Identifying the driver(s) of a process or phenomenon is central to understanding and predicting its future state. In complex hydrometeorological systems, a process can have multiple drivers dynamically coupled to the system across timescales. Thus, a robust method to identify drivers is imperative. In hydrological sciences, methods like multivariate regression and, more recently, Big Data machine-learning approaches rely on finding a co-relation between variables, rather than identifying cause-effect relations. This study evaluates cause-effect discovery (Causal Discovery or CD) algorithms in hydrometeorological systems. Although earlier studies have made important contributions to exploring CD methods, they have primarily focused on bivariate methods in simple synthetic environments. Specifically, we evaluate the following four theoretically distinct multivariate CD algorithms, (i) TCDF, (ii) VARLiNGAM, (iii) PCMCI+, and (iv) DYNOTEARS. We evaluate these algorithms within a large, complex simulated environment of the Global Land Data Assimilation System (GLDAS) where the drivers, reference truth, are known perfectly. We evaluate the drivers identified by CD methods against this reference truth and also contrast its results with the widely used method of co-relation identification, Pearson’s Correlation Coefficient (PCC). The results show that CD methods identify fewer false drivers compared to PCC, across a range of Köppen-Geiger climate types. For example, PCC failed to distinguish true drivers from instantaneous and lagged cross-correlations, typically present in hydrometeorological systems. Whereas, CD methods eliminate a higher number of false instantaneous and lagged drivers. Thus, though PCC identifies the highest number of true drivers, it suffers from high false drivers. Overall, CD methods perform similar to or better than PCC, while PCMCI+ and DYNOTEARS performed the best. Further, we test whether time-series prediction models perform better when predictors are limited to those identified as causal by CD methods. Evaluation of surface soil moisture predictions during drought shows that CD-based models outperform PCC-based models and are more parsimonious. Thus, we demonstrate the effectiveness of using causal discovery to eliminate spurious relations and obtain a robust set of drivers for prediction and process understanding across different climate conditions. This study overviews, demonstrates and tests efficacy of CD methods in studying cause-effect relations in hydrometeorological systems. By exposing their capabilities and differences in a simulated environment, we hope to encourage their use in the real world and move beyond co-relation.

Estimating water availability in terms of runoff volume over time and space is crucial for identifying suitable locations for micro-scale rainwater harvesting (RWH) structures such as loose stone check dams and gabions, which are constructed along stream networks. Traditional hydrological assessments often rely on empirical methods, such as the Soil Conservation Service-Curve Number (SCS-CN) method, which have limitations in representing spatial variability in streamflow availability along drainage networks across different time scales. While rainfall-runoff models can model streamflow at variable time scales, they provide streamflow values only at the outlet of the catchment. To address these gaps under the conditions of limited gauge availability, this study proposes a lumped-spatial hybrid framework that combines a rainfall-runoff model with a GISbased flow accumulation scheme to estimate spatiotemporally varying streamflow along the drainage network of small-sized catchments. Two rainfall-runoff models, GR4J and TOPMODEL, were selected for streamflow modeling. The combined approach was applied to an experimental catchment of 1150 km2 in the Ovens River in southeast Australia, covering the period from 1999 to 2020. The streamflow models calibrated for the catchment were integrated with a flow accumulation scheme in GIS to develop a grid-based stream discharge accumulation (GSDA) map at a 10 m resolution. The spatiotemporal GSDA maps were analyzed at annual, monthly, and seasonal timescales. Monthly GSDA maps were validated against upstream gauging station data from Bright, Harrislane, and Harrietville, resulting in NSE values of 0.75, 0.75, and 0.72 for GR4J and 0.62, 0.6, and 0.45 for TOPMODEL, respectively. The annual GSDA highlights wet and dry years, while monthly and seasonal maps capture flow peaks, low-flow periods, and temporal runoff trends across the catchment. Unlike the SCS-CN method, which provides simplified runoff estimates, the GSDA framework offers temporally dynamic and spatially explicit insight into streamflow availability relevant for RWH planning. By estimating available water volumes at candidate stream locations and accounting for temporal variability in harvestable runoff, the proposed approach enhances the reliability of RWH site selection and structure design based on water availability.

