Geodetic Assessment of Drought Intensity and Hydrological Dynamics in the Cantareira System, Southeastern Brazil

Abstract

Hydrological drought, expressed as anomalously low water availability in rivers, aquifers, and reservoirs, poses a growing threat to water security in densely populated regions such as the Metropolitan Region of Sao Paulo (MRSP), Brazil. Characterizing how drought propagates into large-scale terrestrial water storage (TWS) deficits, and how these deficits translate into measurable surface deformation, remains challenging in complex, human-managed hydrological systems such as the Cantareira Water Supply System. This study conducts an integrated, multi-sensor geodetic assessment of drought-related hydrological dynamics in the Cantareira System, combining vertical/LOS ground displacement derived from continuous GPS observations and Sentinel-1 InSAR time series with terrestrial water storage anomalies (TWSAs) from GRACE/GRACE-FO, groundwater level records, reservoir storage, and meteorological drought indicators. Cross-correlation analysis reveals a strong and statistically significant coupling between GRACE-TWSA and GPS-derived vertical displacement, with correlation coefficients of $r=-0.8$ (Upper Tietê basin) and $r=-0.6$ (PCJ basin), consistent with an elastic crustal response to hydrological loading. Reservoir storage in the PCJ basin is similarly correlated with regional TWSA (up to $r=0.7$ for the Jaguari–Jacareí reservoir), reinforcing GRACE’s sensitivity to the main upstream storage component of the Cantareira System. Empirical Orthogonal Function (EOF) decomposition of the InSAR deformation fields, retaining the first four modes (72% of variance in Upper Tietê, 66% in PCJ), further demonstrates that deformation patterns are strongly influenced by hydrogeological controls, with distinct spatial responses between the PCJ basin, where hydroclimatic forcing is more pronounced, and the more urbanized and anthropogenically influenced Upper Tietê basin. The integrated dataset captures major drought episodes between January 2020 and Dicember 2025, demonstrating the capability of combining GPS, InSAR, and GRACE observations to quantitatively link climatic drought forcing, terrestrial water storage deficits, and surface deformation, thereby providing a robust approach for monitoring water storage changes and supporting water resource management in densely populated regions.

1. Introduction

Droughts have become more frequent and severe in many regions of the world as a consequence of climate change, land-use transformation, and increasing water demand. Southeastern Brazil, and particularly the Metropolitan Region of São Paulo (MRSP), has experienced recurrent water shortages over the past decade, culminating in critical events that severely impacted domestic, industrial, and agricultural water supply. The 2014–2015 drought in southeastern Brazil alone resulted in economic losses exceeding 5 billion dollars [1].

Drought is a complex hydroclimatic phenomenon that can be expressed through different components of the hydrological cycle. Meteorological drought is generally associated with a persistent precipitation deficit, whereas hydrological drought refers to anomalously low water availability in surface and subsurface water systems, including rivers, lakes, groundwater, and reservoirs [2,3]. These different drought types are interconnected but do not necessarily develop simultaneously, as the propagation of precipitation deficits through the hydrological system is modulated by catchment characteristics, water storage, evapotranspiration, and human activities [2,4]. The occurrence and evolution of droughts are also influenced by large-scale climate variability, including ENSO, which can modify regional precipitation patterns and consequently affect hydrological conditions in southeastern Brazil. However, the relationship between climate variability and hydrological drought is not necessarily direct, because the transformation of atmospheric anomalies into hydrological deficits depends on the temporal persistence of the forcing and the storage and response characteristics of the hydrological system [2,4].

Although Brazil has 14% of all the world’s freshwater, its spatial distribution is highly heterogeneous. In the Amazon region, for example, the per capita water availability is around 700,000 m³/year, while in the MRSP, it is only 280 m³/year [5]. This difference generates even greater impacts on drought scenarios. Several studies indicate that, since 2015, the river basins that supply the MRSP for energy production and human consumption have not recovered, remaining in critical condition [6]. The Cantareira System (CS), which supplies water to nearly nine million people, plays a central role in regional water security and is highly sensitive to prolonged meteorological and hydrological droughts.

