Groundwater under the lens: a remote sensing exploration of storage patterns and trends in Bangladesh Open Access

Groundwater (GW) is an indispensable resource that regulates residential, commercial, and agricultural activities and provides water to billions of people (Siebert et al. 2010). It is also necessary for maintaining ecosystems, assisting with climate adaptation, and ensuring global water and food security because of its consistent and crucial role in supplying fresh water (Taylor et al. 2013). In addition, GW plays a crucial role in global consumption because it is easy to access, economical, and less vulnerable to contamination than surface water (Vohra et al. 2023). However, GW depletion is a global issue driven by increased agricultural requirements, climate change, industrialization, urbanization, and unsustainable use of GW (Grogan 2016). Furthermore, the hydrological cycle is expected to be altered by rising global temperatures, resulting in lower GW recharge and more substantial evaporation (Stagl et al. 2014). In this respect, GW levels in Bangladesh are declining as a result of intensive irrigation practices, and this issue has become severe because of the exploitation of GW for agricultural purposes, particularly in the northwest (Hasanuzzaman et al. 2017). In addition, reduced rainfall, frequent droughts, and unsustainable irrigation with GW are among the factors contributing to Bangladesh's GW depletion, which leads to water scarcity for both agricultural and drinking purposes (Rahman et al. 2018). Bangladesh's irrigated agriculture is mostly dependent on GW. According to reports from 2019, 79% of Bangladesh's irrigated areas receive irrigation water from GW (Mainuddin et al. 2020). However, excessive extraction of GW has resulted in significant decreases in GW levels, with some regions experiencing annual decreases of up to 6.6 cm (Islam et al. 2021). Consequently, overextraction and alterations driven by climate change pose increasing stresses on Bangladesh's GW resources, emphasizing the importance of monitoring and understanding the spatiotemporal variations in groundwater storage (GWS) to address the risks of water scarcity and long-term sustainability.

A comprehensive understanding of GWS change is crucial for sustainable water management, although significant knowledge gaps persist. Real-time monitoring of GWS is complicated because of varying depths in sedimentary basins, affecting the relationship between storage changes and GW levels. In addition, the subsurface position of GW makes monitoring difficult, triggering sophisticated techniques such as satellite remote sensing and improved water-cycle monitoring to assess storage dynamics and variations accurately (Xu et al. 2024). Furthermore, conventional techniques for monitoring GW, such as well-based measurements, are frequently expensive and do not provide thorough spatiotemporal coverage. Few studies have used sophisticated statistical and machine-learning techniques to identify patterns, anomalies, and long-term trends in GWS, although many studies have investigated the prediction of GW levels and water quality (Ali et al. 2024). In particular, comprehensive studies employing satellite-derived GW data for systematic analysis, including long-term trend analysis, are lacking in Bangladesh. To address this gap, GW pattern and trend analysis identification can be enhanced by combining geographical and spatiotemporal factors with a log-additive neural model (Pagendam et al. 2023). In addition, merging satellite data with ground-based observations can increase model dependability and provide a more thorough understanding of GW dynamics (Chakraborty 2019). Bangladesh could improve the management of its water resources and make policy decisions that support the sustainable use of GW by applying these approaches (Qureshi et al. 2015; Zzaman et al. 2022).

Remote sensing has become an innovative instrument in GW research because it provides continuous spatial and temporal coverage and may exceed the limits of ground-based observations (Ibrahim et al. 2024). Remote sensing data, such as GLDAS, is essential for understanding GW dynamics (Ouma et al. 2015; Shen et al. 2022), addressing data scarcity (Maksud et al. 2023), and offering a cost-effective alternative to traditional methods (Fahim et al. 2024), particularly in data-limited regions like Bangladesh. The constraints of conventional ground-based approaches are overcome by this integration, which enables continuous monitoring and study of GWS patterns. GW patterns and anomalies can be identified via machine-learning algorithms to examine vast amounts of geographical data collected via remote sensing technologies (Ibrahim et al. 2024; Saha & Chandra Pal 2024). In addition, GW management strategies could be successful because of methods such as hybrid deep-learning models, which have demonstrated excellent accuracy in identifying GW recharge zones (Al-Ruzouq et al. 2024). The utilization of satellite data may provide insightful information on the dynamics of GW, which is a novel approach in Bangladesh, where comprehensive GWS studies have been limited.

This study evaluates Bangladesh's GWS dynamics via data from NASA's Global Land Data Assimilation System (GLDAS 2.2) from 2004 to 2023 to address these knowledge gaps. The research adopts a combined approach, integrating geospatial analysis and trend analysis to accomplish two primary goals: (i) determining distinct spatial patterns in GWS via the K-means clustering algorithm and (ii) analyzing the long-term patterns in GWS using the modified Mann‒Kendall test.

