Investigation of Water Storage Dynamics and Delayed Hydrological Responses Using GRACE, GLDAS, ERA5-Land and Meteorological Data in the Kızılırmak River Basin

Abstract

Monitoring groundwater dynamics and basin-scale water budget closure is critical for sustainable water resource management, especially in regions facing climate stress and overexploitation. This study examines the temporal variability of total water storage and groundwater trends in Türkiye’s Kızılırmak River Basin by integrating GRACE/GRACE-FO satellite gravimetry, GLDAS-Noah land surface model outputs, ERA5-Land reanalysis products, and local meteorological observations. Groundwater storage anomalies (GWSAs) were derived from the difference between GRACE-based total water storage anomalies (TWSAs) and GLDAS-modeled surface storage components, revealing a long-term groundwater depletion trend of −9.55 ± 2.6 cm between 2002 and 2024. To investigate the hydrological drivers of these changes, lagged correlation analyses were performed between GRACE TWSA and ERA5-Land variables (precipitation, evapotranspiration, runoff, soil moisture, and temperature), showing time-shifted responses from −3 to +3 months. The strongest correlations were found with soil moisture (CC = 0.82 at lag −1), temperature (CC = −0.70 at lag −3), and runoff (CC = 0.71 at lag 0). A moderate correlation between GRACE TWSA and ERA5-based water storage closure (CC = 0.54) indicates partial alignment. These findings underscore the value of satellite gravimetry in tracking subsurface water changes and support its role in basin-scale hydrological assessments.

1. Introduction

Water resources are vital to sustaining life and maintaining ecological balance. Throughout human history, access to water has played a defining role in the rise and fall of civilizations. While approximately 2.5% of the world’s water is freshwater, much of it is stored in glaciers and aquifers and thus remains inaccessible. Only about 0.3% of global freshwater is readily available in rivers, lakes, and reservoirs [1].

Groundwater is a critical resource that supports ecosystems, feeds surface water bodies, and sustains human populations. However, global groundwater reserves have been increasingly threatened by climate change and excessive, unregulated exploitation, leading to groundwater depletion [2,3]. Although considered renewable, groundwater can lose its sustainability due to overuse and prolonged drought. Therefore, accurately calculating and continuously monitoring groundwater budgets is crucial for developing and maintaining groundwater and drought early warning systems based on long-term terrestrial water storage anomalies derived from GRACE and model datasets. This study focuses on long-term groundwater dynamics and water storage changes, excluding short-term hydrological events such as torrential rainfall or flash floods. However, several structural challenges continue to hinder sustainable water management. These include the lack of regulation for illegal groundwater extraction, weaknesses in policy implementation, insufficient monitoring infrastructure, and the high costs associated with collecting and accessing hydrological and climatic data [4,5,6,7,8,9].

Over the past two decades, advancements in satellite-based gravity missions and remote sensing technologies have revolutionized hydrological monitoring and prediction [10]. The Gravity Recovery and Climate Experiment (GRACE) and its successor, GRACE Follow-On (GRACE-FO), have provided groundbreaking observations of global mass variations. As detailed by Landerer et al. [11], these missions have significantly enhanced our understanding of large-scale changes in polar ice, soil moisture, surface and groundwater storage, and ocean mass distribution. GRACE-based products are essential for monitoring deviations in total water storage anomalies (TWSAs) and groundwater storage anomalies (GWSAs), offering a unique opportunity to track significant water changes on a broad scale. Meanwhile, hydrological models and meteorological data help decompose these satellite-based observations into their individual components. Incorporating local meteorological data also improves the regional accuracy of water budget analyses.

GRACE and GRACE-FO have been widely used to study changes in polar ice [12,13], soil moisture [14,15], surface and groundwater storage [7,16,17,18,19,20,21,22], and ocean mass distribution [23,24]. In addition, GRACE-based regional-scale TWSA values are widely used in many hydrological applications. GRACE products provide information on unusual climate events, droughts [25,26,27,28,29,30,31], floods [9,32], snow accumulation [12,33,34], and hydrological changes (water resource consumption, evapotranspiration or runoff) [6,35,36,37,38,39] across local and global scales.

Gravity variations resulting from mass redistribution can be conceptualized as shifts in a thin layer of water near the Earth’s surface. These variations typically originate from changes in total water storage (TWS), mass exchange between the atmosphere, land, and ocean, or the melting of ice sheets and glaciers. GRACE/GRACE-FO missions detect these changes in Earth’s gravity field, which can be translated into TWSA by estimating the vertically integrated mass variations [40]. The GRACE/GRACE-FO mission provides satellite-derived total water storage anomaly from GRACE, observed via satellite gravimetry, representing the sum of surface water, soil moisture, groundwater, snow, vegetation water content, and other components.

Conversely, Global Land Data Assimilation System (GLDAS) provides a model-based terrestrial water storage anomaly, derived from simulated soil moisture, snow water equivalent, and plant canopy surface water), enabling finer-scale interpretation of water budget components. To analyze each water storage component individually, complementary datasets such as the (GLDAS) [41] and European ReAnalysis ERA5-Land [42] are commonly used to provide variables like soil moisture, snow cover, and evapotranspiration.

GRACE/GRACE-FO data have been extensively utilized in numerous studies to analyze water storage dynamics across various spatial scales, ranging from small regional basins to major global watersheds. Rzepecka and Birylo (2020) [40] investigated groundwater storage (GWS) changes derived from GRACE and GLDAS in the Vistula (194,500 km²) and Odra (118,900 km²) river basins in Poland. Their study showed strong agreement between GRACE observations and well-based groundwater level measurements, emphasizing the importance of optimally integrating GRACE and GLDAS data for accurate monitoring. Similarly, Wang et al. [43] assessed groundwater sustainability in the arid Hexi Corridor (China) by validating GRACE-derived data with in situ measurements and analyzing the combined impact of climate change and human activity. The GWSA results from GRACE-GLDAS, and water table fluctuation (WTF) methods were consistent between 2002–2010. Nikraftar et al. [44] examined 21 years of GRACE data (2002–2022) for the Middle East and reported significant groundwater depletion across the region, with most basins deemed unsustainable. The study also identified human activity as a key contributor to this decline. Similarly, Zhang et al. (2025) [45], investigated the Shule River Basin in northwestern China using GRACE and GLDAS data from 2003 to 2023, revealing a persistent groundwater decline at an average rate of −0.31 cm yr⁻¹. Their results indicated that after 2016, anthropogenic factors—such as irrigation expansion, urbanization, and economic growth—surpassed climatic influences and became the dominant drivers of groundwater depletion.

Over the past two decades, extensive research has utilized GRACE and GRACE-FO observations to investigate total water storage anomalies (TWSA) and groundwater storage anomalies (GWSA) at global and regional scales. International studies have revealed severe groundwater depletion in semi-arid and irrigated basins [5,7,18,20,22,29,40,44]. In Türkiye, GRACE-based analyses have also been increasingly applied to quantify hydrological variability [35,46,47,48]. Despite these advances, Türkiye-focused studies are not comprehensive and often lack integrated assessments that combine GRACE/GRACE-FO, GLDAS, ERA5-Land, and local meteorological observations for basin-scale water budget evaluation.

Changes in the time-varying gravity field are assumed to represent the motion of water masses over land, but this assumption should be validated with ground-based observations whenever possible [37,38]. Policymakers and stakeholders need systematic, multidisciplinary approaches to accurately determine water budgets and ensure sustainable management. Although many studies have used GRACE/GRACE-FO, GLDAS, and ERA5 data, this study focuses on the Kızılırmak River Basin, combining satellite, model, and ground-based meteorological observations. This multi-source framework enables basin-scale assessment of water budget components and site-specific hydrological evaluation. Using this integrated approach, the spatiotemporal dynamics of water in the Kızılırmak Basin were evaluated more comprehensively.

Various GLDAS land surface models available during the study period were compared to assessing their consistency with GRACE-derived TWSA. Among them, GLDAS-Noah v2.1 yielded the highest correlation (CC = 0.73) and was selected for further analysis. Table 1 summarizes the model performances.

Table 1. Correlation between GRACE/GRACE-FO TWSA and GLDAS Model Outputs.

GLDAS Land Surface Model	Correlation with GRACE/GRACE-FO
GLDAS-Noah	0.73
GLDAS-VIC	0.38
GLDAS-CLSM	0.64

These findings are consistent with previous studies highlighting the agreement between GLDAS-Noah and GRACE-based TWSA, particularly in semi-arid regions [38]. ERA5-Land model was also selected due to its high spatial/temporal resolution and its improvement over earlier versions like ERA-Interim.