Traditional drought monitoring relies on in situ observations such as precipitation, temperature, reservoir levels, and groundwater wells. While these data are essential, they are often spatially sparse and may not fully characterize deficits in complex and/or deep hydrological systems. In recent years, satellite geodetic techniques have emerged as powerful complementary tools for monitoring the Earth’s hydrological cycle [7]. Variations in water mass produce measurable elastic deformations in the crust, detectable by observations from Global Navigation Satellite Systems (GNSSs) and Interferometric Synthetic Aperture Radar (InSAR).

InSAR techniques allow the detection of ground deformations by comparing the phase of radar signals acquired at different times. Through multi-temporal approaches, such as the Persistent Scatterer InSAR (PS-InSAR) methodological principle [8] and Small Baseline Subset (SBAS) [9], it is possible to generate displacement time series that allow for the analysis of subsidence evolution associated with aquifer exploitation [10,11]. These methodologies have proven particularly useful in urban environments, where the abundance of stable reflectors improves radar signal coherence [11,12].

GPS (Global Positioning System), as the most widely used constellation within the broader GNSS framework, provides positioning based on continuous monitoring stations and is widely applied in hydrological studies [13]. It offers high temporal resolution (sub-daily), making it particularly sensitive to local fluctuations in hydrological loading. InSAR, on the other hand, provides extremely high spatial resolution, allowing the mapping of ground deformations on the order of millimeters. In addition to monitoring crustal deformations, the GRACE and GRACE-FO (Gravity Recovery and Climate Experiment–Follow-On) missions allow for the monitoring of variations in water storage on a regional scale (~300 km) with monthly temporal resolution, ideal for characterizing medium and long-term hydrological events [14].

GRACE data in integrated analysis allows for the direct linking of surface deformation with changes in mass storage [15]. This integration has been particularly relevant in studies of subsidence induced by groundwater extraction and drought.

Recent advances in data assimilation, statistical reconstruction, and machine learning have expanded the spatial detail of terrestrial water storage estimates derived from GRACE. These approaches combine GRACE/GRACE-FO observations with additional information from hydrological models and other ancillary datasets, allowing for the reconstruction of terrestrial water storage at more refined spatial scales than the effective resolution of the original observations [16,17,18,19]. However, the evaluation of these products at regional and sub-basin scales still depends on the availability of independent observations, whose spatial coverage can vary considerably between regions. In this context, conventional GRACE mascon solutions continue to provide a consistent satellite-based benchmark for assessing the variability in terrestrial water storage at a regional scale. The approach uses InSAR to provide the spatial distribution of relative deformation, GPS to validate and complement absolute measurements, and GRACE to interpret these deformations in terms of variations in water mass [20]. This combination facilitates the development of geodynamic models that relate hydrological loading to the elastic or inelastic response of the Earth’s crust [21].

Recently, joint inversion approaches have been developed that integrate these three data sources (e.g., [22,23]), allowing for a more precise estimation of groundwater loss and its impact on ground deformation. In addition, the integration of these independent datasets offers a unique opportunity to investigate the spatiotemporal relationships and dynamics between drought indicators, water storage changes, and crustal deformation. Previous studies have demonstrated the potential of combining GPS, InSAR, and GRACE to quantify groundwater depletion, aquifer recharge, and hydrological loading effects. However, integrated analyses of these observations associated with local hydrogeological characteristics remain limited in Brazil, particularly in complex socio-hydrological systems such as the Cantareira System.

The main objective of this study is to conduct an integrated, multi-sensor geodetic assessment of drought intensity and hydrological dynamics in the Cantareira System. By jointly analyzing deformation, water storage anomalies, meteorological drought indices, and reservoir data, we aim to improve the understanding of drought processes and support sustainable water resource management in southeastern Brazil.