To accomplish these primary objectives, we implemented the kriging method for the purpose of spatial interpolation after extracting data via Google Earth Engine (GEE). Kriging, a geostatistical method, was employed for interpolation, ensuring accurate spatial representation of GW levels (Hasan et al. 2021). The clustering that identified homogeneous zones of GWS was conducted via the K-means clustering algorithm, which efficiently revealed patterns in water distribution. In addition, significant trends and monotonic fluctuations in GWS throughout the study period were found via the modified Mann‒Kendall test, a trustworthy non-parametric technique (Cui et al. 2018). ArcGIS software conducted postprocessing and visualization of the maps and graphical outputs.

This study provides an innovative and comprehensive way to study GWS dynamics by postprocessing raw satellite data to provide a unique, long-term analysis of GWS changes. The findings of this research not only improve the understanding of GWS in Bangladesh but also provide significant insights for policymakers to help sustainable GW management in Bangladesh by highlighting important patterns and trends.

The initial phase of the data acquisition methodology involved importing GWS (GWS_tavg) data derived from the GLDAS 2.2 into GEE.

Subsequently, Bangladesh's geographical limits and the temporal span from 2004 to 2023 were established. GEE aggregates the daily GWS_tavg data to compute annual averages for each year encompassed within this timeframe. Sophisticated interpolation techniques utilized a 5,000 m radius to address the NoData values. The data points, which included latitude, longitude, and GWS_tavg value coordinates, were then sampled at a 25,000 m scale and subsequently exported to Google Drive as Comma-Separated Values (CSV) files.

The interpolation methodologies chosen for the study were ordinary kriging, universal kriging, and inverse distance weighting (IDW). These methodologies were evaluated via leave-one-out cross-validation (LOOCV). A range of models, including linear, spherical, exponential, and Gaussian, were examined for ordinary kriging, whereas universal kriging incorporated both linear and quadratic drift components. IDW was assessed utilizing powers of 1, 2, and 3. The efficacy of each methodology was appraised through the application of the root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R²) as evaluative metrics.

Given a dataset of n data points, with input features X = {X₁, X₂, … ,X_n} and corresponding target values Y = {Y₁, Y₂, … ,Y_n}, the LOOCV process involves the following:

For each data point i in the dataset:

• • Leave the ith point out as a test sample:

• • o Training set: X_train = X\X_i, Y_train = Y\Y_i

  • o Test set: X_train = X_i, Y_train = Y_i

In Equations (1)–(4), n is the number of observations, Y_i is the actual value, is the predicted value, and Y′ is the mean of the actual values.

GWS data for Bangladesh were processed in CSV format, which included latitude, longitude, and GWS values, to carry out the interpolation. The shapefile of the Bangladesh boundary was used to create a grid, and GWS values were estimated over the grid via the selected interpolation technique. The interpolated raster layers were clipped via the Bangladesh boundary shapefile after being stored as temporary Geographic Tagged Image File Format (GeoTIFF) files. The clipped rasters (GeoTIFF) with a cell size of 0.023378507 and 0.023378507 were utilized for K-means clustering and the modified Mann–Kendall test.

In Equation (12), Φ(Z) is the cumulative distribution function of the standard normal distribution.

In Equation (13), x_j and x_i are the corresponding time points, m is the estimated slope (Sen's slope), and y_j and y_i are the time-series values at periods j and i, respectively.

The Z-statistic in Equation (11) and the P-value in Equation (12) were used to determine the trend's statistical significance. They focused on areas where the average P-value was less than 0.05, and the average Z-value was less than −1.96, highlighting key relevant areas.

Table 1 summarizes the interpolation performance metrics for IDW (IDW with powers of 1, 2, and 3), universal kriging (regional linear and regional quadratic models), and ordinary kriging (linear, spherical, exponential, and Gaussian models). The RMSE, MAE, MAPE, and R² were used to evaluate each model.

The accuracy and predictive power of interpolation techniques were evaluated. Ordinary kriging (exponential) had the lowest MAE (25.91) and RMSE (39.84) with a high R² (0.9372), indicating superior accuracy. A high R² reflects a strong correlation, while low RMSE and MAE suggest minimal prediction errors. In contrast, ordinary kriging (linear) and all IDW models (powers 1, 2, and 3) had negative or near-zero R², showing poor predictive ability. Universal kriging (regional linear and quadratic) also performed worse than the exponential model. Thus, ordinary kriging (exponential) emerged as the most accurate interpolation technique in this study, aligning with Asadi & Adhikari (2022), who found that the exponential ordinary kriging reduces GW prediction errors more effectively than IDW, and Shahmohammadi-Kalalagh & Taran (2021), who identified kriging as an effective method for GW estimation.

