
Naïve Bayes | 0.951 | 0.944 | 0.905 | 0.955 | 0.929 | 0.994 |
Neural Net | 0.951 | 0.948 | 0.913 | 0.955 | 0.933 | 0.998 |
Logistic Regression | 0.958 | 0.951 | 0.921 | 0.955 | 0.938 | 0.998 |
QDA | 0.963 | 0.955 | 0.908 | 0.982 | 0.943 | 0.995 |
CatBoost | 0.970 | 0.976 | 0.972 | 0.964 | 0.968 | 0.995 |
Nearest Neighbors | 0.990 | 0.986 | 0.982 | 0.982 | 0.982 | 0.995 |
Decision Tree | 0.994 | 0.965 | 0.924 | 0.991 | 0.956 | 0.980 |
Random Forest | 0.996 | 0.979 | 0.981 | 0.964 | 0.972 | 0.999 |
AdaBoost | 0.999 | 0.986 | 0.973 | 0.991 | 0.982 | 0.999 |
XGBoost | 0.999 | 0.986 | 0.973 | 0.991 | 0.982 | 0.991 |
ML classifiers . | Train accuracy . | Test accuracy . | Precision . | Recall . | F1 Score . | AUROC . |
---|---|---|---|---|---|---|
SVM | 0.966 | 0.972 | 0.990 | 0.936 | 0.963 | 0.997 |
Naïve Bayes | 0.951 | 0.944 | 0.905 | 0.955 | 0.929 | 0.994 |
Neural Net | 0.951 | 0.948 | 0.913 | 0.955 | 0.933 | 0.998 |
Logistic Regression | 0.958 | 0.951 | 0.921 | 0.955 | 0.938 | 0.998 |
QDA | 0.963 | 0.955 | 0.908 | 0.982 | 0.943 | 0.995 |
CatBoost | 0.970 | 0.976 | 0.972 | 0.964 | 0.968 | 0.995 |
Nearest Neighbors | 0.990 | 0.986 | 0.982 | 0.982 | 0.982 | 0.995 |
Decision Tree | 0.994 | 0.965 | 0.924 | 0.991 | 0.956 | 0.980 |
Random Forest | 0.996 | 0.979 | 0.981 | 0.964 | 0.972 | 0.999 |
AdaBoost | 0.999 | 0.986 | 0.973 | 0.991 | 0.982 | 0.999 |
XGBoost | 0.999 | 0.986 | 0.973 | 0.991 | 0.982 | 0.991 |
According to the Boruta technique, all the hydrometeorological factors were confirmed as the most prioritized decision for the water storage analysis (Table 4). The Boruta feature importance, based on RF, was employed for parameter sensitivity analysis with 99% accuracy. The results of the Boruta method revealed that precipitation had the lowest mean importance at 5.8, while AET had the highest mean importance at 15.84. In contrast, the AET showed the highest importance for the sensitivity analysis of the groundwater suitability followed by the Tmin, VPD, GWS, DEF, TWS, Tmax, PET, precipitation, and runoff parameters, respectively (Supplementary material, Figure S14(b)).