2. Materials and Methods

2.1. Study Area

The study area comprises the Piracicaba, Capivari and Jundiaí river basins (PCJ), as well the Upper Tietê river basin (UT), where the reservoirs of the Cantareira System are located (Figure 1). This region is characterized by complex hydrogeological settings, strong seasonal climate variability, and intense anthropogenic pressures related to urbanization, industry, and agriculture.

The UT and PCJ basins constitute the central framework of the water supply system for the MRSP. The UT basin encompasses most of the MRSP, a highly urbanized and industrialized region, resulting in high-water demand. The PCJ basin, located upstream of the Tietê River, plays a critical role in surface water production and abstraction, supplying approximately 50% of the water consumed in the MRSP [24].

To address the growing water demand in the MRSP, the Cantareira System was implemented starting in the 1970s. It represents one of the largest interbasin water transfer systems worldwide, comprising five reservoirs (Jaguari–Jacareí, Cachoeira, Atibainha, Paiva Castro, and Águas Claras) covering a total area of approximately 2300 km² (Figure 1). The first four reservoirs capture water directly from the PCJ basin, which is subsequently transferred to the UT basin through a network of tunnels, canals, and pumping stations [24].

As a result, the PCJ basins are responsible for supplying water to more than 14 million people, including approximately 6 million within the basin itself and about 9 million in the MRSP. Within the PCJ basin, domestic use accounts for the largest share of water consumption (64%), followed by industrial use (26%) and irrigated agriculture (5%) [25]. In the MRSP, 47% of the total water demand is met by the Cantareira System, increasing to 65% within the city of São Paulo.

From a hydrogeological perspective, the PCJ basin is predominantly underlain by sedimentary units of the Paraná Basin, including sandstones and siltstones, as well as basalts of the Serra Geral Formation [26]. These lithologies are associated with well-developed Oxisols and Ultisols, in addition to Entisols in steeper terrains [27]. Such conditions favor the development of porous aquifer systems with high storage capacity, as well as fractured aquifers within basaltic units. In contrast, the UT basin is located within the Atlantic Orogenic Belt, where Precambrian crystalline rocks (granites, gneisses, and migmatites) predominate. Soils in this region are generally shallower and more heterogeneous, mainly Inceptisols and Ultisols, which restrict deep infiltration and promote fractured aquifers characterized by lower productivity and high spatial variability.

In morphoclimatic terms, both basins exhibit a humid tropical to subtropical highland climate, with mean annual precipitation ranging from 1200 to 1600 mm, concentrated during the austral summer. However, the PCJ basin displays a transitional geomorphological setting, consisting of gently undulating plateaus interspersed with prominent basaltic cuestas, which favor regional groundwater recharge. In contrast, the UT basin is characterized by a more rugged and compartmentalized relief, including the Serra do Mar and the Planalto Paulistano, combined with intense urbanization, high surface impermeability, and rapid hydrological responses. These contrasting features result in significant differences in water availability and groundwater storage dynamics between the two basins.

2.2. Dataset

2.2.1. GPS Vertical Displacement

Daily positioning time series from four GPS stations located within the study area (Figure 1) were used for the period 2020–2026. In this study, we employed the daily vertical component solutions provided by the Nevada Geodetic Laboratory (NGL) [28], which have been widely demonstrated to be suitable for detecting geophysical signals, particularly elastic deformation associated with hydrological loading processes. Hereafter, GPS-derived vertical displacement is denoted as GPS-VD.

The GPS solutions were generated using Precise Point Positioning (PPP) with ambiguity resolution and include corrections for solid Earth tides, ocean tidal loading, pole tide, tropospheric and ionospheric effects, as well as antenna phase center variations. The processing was carried out by the NGL using the GipsyX software (version 1.0), and the resulting coordinates are linked to the IGS20 reference frame.