Türkiye, a geo-strategic country in the Eastern Mediterranean, is highly vulnerable to the impacts of climate change due to its geography and water policies [49]. Additionally, Türkiye is located in the Mediterranean basin, a critical region that hosts a large portion of the world’s water-stressed population and includes countries that utilize more than half of the world’s renewable water resources. According to the UN-Water Report, water scarcity in Türkiye, excluding the Black Sea Region, will reach an insurmountable level within the next 30 years [50,51]. National Aeronautics and Space Administration (NASA) reported that the Eastern Mediterranean region, where Türkiye is located, was experiencing the driest period in the last 900 years [52]. According to the Intergovernmental Panel on Climate Change Sixth Assessment Report (IPCC–AR6) [53] the Mediterranean region, including Türkiye, is projected to experience a mean surface temperature increase of approximately 1–3 °C by 2100, with hotspot areas reaching 4–5 °C under the RCP4.5 and SSP5-8.5 scenarios. Consistent with these projections, Gümüş et al. (2023) [54] reported that Türkiye could face a temperature increase of up to 7.5 °C and a precipitation decrease of up to 20%, particularly across the Aegean and Mediterranean regions, confirming the intensification of regional water scarcity risks. According to NASA’s drought map for Türkiye (2021), available water storage in the region will decrease significantly, and a water shortage will occur in Türkiye by 2050 [55].

Despite being in a strategic location affected by the global climate crisis, about 75% of Türkiye ‘s water is used for agricultural irrigation. Therefore, monitoring groundwater and water resources, which constitute a large part of the hydrological cycle, is of vital importance for Türkiye. The Kızılırmak River basin is Türkiye ‘s second-largest basin and has the longest river. Agricultural land occupies 56% of its surface area, making it one of the most vulnerable river basins to drought. Current irrigation practices in the basin primarily rely on unsustainable groundwater. As a result, the Kızılırmak River basin, which is a critical basin to be analyzed in terms of water budget, was selected as the study area.

Recent studies in the Kızılırmak River Basin were initially limited to the assessment of meteorological or hydrological trends and drought conditions [56,57,58]. However, in recent years, GRACE-based studies have also been introduced. Deliry [46] developed a GIS-based water budget framework using GLDAS-Noah outputs to map total water storage variations across the basin. Khorrami et al. [47] downscaled GRACE TWSA estimates to a 1-km spatial resolution, revealing sub-basin hydrological changes, while Khorrami and Gündüz [48] combined GRACE/GRACE-FO and FLDAS datasets through a machine learning model to evaluate drought intensity and water storage dynamics at high spatial detail.

Nevertheless, these previous efforts were generally focused on short-term or component-specific analyses. In this study, GRACE/GRACE-FO, GLDAS-Noah, and ERA5-Land datasets were jointly employed to provide a comprehensive long-term (2002–2024) water budget assessment of the Kızılırmak River Basin. This research is expected to pioneer basin-scale water budget monitoring studies to be conducted across Türkiye. Accordingly, a sustainable monitoring methodology is proposed to identify water-related threats in climate-vulnerable strategic regions and to develop appropriate preventive measures.

The objectives of this research are threefold: (i) to quantify the magnitude and statistical significance of groundwater storage depletion derived from the integration of GRACE/GRACE-FO and GLDAS datasets, (ii) to identify the temporal lag relationships among precipitation, evapotranspiration, soil moisture, and runoff that explain the phase shifts in total water storage dynamics, and (iii) to assess how the integration of GRACE, GLDAS, ERA5-Land, and ground-based meteorological data improves basin-scale water budget closure and the diagnostic interpretation of hydroclimatic processes.

The novelty of this study lies in being the first long-term, time-lagged water budget analysis conducted at the basin scale in Türkiye. Furthermore, by integrating GRACE/GRACE-FO, GLDAS, ERA5-Land, and in situ meteorological datasets, the study performs a temporal analysis of hydrological interactions and an uncertainty assessment, thereby enhancing the accuracy of groundwater monitoring and the effectiveness of regional water management strategies.

2. Materials and Methods

2.1. Study Area

The Kızılırmak River Basin is the second largest basin in Türkiye, located between latitudes 37°58′–41°44′ N and longitudes 32°48′–38°22′ E, as shown in Figure 1. The total area of the basin is approximately 82.182 km², covering around 11% of Türkiye’s surface area [59]. The Kızılırmak is also the longest river entirely within Türkiye, with a length of 1.151 km. With an annual runoff volume of 6.48 billion m³, the basin accounts for approximately 3.5% of Türkiye ‘s total water potential [60].

Figure 1. (a) Location of the Kızılırmak River basin within Türkiye, (b) Distribution of the meteorological stations in the basin on the topographic map.

Climatically, the basin features both maritime and continental influences. The northern part, draining into the Black Sea, is characterized by a maritime climate with annual average temperatures ranging from 13 °C to 15 °C and precipitation levels between 1.000–1.200 mm, mostly concentrated in winter. In contrast, the central Anatolian section exhibits a continental climate, with average annual temperatures between 9 °C and 12 °C and lower annual precipitation levels, ranging from 300–600 mm. Winter temperatures in these areas often fall below zero [56,57,58].

The basin’s economy largely depends on agriculture and livestock production, with approximately 56% of its land used for agricultural purposes. Furthermore, the basin includes 57 dams, 275 ponds, 1.154 irrigation facilities, and 37 hydroelectric power plants [59]. Despite existing regulations aimed at managing groundwater resources, the number of unauthorized and unmonitored wells has been increasing. This uncontrolled extraction poses a serious threat to the sustainable management and planning of groundwater resources in the region.

2.2. Data Acquisition and Processing

In this study, each dataset serves a distinct role within the integrated water budget framework. The GRACE/GRACE-FO mission provides satellite-derived TWSA representing the combined effects of surface and subsurface mass variations detected through satellite gravimetry. The GLDAS-Noah model supplies land surface components—including soil moisture (SM), snow water equivalent (SWE), and plant canopy surface water (PCSW)—which are subtracted from GRACE TWSA to estimate groundwater storage anomalies (GWSA). The ERA5-Land reanalysis dataset provides precipitation (P), evapotranspiration (ET), and runoff (RO) variables, from which monthly anomalies relative to the 2004–2009 mean period (the GRACE baseline) were computed to obtain precipitation, evapotranspiration, and runoff anomalies (PA, ETA, ROA). These anomalies were subsequently used to calculate the theoretical water storage change anomaly (WSCA = PA − ETA − ROA) for the basin’s water budget closure analysis. Finally, local meteorological station data (temperature and precipitation) were employed to validate the reanalysis outputs and assess the temporal consistency between modeled and observed hydroclimatic responses across the basin.

2.2.1. GRACE/GRACE-FO Data

The GRACE satellite mission, launched on 17 March 2002, is a joint project by NASA and the German Aerospace Center (DLR), and it has significantly advanced our understanding of Earth’s temporal gravity field variations [61]. To continue the success of its predecessor, NASA and the German Research Centre for Geosciences (GFZ) launched the GRACE-FO mission in May 2018.

The GRACE/GRACE-FO missions offer a unique opportunity to study the spatio-temporal dynamics of TWSA at both global and regional scales [62,63,64]. These missions have greatly enhanced our understanding of large-scale water and mass changes across the Earth system [37,61,65,66,67,68,69,70].

The 11-month hiatus between the GRACE and GRACE-FO missions and several short (1–2 month) data gaps are known limitations of satellite gravimetry. However, GRACE/GRACE-FO Mascon products—characterized by lower noise than spherical harmonic solutions—minimize the impact of these interruptions [65,67,68,69]. These discontinuities exert only a negligible influence on long-term trend detection when suitable gap-filling or multi-mission reconstruction methods are applied [71,72]. In addition, long-term hydrological datasets and model outputs have shown robust behavior even under irregular temporal sampling [73,74]. Therefore, no interpolation was performed; instead, the missing months were excluded from GLDAS, ERA5-Land, and in situ datasets to preserve data integrity and ensure consistent temporal alignment across all records.

This study utilized GRACE/GRACE-FO Mascon Release 06.1 Version 03 products, covering the period from April 2002 to April 2024, provided by NASA’s Jet Propulsion Laboratory (JPL) https://grace.jpl.nasa.gov (accessed on 5 September 2024) to examine TWSA in the Kızılırmak Basin. To minimize signal leakage between land and ocean, the Coastal Resolution Improvement (CRI) filter was applied to the Mascon dataset as recommended by [40,69].

The CRI filter reduces spurious coastal contamination by redistributing mascon values through a regularized weighting scheme that constrains boundary conditions along coastlines. The CRI-corrected JPL-Mascon Level-3 products (Release 06.1, v03) were used directly at 0.5° spatial and monthly temporal resolution. Since the correction is embedded in the official dataset, only basin-scale spatial averaging was applied.

TWSA values are reported as monthly anomalies relative to the 2004–2009 mean, following the processing standards defined for the GRACE and GRACE-FO Level-3 products [11,64].

2.2.2. GLDAS-Noah Land Surface Model Data

The Global Land Data Assimilation System (GLDAS), developed by NASA’s Goddard Space Flight Center (GSFC) and NOAA’s National Centers for Environmental Prediction (NCEP), integrate satellite- and ground-based observations to generate high-resolution estimates of terrestrial water and energy fluxes, including soil moisture, evapotranspiration, runoff, and surface temperature [41,75,76]. It provides global outputs for multiple land surface models (Noah, CLM, VIC, Mosaic, and Catchment) at spatial resolutions from 0.25° to 1° and temporal resolutions from 3-hourly to monthly intervals [3,75].