Irrigation is a major human intervention in the global land-atmosphere system. However, increasing climate variability and associated regional water scarcity may lead to abrupt reductions in irrigation. In this study, we use the Community Earth System Model to investigate the global near-surface air temperature response to irrigation cessation over Northwest India. Our results indicate that evaporative cooling is the dominant local mechanism regulating temperature in response to irrigation in Northwest India. Following irrigation cessation, average near-surface temperatures over Northwest India increased by similar to 0.24 degrees-0.38 degrees C during November-February. Notably, irrigation cessation over Northwest India also induces temperature changes in remote regions. These remote responses are primarily associated with the advection of warm, moist air, enhanced atmospheric humidity, and increased downward longwave radiation. Our findings highlight that the temperature impacts of irrigation practices can extend beyond local regions and should be considered in assessments of climate and water management.

Machine learning (ML) based approaches for rainwater harvesting (RWH) site suitability mapping often face limited data for training ML models. This study investigates whether the knowledge learned from data-rich catchments can be effectively transferred to the hydrologically and geomorphically similar and distinct basins. The transfer-learning (TL) framework is tested across three Indian catchments of different climatic conditions in the states of - Odisha, Maharashtra, and Tamil Nadu-using seven key influential predictors derived from LiDAR DEMs, geological and soil maps, and GR4J-simulated annual streamflow. A total of 18 experimental cases are designed, viz., intra, direct transfer, adaptation-based transfer, and multi-catchment combinations. Feature similarity between the source and target catchments is quantified using Kullback-Leibler (KL) divergence to systematically assess conditions for transfer feasibility. Four ML models-Support Vector Machine (SVM), Random Forest (RF), XGBoost (XGB), and KNN (K-Nearest Neighbour)-are trained over 100 randomized iterations using curated suitable and unsuitable RWH locations. Results show that performance of direct transfer declines sharply when dominant catchment features exhibit high divergence. RF and XGB are more resilient to cross-domain variability, while SVM and KNN are highly sensitive to feature mismatches. Incorporating a 20% target-domain sample substantially improves kappa and F1 scores by 25-40% in Odisha, 20-30% in Maharashtra, and 35-45% in Tamil Nadu, respectively, and stabilizes feature importance between the source and target domains. Multi-source training, which aggregates data from all three catchments, achieves the highest accuracy (0.95) and kappa (0.85). KL divergence analysis confirms that transferability depends on geomorphic similarity; that is, lower divergence enables effective transfer, whereas higher divergence reduces generalization. Overall, this study demonstrates that TL is feasible and beneficial for RWH site mapping, provides a practical and transferable tool for sustainable RWH planning in data-limited regions and also builds foundation for developing wide area scalable models by continuously updating the weights of influential thematic layers as new training data becomes available.

Evaporation from on-farm storages represents a potentially significant yet poorly quantified component of water loss in agricultural systems. The performance of commonly used open-water evaporation models, many of which were developed for large lakes and reservoirs, on small (many <100 ha) and highly managed storages remains poorly understood due to limited observational data for validation. To address this gap, this study collates in situ water-level records from thirteen monitored storages spanning a major agricultural catchment in Australia and evaluates nine evaporation models, including Penman-based variants, area-dependent mass transfer models, and a remote-sensing-based hybrid model. The ensemble mean and ensemble median derived from the individual models are also included to assess whether ensemble estimates provide more accurate and robust evaporation estimates than individual model outputs. Water-level-derived evaporation is used as an observational benchmark across these storages, while dynamic water surface area is estimated from Sentinel-2 imagery to support model evaluation. Results show substantial differences among individual models. The ensemble approaches generally ranked among the leading methods across the monitored storages and assessment metrics (Pearson correlation, bias, and RMSE). Sensitivity test using alternative benchmark-screening configurations produced similar model rankings. Despite the influence of non-evaporative signals in water-level observations such as seepage, these data offer a valuable and previously unavailable constraint for evaluating evaporation at the storage scale. These findings establish a practical and transferable framework for quantifying evaporation depth from small water bodies. The resulting depth estimates can subsequently be combined with independently derived water-surface areas when volumetric evaporation losses are required for storage-scale or regional water accounting.