Further details regarding the processing strategy and applied corrections can be found in the official NGL documentation (https://geodesy.unr.edu/gps/ngl.acn.txt, accessed on 15 April 2026).

2.2.2. InSAR Data and Processing

The area of interest was covered by a total of 200 SAR images, used as input for each processing step in different frames. These C-band images were acquired as a Single Look Complex (SLC) product, containing the interferometric phase, with a frequency of 5.405 GHz corresponding to a wavelength of 5.55 cm. The SAR images were obtained from the Sentinel-1 active radar SAR sensor in a 156° downward relative orbit, measured line of sight (LOS), using the interferometric wide (IW) mode, which is Sentinel-1’s primary ground acquisition mode. The data were acquired using TOPSAR (Terrain Observation by Progressive Scanning SAR) [29], a SAR imaging technique that uses the satellite antenna to generate high-bandwidth images with a 250 km swath and a spatial resolution of 5 m × 20 m during acquisition. However, for processing purposes, a spatial resolution of 90 m × 90 m was used in range and azimuth.

Table 1. Features of SAR images.

Sensor	S1
Number of dates	81
Start date	4 January 2020
End date	1 December 2025
Mode	IW
Relative orbit	126
Orbit direction	Descending
Wavelength (m)	0.055
Number of looks range	20
Number of looks azimuth	5
Applied filter	Goldstein 0.50

presents the features of the Sentinel 1 images used.

InSAR processing was carried out using the Parallel Small Baseline Subset (P-SBAS) algorithm, an advanced implementation of the SBAS approach that incorporates parallel computing strategies for the efficient handling of large multitemporal SAR datasets. Images acquired by the Sentinel-1 satellite were used, ensuring adequate temporal coverage and interferometric coherence. The processing workflow included the selection of interferometric pairs with small spatial and temporal baselines, precise co-registration of SAR images, generation of differential interferograms, and removal of the topographic phase using a digital elevation model (DEM). Subsequently, adaptive filtering techniques were applied to reduce noise and enhance coherence, followed by phase unwrapping and the estimation of LOS displacements. To ensure the reliability of the retained measurement points, only pixels with an average temporal coherence greater than 0.8 were selected for the time-series inversion, restricting the analysis to observations with high phase stability and reducing the influence of decorrelated or noisy pixels, particularly relevant given the partially vegetated conditions of the study area. Importantly, this coherence threshold was applied as part of the InSAR processing workflow, prior to the subsequent analyses. Therefore, all InSAR measurement points retained in the resulting deformation time series and used in the EOF/PCA analysis already had an average temporal coherence greater than 0.8. Therefore, the points presented in those figures correspond to the measurement points already retained during the P-SBAS processing. The P-SBAS approach [30,31] enables a robust and efficient time-series inversion, providing deformation time series and mean velocity maps, while also mitigating atmospheric effects through spatio-temporal filtering. This methodology has demonstrated high accuracy in detecting millimetric deformations associated with subsidence, making it particularly suitable for regional-scale studies and long-term deformation analysis [32,33,34].

2.2.3. GRACE/GRACE-FO Data

We used the latest Release 06 (RL06) GRACE/GRACE-FO mascon (mass concentration) solutions from three independent processing centers: the Center for Space Research (CSR) at The University of Texas at Austin, the Jet Propulsion Laboratory (JPL), and the Goddard Space Flight Center (GSFC). All three products represent variations in terrestrial water storage (TWS), expressed as anomalies relative to the same mean reference field over the period 2004.000 to 2009.999, commonly referred to as terrestrial water storage anomalies (TWSA). The three products are distributed on different output grids (0.25° × 0.25° for CSR and 0.5° × 0.5° for JPL and GSFC). The mascon approach discretizes the Earth’s surface into finite spatial elements within which mass variations are directly estimated, significantly reducing the effects of spatial filtering, signal attenuation (leakage), and noise amplification that commonly affect spherical harmonic solutions, thereby improving the effective spatial resolution and hydrological interpretability of the data.