Among the GLDAS land surface models, the Noah scheme is commonly adopted in GRACE-related studies for its balanced treatment of soil moisture, evapotranspiration, and energy exchange processes [77,78,79]. It simulates land surface states and fluxes for use in hydrological and climate applications, capturing processes such as soil moisture dynamics, snow accumulation, and land–atmosphere energy exchange.

The study utilized GLDAS-Noah 2.1 LSM https://hydro1.gesdisc.eosdis.nasa.gov/data/GLDAS/, (accessed on 27 September 2024), which has a spatial resolution of 0.25 degrees and a temporal resolution of one month, to analyze the basin from April 2002 to April 2024. To ensure consistency, we resampled the GLDAS data to the 0.5-degree coarser spatial resolution of GRACE and also updated the temporal average as in the GRACE data. This GLDAS model encompasses measurements of soil moisture (SM), snow water equivalent (SWE), and plant canopy surface water (PCSW).

2.2.3. ERA5-Land Reanalyzed Data

ERA5-Land, developed by the European Centre for Medium-Range Weather Forecasts (ECMWF) under the Copernicus Climate Change Service (C3S), represents the land component of the fifth-generation European reanalysis (ERA5). It provides continuous data from 1981 to the present and is being extended to include records from 1950 onward [80]. ERA5-Land offers high-resolution (hourly to monthly) reanalysis of terrestrial water and energy cycles, including variables such as precipitation, evapotranspiration, runoff, soil moisture, snow depth, and surface temperature [42,81].

In this study, ERA5-Land data https://cds.climate.copernicus.eu/, (accessed on 15 September 2024) with 0.1° spatial and monthly temporal resolution were used to analyze precipitation, evapotranspiration, runoff, and soil moisture over the Kızılırmak Basin for April 2002–April 2024. Compared to GLDAS, the higher spatial resolution of ERA5-Land enables more detailed basin-scale assessment of hydrological processes.

The ERA5-Land fields were aggregated to GRACE’s 0.5° grid, and monthly anomalies were computed relative to the 2004–2009 mean to ensure consistency with GRACE-based analyses.

2.2.4. Local Meteorological Data

Meteorological stations within the Kızılırmak River Basin are unevenly distributed across the region. These data were obtained from 155 meteorological observation stations, including 10 manual and 145 automatic stations currently in operation. However, stations lacking long-term and consistent records were excluded from the analysis due to data insufficiency.

As a result, 22 observation stations that provided uninterrupted monthly total precipitation and monthly average temperature data over the entire study period (April 2002–April 2024) were selected for analysis (see Figure 1).

Incorporating local meteorological data significantly enhances the precision and reliability of water budget assessments. Compared to global datasets such as GRACE, GLDAS, and ERA5-Land, local measurements offer more accurate representations of small-scale watershed dynamics. These ground-based observations are also essential for validating global models, thereby strengthening their credibility. As emphasized by [82], in situ observations are indispensable for evaluating and calibrating remotely sensed and reanalysis-based hydrological variables, especially in complex or arid regions. Furthermore, recent research shows that combining GRACE-based water storage signals with land surface models or auxiliary datasets enhances the robustness of hydrological estimates and enables more reliable characterization of basin-scale water storage dynamics [83,84].

Moreover, local meteorological stations capture microclimatic variations and localized hydro-meteorological conditions more effectively, enabling more accurate assessments of real-world environmental behavior. As demonstrated by Syed et al. [85], incorporating ground-based observations alongside satellite and model products improves the representation of basin-scale hydrological variability. Analyses grounded in localized data thus yield more actionable and context-relevant insights for regional water resource planning and sustainable management.

In this study, monthly total precipitation and average temperature data provided by the Turkish State Meteorological Service (MGM) for the Kızılırmak Basin were used to validate the ERA5-Land dataset. Although these local observations are reliable for capturing near-surface atmospheric conditions, their limited spatial coverage posed challenges for integration into broader datasets. An attempt was made to calibrate ERA5-Land data using the local measurements; however, this did not lead to a significant improvement in correlations with GRACE-derived variables.

Therefore, local meteorological data were used solely for validation purposes, serving as a reference to evaluate the accuracy and applicability of the ERA5-Land dataset in the context of water budget analysis. This approach ensured methodological consistency while enhancing the scientific robustness and credibility of the results.

2.2.5. Software and Tools

Spatial data processing and mapping were performed in free, open-source software QGIS 3.44.0. Numerical computations, time-series analyses, and lagged correlation calculations were implemented in MATLAB R2022a and Python 3.10.4. The Python environment employed standard scientific libraries, including NumPy 2.1.0, Pandas 2.2.3, Matplotlib 3.9.2, Seaborn 0.13.2, and OpenPyXL 3.1.5.

2.3. Research Method

In regions smaller than approximately 100.000 km², the signal-to-noise ratio of GRACE/GRACE-FO measurements tends to be low, and it is necessary to address the inaccuracies [11,68]. To overcome this limitation in our study, we combined GRACE/GRACE-FO observations with GLDAS and ERA5-Land datasets through a basin-scale water balance approach and applied appropriate scaling factors to ensure spatial and temporal consistency of terrestrial water storage changes. The scale factor, also known as the gain factor, is an additional feature that may be incorporated into the GRACE/GRACE-FO masks to mitigate damping and leakage problems [86]. Scale-corrected GRACE/GRACE-FO time series were computed using the following equation:

$${\mathrm{g}}^{\mathrm{ı}}\left(\mathrm{x},\mathrm{y},\mathrm{t}\right)=\mathrm{g}\left(\mathrm{x},\mathrm{y},\mathrm{t}\right)·\mathrm{s}\left(\mathrm{x},\mathrm{y}\right)$$

(1)

where x is the longitude index, y is the latitude index, t is the time index, and the scale factor is s (x, y).

The application of the scale factor correction (Equation (1)) aims to restore attenuated signal amplitudes and reduce land–ocean leakage errors that commonly occur in GRACE/GRACE-FO mascon solutions, particularly over medium-sized basins (<200,000 km²). Scale factors are derived from forward modeling and numerical experiments using land surface models and represent gain coefficients that compensate for spatial filtering effects introduced during GRACE data processing. Several studies have shown that the use of scaling factors increases the reliability of GRACE-derived terrestrial water storage estimates and improves their agreement with model-based or in situ observations, especially in regions where signal loss or attenuation is prominent [2,65,68,87].

We utilized scale factor data in conjunction with CRI-filtered GRACE/GRACE-FO JPL-M data covering the full observation period from April 2002 to April 2024. Figure 2 presents a comparative analysis of TWSA data obtained from scaled and unscaled GRACE/GRACE-FO JPL-M data for the Kızılırmak River Basin during 2002–2024. The figure illustrates the performance metrics (MBE, MAE, RMSE, and Pearson correlation) between the two datasets, showing that the application of scale factors significantly improves the agreement with independent datasets.

Figure 2. Comparative analysis of TWSA data from scaled and unscaled GRACE/GRACE-FO JPL-M data over the Kızılırmak River Basin (2002–2024).

This study integrated datasets with different spatial resolutions (GRACE/GRACE-FO: 0.5°, GLDAS-Noah: 0.25°, ERA5-Land: 0.1°). The 0.5° resolution of GRACE ensures reliable monitoring of large-scale total water storage variations; however, it carries the risk of signal damping and leakage when applied to small or heterogeneous basins. In contrast, the higher spatial resolutions of the GLDAS and ERA5-Land models (0.25–0.1°) allow for a more detailed analysis of surface processes such as precipitation, evapotranspiration, and soil moisture, though they remain sensitive to model parameterization and boundary conditions. In this study, all datasets were resampled to the coarser 0.5° resolution of GRACE and integrated at the basin scale using an area-weighted averaging method. This approach preserved the spatial detail advantages of high-resolution models while maintaining consistency with the globally reliable gravity-based GRACE observations. Consequently, the multi-scale framework enabled a holistic assessment of both large-scale trends and local hydrological processes.

GLDAS terrestrial water storage (TWS) and ERA5-Land water storage change (WSC) values were calculated according to Equations (2) and (3), respectively. These calculations were performed to evaluate the correlations between GRACE/GRACE-FO and both GLDAS-Noah and ERA5-Land datasets across all grid cells covering the basin:

$${\mathrm{T}\mathrm{W}\mathrm{S}}_{{\mathrm{G}\mathrm{L}\mathrm{D}\mathrm{A}\mathrm{S}}_{\mathrm{N}\mathrm{o}\mathrm{a}\mathrm{h}}}=\mathrm{S}\mathrm{M}\mathrm{S}+\mathrm{S}\mathrm{W}\mathrm{E}+\mathrm{P}\mathrm{C}\mathrm{S}\mathrm{W} \left(\mathrm{m}\mathrm{m}\right)$$

(2)

where SMS is the sum of the soil moisture in four layers (0–10 cm, 10–40 cm, 40–100 cm, 100–200 cm), SWE is the snow water equivalent, and PCSW is the plant canopy surface water.

In this study, the water storage change (WSC) was estimated using the water balance equation based on ERA5-Land reanalysis data:

$$\mathrm{W}\mathrm{S}\mathrm{C}=\mathrm{P}-\mathrm{E}\mathrm{T}-\mathrm{R}\mathrm{O} \left(\mathrm{m}\right)$$

(3)

where P is monthly total precipitation, ET is evapotranspiration, and RO is surface runoff. All variables were obtained from ERA5-Land and resampled to match the spatial and temporal resolution of GRACE data.