Boruta feature importance analysis and selection of hydrometeorological parameters
Parameters . | Mean importance . | Median importance . | Minimum importance . | Maximum importance . | Decision . |
---|---|---|---|---|---|
VPD | 15.22 | 14.98 | 14.2 | 16.42 | Confirmed |
Tmax | 13.59 | 13.47 | 12.97 | 14.82 | Confirmed |
GWS | 14.48 | 14.57 | 12.72 | 16.73 | Confirmed |
TWS | 13.97 | 13.92 | 12.32 | 15.37 | Confirmed |
Tmin | 15.38 | 15.36 | 14.21 | 16.76 | Confirmed |
Runoff | 5.8 | 5.82 | 4.52 | 6.8 | Confirmed |
Precipitation | 6.93 | 7.11 | 6.27 | 7.49 | Confirmed |
PET | 10.41 | 10.36 | 9.86 | 11.1 | Confirmed |
DEF | 14.4 | 14.32 | 13.72 | 15.68 | Confirmed |
AET | 15.84 | 15.73 | 15.17 | 16.78 | Confirmed |
Parameters . | Mean importance . | Median importance . | Minimum importance . | Maximum importance . | Decision . |
---|---|---|---|---|---|
VPD | 15.22 | 14.98 | 14.2 | 16.42 | Confirmed |
Tmax | 13.59 | 13.47 | 12.97 | 14.82 | Confirmed |
GWS | 14.48 | 14.57 | 12.72 | 16.73 | Confirmed |
TWS | 13.97 | 13.92 | 12.32 | 15.37 | Confirmed |
Tmin | 15.38 | 15.36 | 14.21 | 16.76 | Confirmed |
Runoff | 5.8 | 5.82 | 4.52 | 6.8 | Confirmed |
Precipitation | 6.93 | 7.11 | 6.27 | 7.49 | Confirmed |
PET | 10.41 | 10.36 | 9.86 | 11.1 | Confirmed |
DEF | 14.4 | 14.32 | 13.72 | 15.68 | Confirmed |
AET | 15.84 | 15.73 | 15.17 | 16.78 | Confirmed |
DISCUSSION
Having a deep understanding of the various aspects of climatic variables that can affect natural resources is also important for effective environmental management. Therefore, climate change studies have a significant insights at the local level for reducing natural hazards. Salehie et al. (2022) assessed EWT for the water sustainability approach as more reliable through the GRACE and GLDAS v2.2 datasets. According to this approach, our study showed the early years has a high reliability of water resources during the pre-monsoon season. indicated less reliability in water sustainability, and vice versa. Dhar (2010) conducted a study on future (2040 and 2050) water availability and agricultural crop production in the context of climate change scenarios in the Ajay River Basin. Their research identified declining trends of precipitation, ET, soil moisture, and GWS in the southern part of West Bengal (Katwa and Gheropara). This outcome is comparable with the higher linear trend of the different climatic and hydrological variability in this region. The CMIP6 model provided quantitative insights into the potential impacts of the climate change cycle on future water storage conditions, droughts, and flood events from the upper to lower basins (Yang et al. 2020a, 2020b). In our study, the standardized TWS anomaly index helped identify drought-prone areas using hydrological datasets derived from GRACE and GLDAS. The findings indicate that the middle to upper regions of the river basin have very little water storage because of variations in precipitation and very significant groundwater level depletion as a result of human activity. Due to population pressure and extensive groundwater use for agriculture, one of the main issues is groundwater depletion. However, these datasets are valuable for implementing drought mitigation strategies in sustainable river basin planning (Deliry et al. 2022).
Based on the SAI of precipitation in the early years (2020), which showed low values, we observed a simultaneous high rate of DEF, and a declining trend in GWS and TWS in the southern most part of the area. The downstream portion of the river basin had an abundance of surface water, which resulted in significant water storage. The present work highlights that high runoff trends occur in the southern part (e.g., Gushkara, Gonna, Pratappur, Bhatkunda, Ausgram, Kurumba, Gopalpur, Bhedia, Debshala, and Basudha). These areas are primarily agrarian blocks in West Bengal and are characterized by high water consumption for domestic and agricultural purposes. Conversely, the variability and uncertainty of rainfall negatively impact irrigation scheduling as well as basin-scale groundwater discharge and recharge rates (Sahoo et al. 2021). Similarly, high DEF trends were observed in the middle part of the basin (e.g., Madhupur, Haripur, Lokepur, Sikandarpur, Amba, Charakamra, Banamalipur, Khajuri, Beniganj, Babuijore, and Chhotagholjor). Higher PET values are associated with higher temperatures in this study domain. Relative temperature leads to higher values of AET and PET across the region. Our research findings indicate a strong positive correlation between precipitation and runoff, AET, TWS, and GWS, while an inverse correlation exists with Tmax, Tmin, VPD, and DEF. A similar study on precipitation revealed significant effects at the p < 0.05 level with AET for river basin hydrology-related research (Desai et al. 2020).