All three mascon solutions include corrections for solid Earth tides, ocean tides, polar motion tides, non-tidal atmospheric and oceanic mass variations (AOD1B correction), as well as glacial isostatic adjustment (GIA). The resulting fields represent the integrated contribution of the main continental water reservoirs, including soil moisture, surface water, groundwater, snow (when present), and water stored in vegetation. The dataset consists of 72 monthly solutions per center, covering the period from January 2020 to December 2025, and preserving the nominal spatial resolution of approximately ~300 km, consistent with the resolving capability of the GRACE and GRACE-FO missions.

We generated regional TWSA time series from the three GRACE mascon solutions (CSR, JPL, and GSFC). Although the three exhibit similar seasonal and interannual patterns, consistent with the regional hydrological signal, they also display non-negligible systematic differences among themselves (Figure A1). To quantify the relative uncertainty of each solution in the absence of an independent external reference, the Three-Cornered Hat (TCH) method was applied [35]. The three regional series were first aligned to a common temporal period, over which the variances of the pairwise differences between products were computed, providing the basic input for the method.

Unlike the classical TCH, which assumes zero cross-covariances between the errors of the different products, the extended/generalized variant was adopted [36], which estimates these covariances through constrained optimization, seeking the solution most compatible with low correlation between errors without artificially imposing it. This choice is justified because CSR, JPL, and GSFC do not constitute truly independent measurements, as they are derived from the same raw GRACE/GRACE-FO inter-satellite range observations, making it plausible that part of their error shares a correlated component.

The results of the extended TCH indicate relative uncertainties of 2.1 cm, 2.4 cm, and 2.8 cm for CSR, JPL, and GSFC, respectively. Given that CSR presents the lowest estimated uncertainty among the three solutions, GRACE-based TWSA calculations were subsequently linked to this solution.

For each basin, the regional TWSA time series was computed as the area-weighted average of the mascon grid cells whose centers fall within the basin boundary, following the approach of [37], with cell areas computed on the GRS-80 reference ellipsoid.

2.2.4. Groundwater Variations, Reservoir Storage, and Drought Indicators

Piezometric data from four monitoring wells were used to characterize groundwater-level variability in the study area. All wells are located within the PCJ basin (Figure 1), and the data were obtained from the São Paulo State Water Agency. The monitoring records consist of ground-based observations of groundwater levels collected from January 2020 to December 2025. According to the operational characteristics of the São Paulo piezometric monitoring network, groundwater levels are measured in monitoring wells using manual water-level measurements. The available records contain multiple observations per month, with the number of measurements varying among months. To obtain a consistent monthly time series, all available measurements within each month were averaged for each monitoring well. No piezometric records were available for the Upper Tietê basin. Consequently, these ground-based observations were used solely to support the hydrological characterization of the PCJ basin.

Storage data for the Cantareira System reservoirs were obtained from the monitoring database of the National Water and Basic Sanitation Agency (ANA). The dataset comprises monthly time series of reservoir storage for the system’s main reservoirs, namely Jaguari–Jacareí, which are operated as an integrated storage unit, Cachoeira, Atibainha, and Paiva Castro, spanning January 2020 to June 2025 (Figure 1). Of these four reservoirs, three (Jaguari–Jacareí, Cachoeira, and Atibainha) are located within the PCJ basin, while only Paiva Castro lies within the Upper Tietê basin. These records capture the temporal variability in the water volume stored in each reservoir and were used to characterize changes in surface-water storage throughout the study period.