In addition, local meteorological station data (precipitation and temperature) were not directly included in the water balance equation but were later employed to validate and cross-check the accuracy of ERA5-Land precipitation and ET estimates.

The ERA5-Land soil moisture (SM) dataset provides volumetric soil moisture content across four distinct layers: 0–7 cm, 7–28 cm, 28–100 cm, and 100–289 cm. The total soil moisture storage (SMS) can be computed as the sum of volumetric water content across these layers:

$$\mathrm{S}\mathrm{M}\mathrm{S}=\mathrm{s}\mathrm{w}\mathrm{v}\mathrm{l}1+\mathrm{s}\mathrm{w}\mathrm{v}\mathrm{l}2+\mathrm{s}\mathrm{w}\mathrm{v}\mathrm{l}3+\mathrm{s}\mathrm{w}\mathrm{v}\mathrm{l}4 (\mathrm{k}\mathrm{g} {\mathrm{m}}^{-2})$$

(4)

where $\mathrm{s}\mathrm{w}\mathrm{v}\mathrm{l}$ refers to the volumetric soil water layer values. While the terms SM and SMS are often used interchangeably to indicate soil moisture, in this study, SMS specifically refers to the total volumetric soil moisture derived from the sum of the four ERA5-Land layers. All values are expressed in volumetric units of kg m⁻² (equivalent to mm water thickness).

In this study, ERA5-Land SM data were used in Section 3.4. to examine the lagged correlation between GRACE/GRACE-FO TWSA and soil moisture variations. The rationale for using ERA5-Land SM data in this part is that ERA5-Land provides a physically consistent, state-of-the-art reanalysis product with higher spatial and temporal fidelity, which is well-suited for evaluating short-term variability and lagged responses. On the other hand, GLDAS-Noah SM data were employed in deriving GLDAS based TWSA (terrestrial water storage anomaly) and GWSA components because GLDAS-Noah offers an internally consistent land surface water balance framework (SMS + SWE + PCSW), which is required to separate groundwater storage from total storage. Thus, although both datasets provide soil moisture information and were resampled to the GRACE spatial resolution for comparability, they were deliberately used for different analytical purposes: ERA5-Land SM data for capturing correlation dynamics, and GLDAS-Noah SM data for ensuring consistency in water budget decomposition. This complementary use of two independent reanalysis products also provides an internal robustness check, reducing the dependency on a single model source.

Each monthly GRACE Tellus grid represents the surface mass anomaly for that month with respect to a baseline temporal average (2004–2009). It is essential to compare anomalies with respect to the same temporal average for comparisons with other data or models [11]. In order to evaluate the temporal agreement between TWSA from GLDAS-Noah and water storage change anomaly (WSCA) from ERA5-Land data with GRACE/GRACE-FO TWSA, we calculated monthly anomalies using the following equations:

$${∆}_{\mathrm{T}\mathrm{W}\mathrm{S}\mathrm{A}}={∆}_{\mathrm{T}\mathrm{W}\mathrm{S}}-{∆}_{{\mathrm{T}\mathrm{W}\mathrm{S}}_{\overline{0409}}}$$

(5)

$${∆}_{\mathrm{W}\mathrm{S}\mathrm{C}\mathrm{A}}={∆}_{\mathrm{W}\mathrm{S}\mathrm{C}}-{∆}_{{\mathrm{W}\mathrm{S}\mathrm{C}}_{\overline{0409}}}$$

(6)

where ${∆}_{{\mathrm{T}\mathrm{W}\mathrm{S}}_{\overline{0409}}}$ is the average GLDAS-Noah TWS, ${∆}_{{\mathrm{W}\mathrm{S}\mathrm{C}}_{\overline{0409}}}$ is the average ERA5-Land value has calculated for 2004–2009 which is GRACE/GRACE-FO timeline baseline. The resulting value is subtracted from all monthly TWS/WSC values. Thus, the GLDAS-Noah TWSA and ERA5-Land WSCA values are obtained.

To ensure comparability with GRACE/GRACE-FO data, the GLDAS-Noah (0.25°) and ERA5-Land (0.1°) datasets were first converted to monthly values, consistent with the GRACE temporal resolution (2004–2009 baseline). Subsequently, both datasets were resampled to the 0.5° spatial resolution of the GRACE/GRACE-FO Level-3 mascon products to achieve a common grid scale. Finally, the resampled values were combined using an area-weighted averaging method (Equation (7)) to derive single basin-scale time series of GLDAS-Noah TWSA and ERA5-Land WSCA for the Kızılırmak Basin. For consistency and to avoid unit confusion, all variables and derived products were converted to centimetres (cm) of equivalent water thickness

$${∆\mathrm{T}\mathrm{W}\mathrm{S}\mathrm{A}}_{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{i}\mathrm{n}}={\displaystyle \frac{\sum _{\mathrm{n}=1}^{\mathrm{N}}{\mathrm{T}\mathrm{W}\mathrm{S}\mathrm{A}}_{\mathrm{n}}{·\mathrm{S}}_{\mathrm{n}}}{\sum _{\mathrm{n}=1}^{\mathrm{N}}{\mathrm{S}}_{\mathrm{n}}}}$$

(7)

where ${∆\mathrm{T}\mathrm{W}\mathrm{S}\mathrm{A}}_{\mathrm{b}\mathrm{a}\mathrm{s}\mathrm{i}\mathrm{n}}$ is the basin-averaged total water storage anomaly (cm) computed on the 0.5° × 0.5° grid after resampling all datasets to match the GRACE/GRACE-FO spatial resolution. ${\mathrm{T}\mathrm{W}\mathrm{S}\mathrm{A}}_{\mathrm{n}}$ represents the water storage anomaly in the n-th grid cell, initially computed at the native spatial resolution of each dataset—0.25° × 0.25° for GLDAS-Noah and 0.1° × 0.1° for ERA5-Land. ${\mathrm{S}}_{\mathrm{n}}$ is the area weight of the n-th grid cell, representing the portion of the basin covered by that cell. N denotes the total number of grid cells that fully or partially intersects the basin boundaries.

This procedure preserves the spatial detail of high-resolution models while maintaining full consistency with the coarse resolution but globally reliable GRACE/GRACE-FO observations, resulting in more accurate and physically representative estimates of basin-scale water storage variations.

On the other hand, to calculate GWSA, it is necessary to subtract GLDAS-Noah TWSA ($\mathrm{S}\mathrm{M}\mathrm{S}+\mathrm{S}\mathrm{W}\mathrm{E}+\mathrm{P}\mathrm{C}\mathrm{W}\mathrm{S}$) from the TWS changes provided by GRACE/GRACE-FO.

$$\mathrm{G}\mathrm{W}\mathrm{S}\mathrm{A}={\mathrm{T}\mathrm{W}\mathrm{S}\mathrm{A}}_{\mathrm{G}\mathrm{R}\mathrm{A}\mathrm{C}\mathrm{E}}-{\mathrm{T}\mathrm{W}\mathrm{S}\mathrm{A}\left(\mathrm{S}\mathrm{M}\mathrm{S}+\mathrm{S}\mathrm{W}\mathrm{E}+\mathrm{P}\mathrm{C}\mathrm{W}\mathrm{S}\right)}_{\mathrm{G}\mathrm{L}\mathrm{D}\mathrm{A}\mathrm{S}}$$

(8)

2.3.1. Performance Metrics

This study utilized standard deviation (STD), mean bias error (MBE), mean absolute error (MAE), mean square error (RMSE), and the Pearson correlation coefficient (CC) to assess the relationships between observed and modeled data.

The STD is a quality indicator that shows the distribution of a dataset relative to its meaning (Equation (9)):

$$\mathrm{S}\mathrm{T}\mathrm{D}=\sqrt{{\displaystyle \frac{\sum {\left(\mathrm{X}-\overline{\mathrm{X}}\right)}^2}{\mathrm{n}-1}}}$$

(9)

where $\mathrm{S}\mathrm{T}\mathrm{D}$ is standart deviation, X is each of the values of the data, $\overline{\mathrm{X}}$ is mean value of the data, n is the number of data points.