In light of the rapid advancements in hydrological science research facilitated by cutting-edge remote sensing technologies, such as synthetic aperture radar (SAR), hyperspectral imaging, and Light Detection and Ranging (LiDAR), we have curated a special issue in Remote Sensing of Environment entitled “Emerging remote sensing techniques for hydrological applications”, spanning from October 2022 to April 2024. This special issue comprises 31 publications that highlight methodologies leveraging multi-sensor satellite platforms, unmanned aerial vehicles (UAVs), and advanced physical models and machine learning approaches to improve the monitoring and modeling of key hydrological flux and state variables. These remote sensing retrievals (e.g., river discharge and soil moisture) have been applied to various operational hydrological applications such as real-time flood monitoring and drought risk assessment. To provide a systematic overview, we categorize these publications based upon hydrological themes and the number of publications, covering topics such as water body, soil moisture, river discharge, water level, drought, water storage, and other related areas. Finally, we provide an outlook that envisages how the emerging trends (e.g., multi-sensor integration and machine learning-driven approaches) identified from the published studies will evolve and shape future research directions in hydrological remote sensing.

Field-scale soil moisture (SM) retrieval using Synthetic Aperture Radar (SAR) data under heterogeneous vegetation conditions remains challenging due to complex SAR backscatter responses influenced by vegetation, surface roughness, and soil texture. This necessitates site-specific calibration of retrieval models, which limits its application to broader spatial extent. This study examines whether the parameter transferability of the WCM model extends to different agricultural crops—corn, soybean, wheat and canola—using SMAPVEX12 L-band UAVSAR data. Application of WCM parameters trained for a specific crop to other types of crop resulted in increased SM retrieval error (ubRMSE) by 26–55% for $VV$-pol and 37–63% for $VH$-pol. This deterioration is due to scattering from various plant structures, plant heights, and Vegetation Water Content (VWC) across the crops. Also, a multi-crop model calibration provided moderate performance but was less accurate than crop-specific calibrations. A detailed analysis with corn indicates that within-crop transferability is achievable when a model calibrated in advanced growth stages is applied for SM retrieval in earlier growth stages. These findings help define the conditions under which WCM parameters can be transferred across crops and provide insights for operational SAR-based soil moisture retrieval.

Synthetic Aperture Radar (SAR) offers high-resolution surface soil moisture in all-weather conditions, allowing continuous Earth observation. However, a significant challenge in soil moisture retrieval using SAR data exists in the calibration of retrieval parameters applicable to a broader range of environments and canopy densities. The site-specific calibration methods used in existing SAR soil moisture retrieval algorithms limit their operational applicability. In this study, we evaluated the transferability of soil moisture retrieval parameters across various sensing configurations and canopy densities to reduce the need for site-specific calibration and dependence on extensive ground data. We utilised quad-pol L-band imagery collected by UAVSAR during the SMAP Validation Experiment 2012 (SMAPVEX12), focusing on wheat fields. The backscatter from the vegetation-soil system is modelled using the Water Cloud Model (WCM). This approach resulted in accuracy (RMSE = 0.07 m3m-3, correlation coefficient = 0.7) similar to retrievals using ground-based vegetation parameters. The potential transferability of WCM parameters across fields, canopy densities, polarisations and incidence angles was examined using the posterior distribution of WCM parameters obtained from Markov Chain Monte Carlo (MCMC) calibration and extensive cross-validation. Both the posterior parameter distributions and cross-validation results suggest that WCM parameters are not transferable across different incidence angles nor between co-pol and cross-pol images. However, they are transferable across a wide range of canopy densities, between VV & HH polarisations, and across different locations when UAVSAR data with similar incidence angles are used as input.

Hydrological signatures are statistical metrics useful to quantify and infer behaviours of hydrological processes, but there has been limited use of signatures for non-streamflow variables, such as actual evapotranspiration (AET). AET signatures can assist in tasks such as evaluating remotely sensed products, diagnosing deficiencies in hydrological models, and improving understanding of hydrological processes, such as the role of AET in driving hydrological drought. This study focuses on first of these three applications. To achieve this, the study proposes eight AET signatures defined at temporal scales from daily to annual. Two remotely sensed AET (AETRS) products are assessed against flux tower AET (AETFluxtower) at seventeen FluxNET sites in Australia. The two AETRS products are Moderate Resolution Imaging Spectroradiometer (MODIS,16A2GFv06.1), and CSIRO MODIS Reflectance-based Scaling Evapotranspiration (CMRSET). Annually, median AETRS closely matches AETFluxtower, except in less-arid regions. However, signatures reveal AETRS largely underestimates the variability of flux tower data at both annual and monthly scales. Other monthly indices are better matched, such as indices of water stress and AET asynchronicity with potential evapotranspiration. However, some metrics are better matched in one product than the other, such as the strength and timing of seasonal fluctuations, with MODIS exhibiting a phase shift. Overall, the signatures reveal that regionally-developed CMRSET outperformed globally-developed MOD16A2GFv061. This study, the first to systematically define AET signatures, offers a way of assessing various aspects of AET dynamics across temporal scales. Furthermore, the case study highlights specific deficiencies in AETRS and may assist in selecting appropriate AETRS, including for modelling studies.