As an ancillary dataset, the Brazilian Drought Monitor (https://monitordesecas.ana.gov.br/, accessed on 15 April 2026) provides a standardized and continuously updated assessment of drought conditions across the country, including the state of São Paulo. The monitoring framework integrates multiple drought indicators, including the Standardized Precipitation Index (SPI), the Standardized Precipitation–Evapotranspiration Index (SPEI), the Standardized Runoff Index (SRI), and soil and vegetation moisture indicators, which capture precipitation deficits and combined anomalies of precipitation and atmospheric evaporative demand, respectively [38,39]. In this study, we employed the S2 index for the state of São Paulo, which represents the percentage of the state’s area classified as experiencing severe drought conditions, corresponding to SPI/SPEI values between $-1.3$ and $-1.6$. The S2 index was used to characterize the temporal evolution of severe drought conditions throughout the study period. Although the index is available at the state level, its use is justified to represent the PCJ and Upper Tietê basins, since these basins encompass a substantial portion of the hydrologically monitored areas of the state and include the main headwaters that supply the MRSP. The high density of meteorological stations in these basins minimizes spatial variability, making the index at the state level a reliable indicator of local drought dynamics. For the purposes of this study, the index was used only to allow comparison with other datasets.

A summary of the datasets used in this study, including their spatial and temporal resolution and specific applications, is presented in

Table 2. Summary of the data sets used and its application.

Variable	Spatial Res.	Temporal Res.	Source
GPS-VD	point	daily	NGL
LOS displacements	90 m	11–24 days	ESA
Terrestrial Water Storage	~300 km	monthly	CSR
Groundwater Level	point	monthly	SPWA
Reservoir storage	point	monthly	ANA

.

2.3. Integrated Hydrological Analysis Based on Geodetic Signals

We estimated long-term linear trends, as well as the annual amplitudes and phases, for the four GPS stations using a least-squares adjustment. Prior to this analysis, outliers in the daily coordinate time series were identified and removed using the robust Median Absolute Deviation (MAD) estimator. To reduce high-frequency noise and adjust the temporal resolution of GRACE’s terrestrial water storage, the daily observations were subsequently averaged to obtain monthly time series. To isolate deviations from normal hydrological conditions, a climatology was calculated for each station using the median, which is less sensitive to extreme values. Residual time series were then derived by subtracting the climatological signal, providing deformation anomalies relative to the expected crustal response under typical hydrological conditions in the study region (Figure 2).

To analyze the hydrological signal present in the vertical deformation time series derived from InSAR for each watershed, a spatial averaging strategy was adopted. A radius of influence centered on the barycenter of each basin was defined, and all InSAR measurement points located within this radius were selected. An average time series was then computed to represent the characteristic vertical deformation behavior of each basin. The radius defined with respect to the barycenter differed between basins according to their geometry and spatial extent. A radius of 33 km was adopted for the PCJ basin, whereas a radius of 22 km was used for the Upper Tietê basin. For the PCJ basin, 81 observations were obtained between January 2020 and December 2025, while for the Upper Tietê basin, 117 observations were available between January 2020 and November 2025. To emphasize the hydrological component of the deformation signal, the long-term linear trend was removed from the InSAR time series, assuming that this component is mainly associated with tectonic processes or long-term subsidence unrelated to seasonal hydrological forcing.

Groundwater dynamics were assessed through observations of water levels in piezometric monitoring wells. Well data were processed to remove outliers and inconsistencies, and monthly anomalies were calculated relative to long-term median levels in each well. These piezometric variations were used as an independent indicator of changes in groundwater storage and were compared with both crustal deformation signals and satellite-derived hydrological variables.

Drought conditions during the study period were characterized using drought monitoring records based on precipitation, soil moisture, temperature, and other hydroclimatic variables. In addition, historical drought reports and regional hydrological bulletins were analyzed to identify major drought events and their temporal evolution. These records allowed the classification of drought phases and facilitated comparison with observed variations in crustal deformation, groundwater levels, and terrestrial water storage.