Due to agricultural and hydrological droughts, the combined effects of TWS, GWS, precipitation, rising temperature, DEF, runoff, PET, and AET factors result in deteriorating climatic hydrological conditions throughout the research domain. Padhiary et al. (2019) reported that surface runoff and AET dynamics can determine the climate change adaptation and CRA status of the local livelihood condition. Furthermore, the low TWS and GWS trends are found in the northern part (i.e., Salaria, Patwabad, Rampur, Burhai, Dhamni, Kushaha, Madhupur, and Thari blocks) of the study area. Because groundwater is used for agriculture, the northern border of the region has a high groundwater stress zone due to strong trends in VPD, Tmax, and Tmin and low trends in precipitation. This region belongs to hard crystalline rocks (gneissic complex) that is abstracted negligible recharge for the GWS distribution (Banerjee et al. 2021). Chandra et al. (2019) showed that hard crystalline rocks with fracture networks are the major factors for steady sources of sustainable aquifer reserves. However, moderate TWS and GWS have occurred in the Bindapathar, Pahargara, Dimjuri, Geria, Pagla, Sima, Pindargaria, and Dalberia blocks of the basin. The majority of the basin area comprises agricultural land (57.54%) that required more irrigation and domestic purpose. Singha & Swain (2022) employed that water availability of the basin region is delineated by the land use pattern and hydrological condition that is strongly allied with climatic variability. Several studies reported that the lower part of the basin specified as more vulnerable to flood hazards that affect agricultural activity (Roy & Mistri 2016).
Moreover, this region is characterized by high TWS, GWS, and precipitation trends and moderate to low VPD, runoff, and DEF trends due to climate uncertainty. Climate change-induced severe floods and droughts have a definite impact on GWS and groundwater-dependent ecosystems (Swain et al. 2022). Our current study indicates a positive correlation between GWS, TWS, precipitation, AET, runoff, and an inverse correlation with temperature in the context of environmental and climate change studies (Sharma et al. 2022). The GWS, runoff, precipitation, PET, DEF, and AET are in good agreement of >99%; while TWS and Tmin are <95% significance level for the groundwater resource availability of the study area. The statistical significance of the study's hydrological parameters for groundwater resource availability was evaluated using ML validation analysis. Groundwater level sampling locations were divided into a 70:30% ratio for ML model validation, with associated matrix evaluations. The OLS and Boruta feature importance methods were employed to minimize errors in climate-resilient strategies for this region (Lawal et al. 2023). The Boruta feature importance analysis highlighted that AET is highly sensitive in basin hydrology compared to runoff (Desai et al. 2020).
RECOMMENDATION FOR CLIMATE-RESILIENT AND CLIMATE-ADAPTATION STRATEGY
The current study also highlights climate-resilient groundwater management strategies based on the United Nations' SDGs. We recommend that groundwater management and planning should incorporate ecosystem and biodiversity-based approaches, as well as community-driven adaptation solutions (Datta et al. 2022).
Various long-term strategies and plans are needed to sustainably replenish groundwater resources, effectively turning them into drought reserves.
It is important to consider the incorporation of climate models' projections, socio-economic factors, and hydrogeological aspects. Additionally, the use of multi-model scenarios should be prioritized instead of a single scenario or a single climate model (McDonald et al. 2022).
A range of groundwater and agroecosystem-based sustainability approaches should be emphasized. These include securing ecological veracity, land use improvement, selecting suitable groundwater recharge sites, raising public awareness, purifying saline groundwater, recycling wastewater, improving groundwater recharge, promoting rainwater harvesting, implementing irrigation, and managing organic-plastic ponds and wetlands to build CRA strategy in the study area (Rao et al. 2019).
Utilizing geospatial technology for micro-level planning, along with the creation of informative charts and maps, and engaging in cooperative community-based programs and workshops, early warning systems for weather forecasting, and initiatives like the Pradhan Mantri Krishi Sinchayee Yojana (PMKSY), Mahatma Gandhi National Rural Employment Guarantee Act (NREGA), and collaboration with NGOs should be promoted for surface and subsurface water conservation strategies (Samuel et al. 2022).
In the northern part of the basin, the increasing number of water storage structures should be considered as part of groundwater resources management. These structures can be designed to regulate water flow and facilitate irrigation during severe droughts.