Large-scale perturbations in land surface characteristics have been found to induce disturbances in the overlying atmosphere via land-atmosphere coupling. The perturbations can lead to changes in hydroclimatic variables, such as precipitation and air temperature, or in atmospheric circulation patterns. However, the local and remote atmospheric responses to continental-scale changes in land surface water have not been well studied in Australia. In this study, using the Community Earth System Model 2 (CESM2) of the National Center for Atmospheric Research (NCAR), we investigate the changes in Australian monsoon, which primarily impacts the northern Australian climate, in response to an extreme surface condition: the whole Australia being treated as a shallow lake in model simulations. The simulation results show that a continental-scale lake would extend the Australian monsoon season via earlier onset and later end. We find that the most significant changes in the simulated precipitation occur during the pre-monsoon period (e.g., early October to mid November). Considering that the traditional scheme used to explain monsoonal rainfall by the theory of land-sea thermal contrasts is not consistent with the simulated precipitation patterns, this study analyzes the changes in moist static energy (MSE) budget, the simulation with a hypothetical lake features an atmospheric condition that favors the formation of precipitation: increased moisture convergence and dry static energy divergence, which might be associated with the increased net energetic forcing and export of MSE. We also confirm the dominant role of atmospheric circulation in determining the variability of precipitation over northern Australia in wet season via examining the regional moisture recycling. A relative impact computation upon components in the moisture budget shows that the dynamic component of the vertical advection of moisture contributes the most to the temporal evolution of precipitation over northern Australia in wet season.

Benchmarking farm-level irrigation water productivity (WPI) and water productivity (WP) can assist in understanding the irrigation effectiveness of individual farms and in developing strategies to improve their irrigation management. This study introduces a method to integrate on-farm irrigation measurements, remotely sensed yields and publicly available rainfall data for multi-year farm-level WPI and WP benchmarking. The method was tested over cotton farms located in south-eastern Australia during the 2011-19 cropping seasons. We trained remote sensing (RS)-based machine learning (ML) models - Random Forest Regression (RFR), Gradient Boosting Regression (GBR) and Support Vector Regression (SVR) - to predict yields for over 400 cotton fields with groundtruth yield data. Predicted cotton yields from the best-performing model were then combined with irrigation and rainfall data for WPI and WP benchmarking. We also examined: 1) if the yield model is transferable to unseen years and 2) if sub-field-scale yield data from a harvester over a small number of fields are effective for training ML models, in case field-scale yield data are insufficient. The results showed that field-scale cotton yield could be predicted with the best accuracy using the GBR model (R2 = 0.7, RMSE = 235 kg/ha, mean absolute error = 176 kg/ha and Pearson correlation = 0.84) when applied to the period of training. The average WPI and WP varied between 0.18-0.36 kg/m3 and 0.16-0.23 kg/m3, respectively. However, the RS-based yield model showed reduced performance outside of the training period. In addition, when field-scale yield samples were used in combination with many sub-field-scale samples for calibration, the model performance was biased to favour the sub-field-scale samples. Our findings demonstrate the ability of RS and ML models to provide yields for benchmarking analysis but highlight the potential risk of reduced accuracy of yield prediction in future years.

Accelerating groundwater depletion, driven by climate change and growing groundwater extraction for irrigation, has increased the need for accurate monitoring of this indispensable resource. Traditional methods, such as in-situ water table observations and pumping tests, have proven valuable for continued monitoring of groundwater availability and aquifer characteristics but are limited in assessing groundwater variations at a larger basin scale. In contrast, the Gravity Recovery and Climate Experiment (GRACE) offers a method to estimate basin-scale groundwater changes, although its estimates encompass not only groundwater in the aquifer but also surface water (e.g., lakes, rivers) and soil moisture in the vadose zone. To delineate groundwater variations accurately from GRACE observations, additional data sources are necessary. In this study, we use the European Space Agency’s Climate Change Initiative for Soil Moisture (ESA CCI SM) in the surface layer (top 0-2cm), which is extrapolated to the profile moisture content for the entire root zone (0-120cm). Utilizing the estimated profile soil moisture, we derive groundwater variations in the southern Victoria region of Australia by subtracting the ESA CCI SM derived soil moisture component from GRACE observations. The estimated groundwater variations agree well with the groundwater mass changes estimated from in-situ observations. This study presents an approach that integrates GRACE observations with profile soil moisture estimates derived from the ESA CCI SM product to assess groundwater variations. The validation against in-situ data indicates that satellite observations of soil moisture and gravity changes can provide robust estimation of basin-scale variations in both profile soil moisture and groundwater.