An integrated analysis was conducted to investigate the relationships between drought indicators, groundwater level variations, reservoirs storage, water storage anomalies, and crustal deformation. Cross-correlation analysis was applied to quantify the temporal relationships between GPS-VD, InSAR-derived deformation, GRACE-TWSA, drought indices, and groundwater level variations. For each variable pair, the Pearson cross-correlation coefficient was computed for lags ranging from $-12$ to $+12$ months, and the optimal lag was defined as that maximizing the correlation coefficient. This approach allowed the identification of potential time delays between meteorological forcing, hydrological responses, and the elastic deformation of the crust.

To decompose the spatiotemporal variability in the detrended InSAR deformation fields, an Empirical Orthogonal Function (EOF)/Principal Component Analysis (PCA) was applied separately to the PCJ and Upper Tietê basins. All EOF/PCA calculations were performed using the InSAR deformation time series retained after the P-SBAS processing and coherence thresholding described in Section 2.2.2. Thus, all points included in the analysis had an average temporal coherence > 0.8. For each basin, the data matrix was arranged as space × time, using the detrended InSAR deformation (mm) at each cell. Only temporal centering was applied, without standardization, so the decomposition is based on the covariance matrix and the resulting EOF spatial patterns retain physical units (mm). Pixels with any missing observation across the full time series were excluded from the analysis. This decomposition expresses the deformation field as a set of orthogonal spatial patterns (EOFs) and their associated temporal coefficients (PCs), ranked according to the fraction of variance explained. Modes explaining less than 2.5% of the total variance were not retained for interpretation. Accordingly, the first four modes (PC1–PC4) were retained, jointly accounting for 72% of the variance in Upper Tietê (58%, 7%, 4%, 3%) and 66% in PCJ (45%, 9%, 8%, 4%). This criterion allows the dominant regional hydrological signal (PC1) to be separated from higher-order modes associated with more localized or short-term processes. Information contained in higher-order modes was considered to represent noise rather than coherent deformation signal.

Finally, the procedures described above were combined into an integrated multi-sensor assessment of drought conditions in the Cantareira System. This assessment jointly considers deformation observations from GPS and InSAR, terrestrial water storage anomalies from GRACE, groundwater level variations, reservoir storage data, and meteorological drought indicators to characterize drought phases and their associated hydrological deficits. Figure 3 summarizes the input datasets, the preprocessing applied to each of them, the spatial and temporal scales used for comparison, the statistical analyses performed, the criteria used to delineate drought phases, and the resulting outputs of this integrated assessment. This approach enables the evaluation of drought conditions from complementary hydrological and geodetic perspectives, providing insights into regional water dynamics and supporting improved monitoring and management of water resources in the Cantareira System.

3. Results

The integrated analysis of GPS, InSAR, GRACE, groundwater, reservoirs and drought records reveals consistent spatiotemporal patterns of hydrological variability and associated crustal deformation across the Upper Tietê and PCJ basins during the period 2020–2025. The GPS-VD exhibit marked spatial variability in long-term trends (

Table 3. Linear trend, amplitude and annual phase of GPS time series.

	Latitude	Longitude	Trend	Annual Amplitude	Annual Phase
Station	(Deg.)	(Deg.)	(mm/Year)	(mm)	(Deg.)
EACH	−23.482	−46.500	0.11 ± 0.04	7.39 ± 0.12	52.811 ± 0.898
POLI	−23.556	−46.730	1.00 ± 0.05	6.84 ± 0.19	53.506 ± 1.617
SPBP	−22.926	−46.534	−0.28 ± 0.07	6.69 ± 0.21	63.010 ± 1.820
SPC1	−22.816	−47.063	0.06 ± 0.02	6.27 ± 0.09	58.986 ± 0.840

and Figure 2). The POLI station shows a pronounced uplift (1.0 mm/yr), while EACH and SPC1 display weak positive trends. In contrast, the SPBP station presents a negative trend, indicating subsidence. Despite these differences, the seasonal components are remarkably consistent across all stations, with annual amplitudes ranging between ~6.3 and 7.4 mm and phases clustered within a narrow interval (~53–63°), indicating a coherent seasonal signal across the study area. The temporal evolution of the residual time series (Figure 2) shows recurrent oscillations with similar timing, suggesting common regional forcing.