The MBE (Equation (10)) metric evaluates the average difference between the predictions made by a model (${\mathrm{X}}_{\mathrm{i}}^{\mathrm{m}\mathrm{o}\mathrm{d}}$) and observed data (${\mathrm{Y}}_{\mathrm{i}}^{\mathrm{o}\mathrm{b}\mathrm{s}}$). This metric quantifies the systematic deviation of forecast from the actual values, indicating either an average overestimation or underestimation.

$$\mathrm{M}\mathrm{B}\mathrm{E}={\displaystyle \frac{1}{\mathrm{n}}}\sum _{\mathrm{i}=1}^{\mathrm{n}}\left({\mathrm{Y}}_{\mathrm{i}}^{\mathrm{o}\mathrm{b}\mathrm{s}}-{\mathrm{X}}_{\mathrm{i}}^{\mathrm{m}\mathrm{o}\mathrm{d}}\right)$$

(10)

The MAE (Equation (11)) indicates the mean number of absolute discrepancies between the observed and modeled data.

$$\mathrm{M}\mathrm{A}\mathrm{E}={\displaystyle \frac{1}{\mathrm{n}}}\sum _{\mathrm{i}=1}^{\mathrm{n}}\left|{\mathrm{Y}}_{\mathrm{i}}^{\mathrm{o}\mathrm{b}\mathrm{s}}-{\mathrm{X}}_{\mathrm{i}}^{\mathrm{m}\mathrm{o}\mathrm{d}}\right|$$

(11)

The RMSE (Equation (12)) of a sample is the quadratic means of the differences between the observed value and modeled or predicted ones.

$$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=\sqrt{{\displaystyle \frac{\sum _{\mathrm{i}=1}^{\mathrm{n}}{({\mathrm{Y}}_{\mathrm{i}}^{\mathrm{o}\mathrm{b}\mathrm{s}}-{\mathrm{X}}_{\mathrm{i}}^{\mathrm{m}\mathrm{o}\mathrm{d}})}^2}{\mathrm{n}}}}$$

(12)

The CC (Equation (13)) provides a value between −1 and 1 that describes the extent of linear correlations between predicted and observed data. The CC value of zero (0) means no correlation, while positive (negative) values mean positive (negative) correlation. A value of 1 (or −1) indicates a perfect positive (negative) correlation between predicted and observed values for all combinations also investigated.

$$\mathrm{C}\mathrm{C}={\displaystyle \frac{\sum _{\mathrm{i}=1}^{\mathrm{n}}\left({\mathrm{X}}_{\mathrm{i}}^{\mathrm{m}\mathrm{o}\mathrm{d}}-\overline{\mathrm{X}}\right)\left({\mathrm{Y}}_{\mathrm{i}}^{\mathrm{o}\mathrm{b}\mathrm{s}}-\overline{\mathrm{Y}}\right)}{\sqrt{\sum _{\mathrm{i}=1}^{\mathrm{n}}{({\mathrm{X}}_{\mathrm{i}}^{\mathrm{m}\mathrm{o}\mathrm{d}}-\overline{\mathrm{X}})}^2}\sqrt{\sum _{\mathrm{i}=1}^{\mathrm{n}}{({\mathrm{Y}}_{\mathrm{i}}^{\mathrm{o}\mathrm{b}\mathrm{s}}-\overline{\mathrm{Y}})}^2}}}$$

(13)

Here, ${\mathrm{X}}_{\mathrm{i}}^{\mathrm{m}\mathrm{o}\mathrm{d}}$ represents model values and ${\mathrm{Y}}_{\mathrm{i}}^{\mathrm{o}\mathrm{b}\mathrm{s}}$ represents observed values.

The lagged Pearson correlation coefficient is used to evaluate the linear relationship between two time series when one is temporally shifted by a specified lag (T). This metric, defined in Equation (Equation (14)), consists of two main components: the numerator and the denominator.

• The numerator calculates the covariance between the two series, specifically assessing how deviations from their respective means align after applying the lag. This is done by comparing means ${\mathrm{Y}}_{\mathrm{i}+\mathrm{T}}^{\mathrm{o}\mathrm{b}\mathrm{s}}-\overline{\mathrm{Y}}$ and ${\mathrm{X}}_{\mathrm{i}}^{\mathrm{m}\mathrm{o}\mathrm{d}}-\overline{\mathrm{X}}$, where T is the lag applied to the observed series.
• The denominator normalizes this covariance by the product of the standard deviations of both series, obtained by calculating the square root of the sum of squared deviations from their means.

Together, the formula yields a standardized correlation coefficient ${\mathrm{C}\mathrm{C}}_{\mathrm{T}}$, which ranges from −1 to 1. Positive values indicate a direct relationship, while negative values suggest an inverse relationship.

When T is positive, the observed data series Y is shifted forward in time, implying a delayed effect of the predictor X. Conversely, a negative T shifts the model data X backward, indicating a possible leading influence of the observed variable.

$${\mathrm{C}\mathrm{C}}_{\mathrm{T}}={\displaystyle \frac{\sum _{\mathrm{i}=1}^{\mathrm{N}-\mathrm{T}}\left({\mathrm{X}}_{\mathrm{i}}^{\mathrm{m}\mathrm{o}\mathrm{d}}-{\overline{\mathrm{X}}}_{\mathrm{T}}\right)\left({\mathrm{Y}}_{\mathrm{i}+\mathrm{T}}^{\mathrm{o}\mathrm{b}\mathrm{s}}-{\overline{\mathrm{Y}}}_{\mathrm{T}}\right)}{\sqrt{{\left({\sum }_{\mathrm{i}=1}^{\mathrm{N}-\mathrm{T}}\right({\mathrm{X}}_{\mathrm{i}}^{\mathrm{m}\mathrm{o}\mathrm{d}}-{\overline{\mathrm{X}}}_{\mathrm{T}})}^2}\sqrt{\sum _{\mathrm{i}=1}^{\mathrm{N}-\mathrm{T}}{({\mathrm{Y}}_{\mathrm{i}+\mathrm{T}}^{\mathrm{o}\mathrm{b}\mathrm{s}}-{\overline{\mathrm{Y}}}_{\mathrm{T}})}^2}}}$$

(14)

Here, ${\overline{\mathrm{X}}}_{\mathrm{T}}\mathrm{a}\mathrm{n}\mathrm{d}{\overline{\mathrm{Y}}}_{\mathrm{T}}$ denote the mean values calculated over the overlapping period after applying the lag T. This formulation allows for the investigation of time-dependent dynamics between variables, which is particularly valuable in hydrological and climate studies where system responses may be delayed.

Where (Equations (10)–(14)), ${\mathrm{Y}}_{\mathrm{i}}^{\mathrm{o}\mathrm{b}\mathrm{s}}$ is the GRACE/GRACE-FO TWSA data and ${\mathrm{X}}_{\mathrm{i}}^{\mathrm{m}\mathrm{o}\mathrm{d}}$ is the GLDAS-Noah/ERA5-Land data, $\overline{\mathrm{Y}} \mathrm{a}\mathrm{n}\mathrm{d} \overline{\mathrm{X}}$ is the mean value of the observed and model data respectively.

2.3.2. Uncertainty Assessment of GWSA Estimation

To quantitatively evaluate the uncertainty of the derived groundwater storage anomalies (GWSA), both GRACE/GRACE-FO and GLDAS-Noah uncertainties were incorporated. The uncertainty of GRACE mascon data was obtained from the “uncertainty” layer in the JPL product, rescaled to 0.5° resolution using the official scale factors and averaged over the Kızılırmak River Basin. For GLDAS, the uncertainty of soil moisture, snow water equivalent, and canopy water storage components was estimated based on their 12-month moving standard deviations, and combined following the variance-propagation rule. The resulting total uncertainty of GWSA was computed as:

$${\mathsf{\sigma }}_{\mathrm{GWSA}}=\sqrt{{\mathsf{\sigma }}_{\mathrm{GRACE}}^2+{\mathsf{\sigma }}_{\mathrm{GLDAS}}^2}$$

(15)

3. Results

3.1. Spatial Distribution of GRACE-Derived Total Water Storage Anomalies (TWSA)

The spatial distribution of TWSA derived from GRACE/GRACE-FO observations provides critical insights into long-term hydrological changes across the Kızılırmak River Basin. Figure 3 illustrates the basin-wide TWSA conditions for (a) April 2002, (b) April 2024, and (c) the long-term change between these two periods. In April 2002 (Figure 3a), the basin exhibited predominantly positive TWSA values, particularly in the northern and central regions, indicating relatively high-water storage levels. By contrast, the April 2024 distribution (Figure 3b) reveals a widespread decline in TWSA, especially across the southern and eastern sub-basins. The resulting change map (Figure 3c) shows that most regions experienced significant depletion over the 22-year period, with losses reaching up to −20 cm.

Figure 3. GRACE/GRACE-FO TWSA values over the Kızılırmak River Basin: (a) April 2002, (b) April 2024, and (c) long-term change in TWSA between April 2002 and April 2024.

3.2. Temporal Variations in Water Storage Components

This section investigates the relationship between GLDAS-Noah terrestrial water storage anomaly (TWSA) and GRACE/GRACE-FO total water storage anomaly (TWSA), with particular consideration of temporal lags. In addition, we analyze the groundwater storage anomaly (GWSA), derived from the difference between GRACE/GRACE-FO TWSA and GLDAS-Noah TWSA, to provide a more comprehensive assessment of the basin’s water budget.

Figure 4a,b illustrate the temporal evolution of TWSA obtained from GRACE/GRACE-FO and GLDAS-Noah, as well as the effect of applying a temporal lag to the GRACE/GRACE-FO dataset. TWSA captures variations in total water resources, including groundwater, soil moisture, surface water, and snow cover. While GRACE/GRACE-FO directly observes these variations, the GLDAS-Noah model represents them through numerical simulations.