Table 2 shows different statistics of different GWS clusters in Bangladesh. Cluster 0 (42.81%) is primarily located in western Bangladesh, with some extension into the northeast, and has the lowest mean GWS of 783.49 mm. Cluster 1 (47.16%) spans a significant portion of the northern and eastern regions, extending partially into the central area, with a mean storage of 894.94 mm. Cluster 2 (4.76%) is situated in the southeastern (SE) region and has the highest mean storage of 1,514.90 mm. Cluster 3 (5.27%) is also concentrated in the southeast, with a mean storage of 1,250.12 mm.

Several studies have documented regional differences in GWS throughout Bangladesh, consistent with our findings. Between 2003 and 2013, Khaki et al. (2018) reported that GWS decreased by 32%, especially in areas such as Clusters 0 and 1, where we also reported reduced storage levels. Our results of reduced GWS in Cluster 0, especially in the northwest, are also consistent with those of Kamal et al. (2022), who reported that this region has a low GW level. Furthermore, Shamsudduha et al. (2012) and Sarkar et al. (2022) reported that GW levels varied, with the former decreasing in northwest Bangladesh (corresponding to Clusters 0 and 1) and increasing in southern coastal areas of Bangladesh, which reflects greater storage in Clusters 2 and 3. The significant storage levels that we found in Cluster 3 are further supported by Sarkar et al.’s (2022) findings of greater GW potential in southern and river-adjacent regions.

The highest mean slope is observed in the southwestern region with −4.18 mm/year, followed by the central region (C) with −3.94 mm/year, the northeastern region (NE) with −3.65 mm/year, and the SE region with −3.12 mm/year, and the lowest mean slope is found in the northwestern region with −2.84 mm/year. These findings indicate a substantial and consistent depletion of GWS across all regions, with the southwest experiencing the most significant decline. While the NE region shows an increase in GWS of up to 0.28 mm/yr, this finding is considered insignificant in this study.

The Mann–Kendall test reveals pronounced declines in the southwest, consistent with Janardhanan et al. (2023), who found the southern region experiencing the highest decline rate, while the northern part saw less decline. Purdy et al. (2019) reported a 0.88 cm/year decrease in northwest Bangladesh (2002–2016), which is relevant to this study. Mridha & Rahman (2021) found a 13.5% annual decline in Bogura, while Ouyang et al. (2024) observed severe GW losses in Mymensingh, Rangpur, and Rajshahi, confirming nationwide depletion. In contrast, the marginally positive trends in the northeast align with Roy et al. (2024) and Dey et al. (2023), who identified GW potential in the Bengal Basin.

These significant declines indicate severe GW depletion, likely due to excessive extraction and minimal recharge (Qureshi et al. 2015), with Monir et al. (2024) emphasizing overextraction as the primary cause. Jameel et al. (2023) observed a ∼50% shift in recharge from precipitation to stagnant water bodies. Moreover, Abdissa & Chuko (2024) noted that climate-change-induced rainfall shifts and rising temperatures have worsened GW depletion by reducing recharge. These findings highlight the need for region-specific GW management strategies, urging policymakers to enforce controlled withdrawals, recharge well construction, and alternative water use, i.e., rainwater harvesting, to enhance water availability and limit GW abstraction (Akter et al. 2024) for further depletion prevention.

This study highlights significant GW depletion in southwestern and western Bangladesh due to excessive extraction and limited recharge, worsened by climate change. In contrast, northeastern regions show stable or slightly increasing levels. There is an urgent need for targeted GW management plans to reduce overextraction and enhance recharge potential. Integrating climate resilience initiatives and sustainable farming practices is essential for protecting water resources. Key actions should include setting withdrawal limits and promoting alternative water uses. Future research should explore the links between agriculture, land use, and GW storage, using data-driven methods for better predictions. Incorporating machine-learning techniques and addressing GW quality would further improve sustainable management in Bangladesh.

The authors extend their appreciation to Taif University, Saudi Arabia, for supporting this work through project number TU-DSPP-2024-07.

Conceptualization: M.R., S.N., S.I., T.R.P., M.A.I., N.I.T.; methodology: M.R., S.N., S.I., T.R.P., M.A.I., N.I.T.; experiment conduct: M.R., S.N., S.I., T.R.P., M.A.I., N.I.T.; data curation: M.R., A.G., A.H.; formal analysis: M.R., S.N., S.I., A.G., A.H.; software: M.R., S.N., S.I., A.G., A.H.; writing – original draft preparation: M.R., S.N., S.I., T.R.P., M.A.I., N.I.T., A.G., A.H.; writing – review and editing: A.G., A.H.; supervision: A.G., A.H.; funding acquisition: A.G., A.H.; project administrator: A.G., A.H.; resource: A.G., A.H. All authors are agreed to submit the article in the journal.

This research was financially supported by Sher-e-Bangla Agricultural University, Sher-e-Bangla Nagar, Dhaka, Bangladesh. The study was funded by Taif University, Saudi Arabia (Project No. TU-DSPP-2024-07).

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

The authors declare there is no conflict.