Drought-induced hydrological shifts and subsequent non-recovery have been reported globally, including in Australia. These phenomena involve changes in the rainfall-runoff relationship, so a year of given rainfall gives less streamflow than before. Some authors have indicated that vegetation dynamics played a key role in hydrological shifts during Australia's Millennium Drought (MD, 1997-2009), but such interactions are complex and are yet to be fully examined. This study investigates vegetation responses before, during, and after the MD for the same set of catchments in southeast Australia where hydrological shifts and non-recovery have been reported. The characterisation of vegetation behaviour relies on remotely sensed vegetation indices (VIs), namely Normalised Difference Vegetation Index (NDVI), Fraction of Photosynthetically Active Radiation (FPAR), Enhanced Vegetation Index (EVI), and Vegetation Optical Depth (VOD). Despite the severe multi-year drought, in most locations, the results indicate increased or maintained VIs over the entire period spanning pre-drought to post- drought. However, the link with hydrological shifts is nuanced and depends on how data are analysed. Contrary to expectations, an initial analysis (focussing on raw VI values) indicated that VI shifts were not correlated with hydrological shifts. It was only when the data were reanalysed to better account for the meteorological conditions that the expected correlations emerged. Overall, the results suggest that vegetation was able to maintain indices such as greenness and, by extension, actual evapotranspiration, leaving less rainfall for streamflow. More broadly, this approach provides greater insights into how vegetation affects hydrological behaviour through matched catchments during this and other multi-year droughts.

Soil moisture retrieval using the semi-empirical Water Cloud Model (WCM) is a widely used approach for Synthetic Aperture Radar (SAR) data due to its simplicity. However, applying the model to large regions poses a challenge due to the need for site-specific calibration, which requires a large number of ground samples. In this study, a Bayesian Markov Chain Monte Carlo (MCMC) optimization scheme was employed for parameter estimation, enabling the generation of optimal parameter distributions for different crops under diverse surface conditions. The analysis is based on L-band UAVSAR data acquired during the SMAPVEX12 campaign and includes crop types such as soybean, canola, corn, winter wheat, wheat, and oats. The results highlight that the performance of the calibrated model varies across crop types, and for some crops, the choice of objective function also influences the performance. Two objective functions tested are the Sum of Squared Residuals (SSR) and the Kling-Gupta Efficiency (KGE). This suggests that WCM is potentially deficient in representing backscattering by vegetation and rough soil surfaces. The results show that calibrating WCM against SSR leads to smaller root mean squared error (RMSE), while KGE-based calibration results in a greater correlation coefficient (R) for soil moisture retrievals, particularly for corn and canola. The difference in RMSE and R is small but statistically significant, implying consistent trade-offs between RMSE and R.

The increasing frequency of extreme weather events caused by climate change has highlighted the importance of systematic reservoir management to ensure a stable supply of agricultural water. Dredging, in particular, is one of the key management strategies to prevent reservoir functionality degradation due to sediment accumulation. However, dredging projects are often carried out relying heavily on the subjective judgment and experience of field personnel due to constraints in time and resources. This study proposes a method for objectively prioritizing dredging sites using Sentinel-1 Synthetic Aperture Radar (SAR) imagery. The gradual changes in reservoir capacity, presumably caused by sedimentation or erosion, were estimated by analyzing the slopes of the satellite-derived surface area to field-observed surface area ratio for 10 reservoirs from 2015 to 2023. The results indicated that Dongbu (-2.094e-05 day(-1)), Yedang (-1.263e-05 day(-1)), and Seongju (-8.051e-06 day(-1)) reservoirs experienced the most significant declines in surface area ratio over time, making them the highest-priority sites for dredging. The use of satellite imagery for dredging site selection is expected to support rational decision-making and improve the efficiency of reservoir management.