Figure 4. (a) This figure demonstrates the direct comparison between the GRACE/GRACE-FO data and GLDAS-Noah datasets. The red line represents the GRACE/GRACE-FO TWSA data, while the blue solid line corresponds to the GLDAS-Noah model estimates. The black line illustrates GWSA. Transparency shades the positive (green) and negative (red) anomalies, while a green dashed line illustrates the GWSA trend. (b) presents a comparison between the GRACE/GRACE-FO data and GLDAS-Noah datasets, following the application of a −1 month time latency to the independent variable, GRACE/GRACE-FO data. The GLDAS-Noah model estimates align with the 1-month forward shift in the GRACE/GRACE-FO data. The red line represents the −1 month-shifted GRACE/GRACE-FO TWSA data, while the blue solid line represents the GLDAS-Noah model estimates. The black line again illustrates GWSA, with positive (green) and negative (red) anomalies shaded transparently. A green dashed line illustrates the GWSA trend.

The direct comparison between GLDAS-Noah and GRACE/GRACE-FO data shows a strong correlation coefficient of 0.72. The mean error levels (MAE = 5.17 cm, RMSE = 6.35 cm) are acceptable for large-scale TWSA studies; however, they indicate systematic discrepancies between the datasets. This suggests that the GLDAS model effectively captures seasonal variations, but it shows inconsistencies in magnitude relative to GRACE/-FO, particularly during certain periods. Figure 4b shifts the GLDAS-Noah data one month forward before comparing it with the GRACE/GRACE-FO data. After applying the time lag, the correlation coefficient rises to 0.82, indicating a significant improvement in alignment. The improvement in error metrics (MAE = 4.02 cm, RMSE = 5.14 cm) suggests that the time-shifted GLDAS data is more consistent with the GRACE/GRACE-FO measurements. In both part of Figure 4, the long-term trend of GWSA indicates a persistent water loss in the basin. Figure 4a shows the trend as −9.55 cm, whereas Figure 4b shows a slight difference as −9.52 cm.

To quantify the reliability of this depletion estimate, an uncertainty assessment was performed by combining GRACE/GRACE-FO and GLDAS-Noah uncertainties through variance propagation. The analysis yielded an average monthly combined uncertainty of ±2.6 cm for the basin. The long-term GWSA trend (−9.55 cm between 2002 and 2024) was then reassessed using linear regression and found to be statistically significant (p < 0.001, 95% confidence level). Here, ±2.6 cm represents the mean monthly observational and model uncertainty, confirming that the long-term depletion trend is well beyond the uncertainty range.

3.3. Validation of ERA5-Land and Local Meteorological Datasets

3.3.1. Validation of ERA5-Land Precipitation Data with Local Meteorological Precipitation

Hydrological processes frequently compare and complementarily utilize local meteorological total precipitation data and model-based total precipitation datasets such as ERA5-Land. The findings of this study demonstrate that the ERA5-Land precipitation dataset accurately and reliably represents the meteorological characteristics of the Kızılırmak River Basin in estimating the basin’s monthly total precipitation. Furthermore, the utilization of this dataset provides a significant facilitative tool for studies aimed at determining the basin’s water budget. In this section, we conducted a comprehensive analysis of the correlation between the two datasets.

The accuracy metrics presented in Table 2 demonstrate a strong agreement between local meteorological precipitation data and ERA5-Land precipitation anomalies. The minimal mean bias error (MBE = −1.30 cm) indicates the absence of systematic bias, while the high correlation coefficient (0.91 ± 0.02) reveals a statistically significant linear relationship between the datasets. These findings confirm that the ERA5-Land dataset serves as a reliable and consistent data source for hydrological studies at the basin scale.

Table 2. Comparative accuracy metrics of local meteorological precipitation and ERA5-Land precipitation.

STD (cm)	MBE (cm)	MAE (cm)	RMSE (cm)	CC (%95)
1.27	−1.30	1.43	1.81	0.91 ± 0.02

It is important to note that the spatial distribution of local meteorological stations across the Kızılırmak River Basin is heterogeneous, and many stations contain temporal gaps in precipitation records. As a result, only 22 stations with complete datasets over the full study period (April 2002–April 2024) were included in this analysis. For these local meteorological precipitation and temperature data, a basin-scale mean time series was constructed by averaging the records from the 22 selected stations. Since the stations were not uniformly distributed, both simple arithmetic mean and spatial representativeness were considered. While gridded datasets (ERA5-Land, GLDAS) were aggregated using area-weighted averaging (Equation (7)), station-based data were aggregated using arithmetic averaging to obtain a single basin-scale representative series. In contrast, ERA5-Land precipitation data offer continuous temporal coverage and uniform spatial resolution across the basin.

3.3.2. Validation of ERA5-Land Temperature Data with Local Meteorological Temperature

Monthly average temperature data from 22 meteorological stations with uninterrupted records between April 2002 and April 2024 in the Kızılırmak River Basin were compared with ERA5-Land 2-m air temperature data. The comparison revealed an exceptionally high correlation coefficient (CC > 0.99) between the two datasets, indicating that ERA5-Land reliably captures temperature variability at the basin scale (Table 3).

Table 3. Comparative accuracy metrics of local meteorological monthly average temperature and ERA5-Land monthly 2 m air temperature.

STD (°C)	MBE (°C)	MAE (°C)	RMSE (°C)	CC (%95)
0.56	1.45	1.46	1.55	>0.99 ± 0.01

To further illustrate seasonal temperature variability across the Kızılırmak River Basin, a circular climatology chart was generated using monthly average temperature data from local meteorological stations. The resulting temperature profile, together with the high correlation with ERA5-Land data (CC = 0.998), demonstrates consistent seasonal and interannual temperature patterns across the basin (see Figure 5).

Figure 5. Monthly average temperature climatology (2002–2024) for the Kızılırmak River Basin based on local meteorological station data. (a) Polar chart illustrates seasonal variations. (b) Time series of monthly average temperature (April 2002–April 2024), showing long-term patterns and interannual variability.

Figure 5 presents the seasonal and long-term behavior of monthly average temperatures over the Kızılırmak River Basin between April 2002 and April 2024. The polar diagram (Figure 5a) effectively captures the annual temperature cycle, with peak values observed during July–August and minimum values around January, reflecting the basin’s continental climate characteristics. Meanwhile, the time series plot (Figure 5b) illustrates consistent interannual oscillations alongside a notable pattern in recent years: summer peaks appear increasingly sharper, while winter troughs are relatively less pronounced.

3.4. Delayed Response of ERA5-Land Hydrological Variables to GRACE/GRACE-FO TWSA

In this section, we investigated the delayed response of the ERA5-Land data between GRACE/GRACE-FO TWSA. Figure 6, Figure 7, Figure 8, Figure 9, Figure 10 and Figure 11 illustrates the lagged correlation between GRACE/GRACE-FO TWSA and various ERA5-Land hydrological variables under time lags of +3 to −3 months. Each subplot considers GRACE/GRACE-FO TWSA as the independent variable and visualizes its relationship with the corresponding dependent variable (e.g., temperature anomaly, soil moisture, precipitation, evapotranspiration, runoff, and P–ET–R). Graph vertices are color-coded based on GRACE/GRACE-FO TWSA values, with blue shades indicating low and red shades indicating high water storage levels. The regression line, Pearson correlation coefficient (CC), and error metrics (MAE, MBE, RMSE) with 95% confidence intervals reflect the strength, direction, and uncertainty of the relationship across different lags.

Figure 6. (a–g) depicts the correlation plots between GRACE/GRACE-FO TWSA and ERA5-Land Temperature Anomaly, including lags of 3, 2, 1, 0, −1, −2 and −3 months respectively.

Figure 7. (a–g) depicts the correlation plots between GRACE/GRACE-FO TWSA and ERA5-Land Soil Moisture Anomaly, including lags of 3, 2, 1, 0, −1, −2 and −3 months respectively.

Figure 8. (a–g) depicts the correlation plots between GRACE/GRACE-FO TWSA and ERA5-Land Precipitation Anomaly, including lags of 3, 2, 1, 0, −1, −2 and −3 months respectively.

Figure 9. (a–g) depicts the correlation plots between GRACE/GRACE-FO TWSA and ERA5-Land Evapotranspiration Anomaly, including lags of 3, 2, 1, 0, −1, −2 and −3 months respectively.

Figure 10. (a–g) depicts the correlation plots between GRACE/GRACE-FO TWSA and ERA5-Land Runoff Anomaly, including lags of 3, 2, 1, 0, −1, −2 and −3 months respectively.

Figure 11. (a–g) depicts the correlation plots between GRACE/GRACE-FO TWSA and ERA5-Land WSCA (PA−ETA−ROA), including lags of 3, 2, 1, 0, −1, −2 and −3 months respectively.

To enable consistent comparison with GRACE/GRACE-FO TWSA, all datasets—including ERA5-Land hydrological variables—were converted into anomalies by removing their respective monthly climatological means. This ensured that all time series represent deviations from the same reference period, allowing for accurate correlation and lag analysis in line with the GRACE anomaly framework.

In Figure 6, Figure 7, Figure 8, Figure 9, Figure 10 and Figure 11, each scatter point represents a monthly paired observation between GRACE/GRACE-FO TWSA and the corresponding hydrometeorological variable (e.g., precipitation, evapotranspiration, runoff, temperature, or soil moisture) over the study period (April 2002–April 2024). Thus, the total number of points reflects the length of the overlapping monthly time series used in the correlation analysis (after considering possible lags), and is independent of the number of meteorological stations. This ensures that the figures consistently illustrate basin-scale temporal dynamics derived from aggregated datasets rather than station density.