Implementing climate-resilient and climate-smart agricultural practices, including crop rotation, diversification, mixed cropping, planting rescheduling, cropping pattern adjustments, irrigation scheduling, and micro-irrigation, can enhance rainfed agriculture systems for smallholder farmers.
Encouraging afforestation and the adoption of water preservation actions can help reduce the AET, PET, soil erosion, and runoff losses in the basin. These measures can also help increase soil moisture storage as well as the TWS condition. Precision agricultural practices should be properly managed to ensure agricultural production can withstand the impacts of climate change.
CONCLUSIONS
The aim of the study was to understand the long-term spatiotemporal trends in hydrology (1958–2020), climatic (1958–2050), total water availability (2002–2021), and groundwater levels (2010–2017) changes. Various multi-satellite images, including TerraClimate, CMIP6, GRACE, GLDAS, MODIS, as well as in situ data from the CGWB of India (CGWB, India), were utilized to assess GWS in the Ajay river basin, India. The spatial trends of the SAI and SLR method were applied to analyze climatic and hydrologic trends. Projections from NEX GDDP CMIP6 SSP2-4.5, SSP5-8.5 and historical (i.e., MIROC6 and EC-Earth3) were used to examine future climate trends (2021–2050). The GRACE-derived JPL and CSR and GLDAS 2 CLM data were utilized for analyzing GWS, TWS, and EWT. Statistical correlation, OLS analysis, and Boruta sensitivity analysis were conducted to prioritize and understand the relationships among the observed parameters. The study found that GRACE-derived EWT measurements in recent years have indicated a reduction in water resource reliability, suggesting a declining trend in post-monsoon periods. Linear trends in annual precipitation and Tmax values ranged from −0.04 to 0.10 mm/year and 0.005–0.007 °C, respectively. Additionally, TWS and GWS exhibited higher values in the southern part compared to the northern part. A declining trend in TWS was observed in the northern part, while a high GWS value was noted in the southern part. In 2020, early-time precipitation, AET, runoff, TWS, and GWS were concentrated at lower levels in the southern region, whereas Tmax, Tmin, PET, VPD, and DEF exhibited inverse conditions in this region. The study's findings underscore the direct impact of climate uncertainty on the vulnerability of agricultural productivity in the north-western part of the region. These findings also offer insights into various climate resilience practices that can be applied to sustainable agriculture. However, further research is needed to comprehensively assess the potential impacts of climate change on groundwater practices among smallholder farmers and households by considering cultural and behavioral aspects of the local community. Also, we implemented the meteorological forcing datasets for improvement of the CRA adoption strategy. Different stakeholders' perceptions, and community involvement are considered for the CRA. Future studies are explored to encompass the findings of the current study to a superior range of environments crossways the globe. The study's reported observed data shortage issues and inadequate field validation are its drawbacks.
ACKNOWLEDGEMENTS
The authors are grateful to the TerraClimate, United States Geological Survey (USGS), and the ESA, GEE and CGWB for providing the required hydroclimate, satellite images, and groundwater level data. We acknowledge the project ‘Integration of Digital Augmentation for sustainable Agroecosystem in Western Lateritic Zone under National Hydrology Project, West Bengal’ under which this work is mapped. The author also thanks the International Center for Agricultural Research in the Dry Areas (ICARDA) for providing necessary support for this research work. The authors are also grateful to editors and potential reviewers.
AUTHORS' CONTRIBUTIONS
S.S., C.S., and A.G. were involved in conceptualization; S.S., C.S., and A.G. prepared the methodology; S.S., C.S., and A.G. did software analysis; S.S., C.S., and A.G. validated the study; S.S., C.S., and A.G. did formal analysis; S.S., C.S., and A.G. investigated the study; S.S., C.S., and A.G. collected resources; S.S. and C.S. did data creation; S.S., C.S. A.G., and B.P. were involved in writing – original draft preparation; S.A., T.H.A., S.S., C.S., A.G., H.G.A., A.R.M.T.I., and B.P. did writing – review and editing; A.G. and H.G.A. acquired funding. All authors have read and agreed to the published version of the manuscript.
DATA AVAILABILITY STATEMENT
Data cannot be made publicly available; readers should contact the corresponding author for details.
CONFLICT OF INTEREST
The authors declare there is no conflict.