Digital agricultural services (DAS) rely on timely and accurate spatial information of agricultural fields. Initiatives, including deep learning (DL), have been used to extract accurate spatial information using remote sensing images. However, DL approaches require a large amount of fully segmented and labelled field boundary data for training that is not readily available. Obtaining high-quality training data is often costly and timeconsuming. To address this challenge, we develop a multi-scale, multi-task DL-based novel architecture with two modules, an edge enhancement block (EEB) and a spatial attention block (SAB), using partial training data (i.e., weak supervision). This architecture is capable of delineating narrow and weak boundaries of agricultural fields. The model simultaneously learns three tasks: boundary prediction, extent prediction and distance estimation, and enhances the generalisability of the network. The EEB module extracts semantic edge features at multiple levels. The SAB module integrates the features from the encoder and decoder to improve the geometric accuracy of field boundary delineation. We conduct an experiment in Ille-et-Vilaine, France, using time-series monthly composite images from Sentinel-2 to capture key phenological stages of crops. The segmentation output from different months is combined and post-processed to generate individual field instances using hierarchical watershed segmentation. The performance of our method is superior in both pixel-based (86.42% Matthew's correlation coefficient (MCC)) and object-based accuracy measures (76% shape similarity and 60% intersection over union (IoU)) to existing multi-task models. The ablation study shows that the EEB and SAB modules enhance the efficiency of feature extraction relevant to field extent and boundaries and improve accuracy. We conclude that the developed model and method can be used to improve the extraction of agricultural fields under weak supervision and different settings (satellite sensors and agricultural landscape).

Floods are a leading cause of disaster-related fatalities and economic losses. There is a growing use of remotely sensed information on flood extents to evaluate flooding conditions and to assist in future event predictions. However, remotely sensed products often lack the necessary spatial or temporal resolution needed for disaster mitigation. Recent research has attempted to solve this issue by utilizing high-spatial-resolution optical satellite images collected at sparse time intervals (e.g., Landsat 8/9) to enhance coarse-spatial-resolution images (e.g., Moderate Resolution Imaging Spectroradiometer or MODIS), taking advantage of the latter's high temporal resolution. Nevertheless, the usability of optical images is limited by clouds and aerosols, resulting in missing or poor-quality pixels. This study addresses this issue by developing a feature extraction, mixing and matching (FEMM) methodology to infill missing inundation information in optical satellite derived flood inundation maps. In the FEMM approach, dominant spatial features of flood extents are extracted from long-term simulation results from hydrological and simple flood inundation models. Missing pixel values for a satellite image are then inferred from the non-missing parts of the image based on the dominant spatial features. The Empirical Orthogonal Functions (EOF) technique is used in both the feature extraction and flood extent construction. To evaluate the efficacy of the FEMM methodology, we use synthetic cloud masks of varying patterns and coverage sampled from real images with clouds. The synthetic cloud masks are applied to a cloud free image to produce degraded images, which are then infilled and compared with the original image. A combination of an initial terrain-based local infilling and the FEMM is shown to be highly effective in most missing data scenarios, achieving critical success indices >80%. An exception is when an extensively large patch (for example half of the image) is completely missing from the image at either the upstream or downstream end of the floodplain. The approach has the capacity to enhance remote sensing images by infilling incomplete inundation maps based on hydrological spatial patterns and topographic information, thereby improving flood monitoring for emergency services.

Groundwater depletion, driven by climate change and increasing extraction for irrigation, has increased the need for accurate monitoring. Traditional methods, such as in situ water table observations and pumping tests, are valuable for assessing groundwater availability and aquifer characteristics but are limited in capturing basin-scale variations. The Gravity Recovery and Climate Experiment (GRACE) enables estimation of basin-scale groundwater changes, though its observations also include surface water and soil moisture (SM) in the vadose zone. Therefore, additional data on non-groundwater components are needed to isolate groundwater variations. In this study, we use the profile SM content for the top 0-120 cm of soil as an estimate of vadose zone SM, derived using an exponential filtering technique applied to European Space Agency's Climate Change Initiative for Soil Moisture (ESA CCI SM) and in situ data. This approach addresses limitations of conventional models, such as their inability to represent non-natural or lateral water redistribution. Groundwater storage (GWS) changes in southern Victoria, Australia were estimated by subtracting the filtered SM from GRACE data and validated against in situ groundwater level observations for both unconfined and confined aquifers. The ESA CCI SM-based estimates showed clear improvements in capturing seasonal and interannual variability of in situ GWS compared to conventional model-based estimates. The proposed approach is potentially applicable to GWS estimation at continental scales.