In the lagged correlation analysis between GRACE TWSA and ERA5-Land tempera-ture data (Figure 6f,g), the correlation coefficients at −2 and −3 months were found to be as high as −0.69 and −0.70, respectively.

Assessing the basin water budget requires monitoring the temporal variations of GRACE/GRACE-FO TWSA and soil moisture, which aids in comprehending groundwater and surface water storage dynamics throughout the hydrological cycle. Figure 7. illustrates the time-lagged relationship between GRACE/GRACE-FO TWSA and ERA5-Land soil moisture anomaly across seven latency intervals ranging from −3 to +3 months. The highest positive correlation is observed at a −1-month lag (CC = 0.82; Figure 7e).

The lagged correlation analysis between GRACE/GRACE-FO TWSA and ERA5-Land precipitation anomaly reveals the highest positive association at a −2-month lag (CC = 0.59; Figure 8f). This finding suggests that the impact of precipitation on surface and subsurface water components is reflected in total water storage changes measurable by GRACE approximately two months later. The corresponding error metrics (MAE: 5.71 cm, RMSE: 7.14 cm) confirm that the relationship is statistically significant and consistent.

The lag analysis revealed that the strongest negative correlation between GRACE/GRACE-FO total water storage anomalies (TWSA) and ERA5-Land evapotranspiration (ET) anomalies occurs at a +2-month lag (CC = −0.69; Figure 9b).

Conversely, the positive correlation observed at a −3-month lag (CC = 0.55; Figure 9g) indicates that increases in evapotranspiration may lead to detectable water loss in GRACE data approximately three months later.

The lagged correlation analysis between GRACE/GRACE-FO TWSA and ERA5-Land runoff anomaly, given in Figure 10, revealed the strongest positive relationship at zero-month lag (CC=0.71; Figure 10d).

In the lagged correlation analysis between GRACE/GRACE-FO TWSA and ERA5-Land variables, individual hydrological components (e.g., precipitation, evapotranspiration, runoff) tend to show the strongest correlations at lags of 1 to 3 months. In contrast, the correlation between GRACE TWSA and the model-based theoretical water storage closure (WSCA = PA − ETA − ROA) peaks at zero-month lag (Figure 11d).

4. Discussion

This study provides an integrated assessment of terrestrial water storage (TWS) dynamics in the Kızılırmak River Basin by combining multiple data sources, including GRACE/GRACE-FO satellite gravimetry, GLDAS land surface modeling, ERA5-Land reanalysis, and local meteorological observations. The results revealed pronounced seasonal and interannual variability with evident phase lags among precipitation, evapotranspiration, soil moisture, and groundwater storage. Such time-shifted responses are consistent with earlier investigations in semi-arid and Mediterranean-type basins, where delayed groundwater recharge and evapotranspiration feedbacks shape the hydrological cycle [3,4,14,25,85,88,89]. Furthermore, the results indicate a significant decrease in total water storage across much of the basin. This is likely due to prolonged drought, excessive uncontrolled water abstraction, and climate-related factors.

The strongest lagged correlations identified for each variable were statistically significant at the 95% confidence level (p-value < 0.001), while weaker correlation lags showed no significant relationship. In addition, 95% confidence intervals were computed and are presented in Figure 6, Figure 7, Figure 8, Figure 9, Figure 10 and Figure 11 as ± values accompanying each correlation coefficient, indicating the uncertainty range of the estimates. These results confirm that the observed correlations between GRACE/GRACE-FO TWSA and ERA5-Land hydrological variables are statistically robust and not due to random variability. However, it is important to distinguish between statistical significance and practical (hydrological) relevance. While the correlations are statistically significant, their practical interpretation depends on the physical and hydrological context of the basin processes. Therefore, in this study, statistical results were interpreted together with their hydrological implications rather than relying solely on p-values.

In terms of model performance, error metrics reveal within acceptable bounds for basin-scale hydrological validation. These results confirm the utility of GRACE/GRACE-FO observations not only as independent references for validating land surface model outputs but also as critical benchmarks for improving regional water budget closure and guiding sustainable water management decisions under conditions of short-term hydrological variability.

In addition, sensitivity analysis confirmed that resampling GLDAS (0.25°) and ERA5-Land (0.1°) data to GRACE’s 0.5° resolution caused negligible information loss (r = 0.999, p < 0.001; MAE < 1.5 cm), validating the suitability of 0.5° resolution for medium-scale basins such as the Kızılırmak.

Compared with previous GRACE-only studies, the multi-source integration applied here provided a more robust evaluation of basin-scale water budget closure. While GRACE and GRACE-FO observations remain indispensable for capturing long-term storage anomalies and secular groundwater depletion trends, the combination with ERA5-Land and in situ meteorological data improved the temporal resolution of short-term variability and enhanced interpretation of hydro-climatic drivers.

Validation using local meteorological observations showed that ERA5-Land precipitation (CC = 0.91) and temperature (CC > 0.99) effectively capture basin-scale hydro-climatic variability across the Kızılırmak River Basin. The validation results confirm the robustness of ERA5-Land temperature estimates, consistent with previous studies reporting very high correlations (CC > 0.99) between ERA5 2 m air temperature and ground-based observations across diverse climatic regions such as China and Central Asia [90]. Similarly, the strong agreement in precipitation supports its applicability for basin-level hydrological modeling and long-term water budget estimation. Importantly, local temperature and precipitation records were used solely for validation—not calibration—ensuring methodological independence and enhancing confidence in the ERA5-Land dataset as a reliable proxy for hydrological analysis. Although integrating local rainfall observations did not markedly improve the correlation with GRACE TWSA, it provided valuable cross-validation that reinforced the reliability of ERA5-Land inputs for basin-scale water balance studies.

Furthermore, the seasonal temperature profile, supported by both local observations and ERA5-Land estimates, underlines the capability of reanalysis products to capture basin-scale temperature variability and reinforces their suitability for hydroclimatic correlation studies involving evapotranspiration, soil moisture, and runoff dynamics.

Additionally, the ERA5-Land dataset provides uniform spatial coverage and continuous temporal records, which helps compensate for the limitations of ground-based measurements. Therefore, the comparison between local and ERA5-Land datasets serves a dual purpose: (i) validating the reliability of the ERA5-Land data against in situ measurements, and (ii) highlighting the representational limitations of sparse and unevenly distributed local observations. This dual assessment improves the interpretability and credibility of the hydrological results by integrating the complementary strengths of both data sources.

The lagged correlation analysis between GRACE TWSA and ERA5-Land temperature data (Figure 6) revealed strong negative relationships, with correlation coefficients of −0.69 and −0.70 at −2 and −3-month lags, respectively. These strong negative correlations indicate that temperature increases have a delayed but significant impact on the total water storage capacity of the basin. This is primarily because temperature changes do not affect total water storage directly, but rather through enhanced evaporation and evapotranspiration processes, which deplete both surface and subsurface water components. In particular, temperature rises reduce soil moisture and intensify evapotranspiration, leading to water loss that becomes detectable by GRACE satellites as a measurable mass change approximately two to three months later. Similar results were reported by [91], who found that terrestrial water storage in permafrost regions responds to surface temperature increases with a lag of several months. Such delayed responses also reflect the hydrological buffering capacity of the basin and the time required for water to percolate from the surface into deeper subsurface layers. Moreover, the findings of [92] highlighted that GRACE TWSA exhibits delayed correlations with climatic variables such as temperature, evapotranspiration, and soil moisture.

The 1–3-month lagged responses between GRACE-derived TWSA and variables such as soil moisture and temperature can be explained by the hydro-geological and land-use characteristics of the Kızılırmak Basin. The basin consists of extensive alluvial plains (e.g., Kayseri, Avanos, Bafra) underlain by low-permeability clay–marl formations that delay infiltration and groundwater recharge [56]. Annual precipitation (≈446 mm) mainly occurs between October and May, and the semi-arid climate, together with deep aquifer systems, prolongs the percolation of surface water to the root zone. Intensive agricultural activity and vegetation coverage enhance evapotranspiration and soil-water retention, contributing to the phase lags observed between surface fluxes and storage changes. Similar processes were reported in the Fırat Basin [93], where a ~3-month lag was attributed to percolation time, and over Iran [3], where 2–3-month delayed responses of GRACE-based water storage to precipitation and temperature were found. These mechanisms confirm that the observed lags in the Kızılırmak Basin stem from intrinsic hydro-geological response times and evapotranspiration feedbacks.

As shown in Figure 7, GRACE/GRACE-FO TWSA exhibits a strong relationship with ERA5-Land soil moisture anomaly, reflecting coherent variations between groundwater and near-surface storage components. The highest positive correlation is observed at a −1-month lag (CC = 0.82; Figure 7e), indicating that variations in total water storage are most strongly manifested in soil moisture conditions approximately one month later. This time shift reflects the basin’s hydrological response delay, during which precipitation and recharge influence surface and subsurface water components, which in turn increase soil moisture through infiltration, capillary action, and upward flux from shallow groundwater. The consistent lagged correlation structure emphasizes the interconnected dynamics between total water storage and near-surface moisture processes. GRACE-derived TWSA, by integrating groundwater, soil moisture, and surface water signals, provides a robust dataset for evaluating subsurface hydrological responses—particularly in regions where in situ data are sparse. Furthermore, the highest correlation achieved at a −1-month lag, compared to lower correlations at other time shifts, suggests that this offset yields the most coherent temporal agreement between the satellite and model outputs.

The cumulative effects of post-precipitation processes such as infiltration, surface runoff, and groundwater recharge take time to manifest in integrated water storage signals (see Figure 8). Similarly, studies have shown that GRACE is particularly sensitive to slowly responding hydrological components such as groundwater [94] and exhibits delayed responses emphasizing that GRACE TWSA provides an integrated but time-delayed hydrological signal [85,88]. This behavior is also reflected in the moderate correlation observed between GRACE TWSA and ERA5-Land precipitation, with the strongest relationship occurring at a −2-month lag (CC = 0.59), indicating that precipitation-driven water storage changes typically emerge in GRACE signals after a delay of approximately two months.

In addition to rainfall-related delays, thermal and evaporative processes also play a crucial role in shaping basin-scale storage variability. Overall, the obtained correlation values confirm that temperature-induced evapotranspiration (Figure 9) leads to a measurable decline in groundwater and other storage components within approximately two months. This behavior is consistent with the moderate negative correlation observed between GRACE TWSA and ERA5-Land evapotranspiration, with the strongest relationship occurring at a −2-month lag (CC = −0.69), indicating that increased evapotranspiration reduces total water storage with a delayed effect. This demonstrates the sensitivity of GRACE TWSA to climatic forcings and underlines its importance as an indicator of basin-scale hydrological dynamics. This also suggests that increases in total water storage tend to have a delayed and suppressing effect on evapotranspiration, typically becoming noticeable around two months later. The delayed response may be explained by water being retained in deeper subsurface layers or moving away from the surface, thereby reducing the moisture available for evaporation.

Collectively, these results confirm the reliability of GRACE/GRACE-FO observations as robust indicators for tracking precipitation-driven (Figure 8) hydrological responses and evapotranspiration-induced (Figure 9) variations in basin-scale water storage. Moreover, GRACE effectively captures both slow-responding subsurface components such as groundwater and rapidly varying surface processes like runoff. This coherence with previous findings [89], which reported that the correlation between runoff and GRACE-derived TWSA is typically strongest at zero or +1-month lag, reinforces the ability of GRACE to represent the full spectrum of hydrological dynamics and to account for inherent time lags in regional water-budget assessments.

Unlike the delayed responses observed for precipitation and evapotranspiration, runoff shows a nearly immediate effect on total water storage. This finding indicates that runoff generated after precipitation is reflected in GRACE-observed total water storage changes without delay. The high correlation at zero lag (CC = 0.71; Figure 10d), along with similarly strong correlations at +1 month (CC = 0.69; Figure 10c) and −1 month (CC = 0.56; Figure 10e), suggests that runoff is a rapidly occurring process within the basin and contributes promptly to the GRACE signal. In contrast, the noticeable decrease in correlation values at −2 and −3-month lags (CC = 0.28 and CC = −0.03; Figure 10f,g respectively) confirms that the influence of runoff is temporally limited and short-lived.

The observed lag characteristics can be attributed to the physical nature of GRACE observations. This is because GRACE does not measure hydrological processes directly but senses their cumulative mass effects on the Earth’s gravity field, which develop over time. GRACE detects total water storage changes physically through gravimetric measurements, whereas precipitation and subsequent infiltration, evapotranspiration, runoff, and recharge processes influence these mass changes with inherent delays. Consequently, lagged responses are expected when correlating GRACE with individual components. However, WSC represents the instantaneous theoretical storage change derived from simultaneous hydrological fluxes, and therefore, a synchronous relationship with GRACE is theoretically anticipated. Consistent with this, our results showed the highest correlation at lag-0 between GRACE TWSA and WSCA (CC = 0.54; Figure 11d).

This agreement indicates that the water balance computed from ERA5-Land components is consistent with GRACE-observed storage variations and reinforces the reliability of GRACE/GRACE-FO as a robust reference not only for process-level analysis (via lagged relationships with individual fluxes) but also for budget-level validation of reanalysis data.

Beyond the consistency with ERA5-Land, GRACE/GRACE-FO observations also showed strong coherence with GLDAS-Noah model outputs, enabling the derivation of groundwater storage anomalies (GWSA) through their difference. This integration bridges satellite gravimetry and land surface modeling, allowing groundwater dynamics to be isolated from total water storage changes. The comparative analysis between GRACE and GLDAS highlighted that the long-term depletion trend detected in GWSA is consistent with GRACE-based storage declines, confirming that groundwater variations are the dominant driver of basin-scale water loss.

The integration of satellite gravimetry, reanalysis products, and ground-based observations thus proved highly effective for basin-scale hydrological assessment, enabling multiple cross-validations and enhancing confidence in water budget closure estimates. The long-term groundwater depletion (−9.55 cm between 2002 and 2024) emphasizes the urgent need for improved groundwater regulation and sustainable water-use strategies in agriculturally intensive regions of central Türkiye.

The consistency between the long-term GWSA trend (−9.55 ± 2.6 cm) and the associated uncertainty analysis confirms the robustness of the detected signal, indicating that the observed decline is not an artifact of GRACE or GLDAS errors but a genuine, basin-wide groundwater loss driven by both climatic variability and anthropogenic abstraction.

Moreover, the temporal evolution of groundwater storage anomalies (GWSA) indicates a persistent and long-term water loss in the basin. Negative GWSA values closely align with the observed decreases in GRACE-based TWSA, confirming that groundwater variations are the primary contributor to total water storage changes. In contrast, positive GWSAs generally preceded increases in TWSA, highlighting the link between groundwater recovery and surface water dynamics.

Between 2010 and 2014, significant positive GWSA values were detected, followed by pronounced negative anomalies after 2015. This temporal pattern indicates that the GLDAS model alone cannot precisely capture the timing and amplitude of these seasonal variations without a time-lag adjustment. After applying a −1-month lag correction, the adjusted GLDAS dataset reproduced groundwater fluctuations more accurately and achieved stronger agreement with GRACE/GRACE-FO observations, improving the correlation from 0.72 to 0.82 and reducing both MAE and RMSE.

During major drought episodes (e.g., 2006–2007, 2013–2014, 2017, 2020–2021, and post-2021), the lag-corrected GLDAS anomalies aligned more closely with GRACE-derived TWSA signals, demonstrating that the applied correction effectively represents delayed groundwater–surface water interactions in the basin. These results confirm that temporal synchronization between model simulations and satellite observations is essential for accurately capturing basin-scale hydrological dynamics.

5. Conclusions

By integrating GRACE/GRACE-FO satellite gravimetry, GLDAS-Noah land surface modeling, ERA5-Land reanalysis, and ground-based meteorological observations, this study provides a comprehensive assessment of the water budget dynamics in the Kızılırmak River Basin. The results indicate a sustained groundwater depletion of −9.55 ± 2.6 cm between 2002 and 2024, reflecting unsustainable groundwater abstraction, especially in agriculturally intensive areas.

The strong agreement between GRACE-derived total water storage anomalies and the lagged correlation analysis of ERA5-Land hydrological variables and GLDAS components confirms the capability of satellite gravimetry to capture both rapid surface and slow subsurface hydrological responses. The application of lagged correction between GRACE and GLDAS datasets improved temporal consistency and reduced model error, highlighting the importance of accounting for time offsets in multi-source hydrological assessments.

Despite limitations such as missing GRACE/GRACE-FO observations, coarse spatial resolution, and model-based uncertainties [92,95,96,97], the integrated approach successfully reproduced basin-scale spatiotemporal variability. Validation with in situ meteorological data confirmed the reliability of ERA5-Land precipitation and temperature estimates for hydroclimatic analysis.

This study represents one of the first basin-scale water budget assessments in Türkiye that combines satellite gravimetry, land-surface models, and local observations within a fully reproducible and transferable analytical framework.

Future work should (1) fill GRACE data gaps using machine learning or data assimilation, (2) enhance GRACE–model coupling through local calibration, and (3) expand groundwater and meteorological monitoring networks to reduce uncertainties. Such advances will strengthen basin-scale hydrological modeling and support sustainable water management under increasing climatic and anthropogenic pressures.

From a management perspective, the observed groundwater decline calls for basin-specific interventions. In Türkiye, approximately three-quarters of total water withdrawals are allocated to agricultural irrigation [98,99]. Therefore, improving irrigation efficiency, promoting low-water-demand crops, modernizing conveyance infrastructure, and enforcing groundwater licensing and metering are essential steps toward sustainable aquifer recovery and long-term water security in the Kızılırmak River Basin.

Overall, this study demonstrates that the integration of GRACE/GRACE-FO satellite data, land surface model, atmospheric reanalysis model, and local meteorological data provides a reliable framework for long-term groundwater monitoring and lagged hydrological assessments. The findings of the study offer actionable guidance for sustainable groundwater management in climate-vulnerable basins.