The GRACE-TWSA time series extracted for the PCJ and Upper Tietê basins exhibit strong interannual variability characterized by quasi-seasonal oscillations (Figure 4). Periods of strong negative anomalies are observed around 2021–2022 and again during 2024–2025, while positive anomalies peak during 2023. Both basins display a high degree of temporal coherence, consistent with the fact that they lie within, and largely share, the same ~300 km GRACE footprint. As such, the two series should be interpreted as regional-scale indicators of water storage variability across the broader Cantareira System rather than as independently resolved, basin-specific estimates. Within this regional signal, the extracted series nonetheless differ in magnitude, with the PCJ pixel-average showing larger negative excursions, reaching values close to $-400$ mm, whereas the Upper Tietê pixel-average exhibits comparatively attenuated variations. In terms of long-term behavior, the Upper Tietê series shows near-stable to slightly positive tendencies, while the PCJ series presents a persistent negative trend over the analyzed period.

A cross-correlation analysis was conducted between the detrended GPS-VD and the GRACE-TWSA extracted at each basin. The linear trend was removed from each GPS-VD series prior to the analysis to exclude possible long-term tectonic contributions from the hydrologically driven signal of interest. Following the convention, positive lags indicate that GPS-VD leads GRACE and negative lags indicate that GRACE leads GPS-VD, over a tested range of $\pm 6$ months. Statistical significance was assessed with a two-tailed Student t-test on the Pearson correlation coefficient, evaluated at $\alpha =0.05$; the test confirmed that the correlations obtained at both basins are statistically significant.

At Upper Tietê, the common analysis period spans 70 months (January 2020 to December 2025). The correlation peaks at a lag of $-1$ month ($r=-0.80$), indicating that GRACE TWSA leads the GPS vertical response by approximately one month. At PCJ, over a common period of 50 months (January 2020 to February 2024), the strongest correlation occurs at zero lag ($r=-0.64$) (Figure A2). In both basins, the negative sign of r is consistent with the expected elastic response to surface loading, whereby an increase in terrestrial water storage (positive TWSA) is accompanied by a downward (negative) vertical displacement at the GPS stations. The near-zero lag obtained at both basins (0 to $-1$ month) indicates an essentially immediate elastic response of the crust to hydrological loading, consistent with the monthly temporal resolution of GRACE. The comparatively stronger correlation at Upper Tietê than at PCJ, despite both basins being sampled within the same ~300 km scale GRACE footprint, more likely reflects differences in the local GPS deformation response than a genuine sub-basin difference resolved by GRACE itself.

The detrended InSAR time series for both basins reveals clear temporal variability that aligns with hydrological conditions (Figure 5). Periods characterized by reduced water storage correspond to positive deformation (uplift), whereas intervals of increased water availability coincide with negative deformation (subsidence). The temporal evolution of deformation follows a similar pattern to TWSA and drought records, with pronounced anomalies during dry periods and reduced variability during wetter phases. The amplitude of deformation varies between basins, with the Upper Tietê basin showing stronger signals compared to the PCJ basin.

Groundwater level anomalies in the PCJ basin exhibit a clear temporal evolution consistent with the other hydrological indicators (Figure 6). Between 2020 and mid-2022, groundwater levels show a progressive decline, reaching minimum values close to $-1$ m. A rapid increase is observed from late 2022 to 2023, with positive anomalies exceeding 1.5 m. Subsequently, during 2023–2024, groundwater levels fluctuate around near-equilibrium conditions, followed by a renewed decline toward 2025. These variations display temporal correspondence with both TWSA (Figure 4) and InSAR-derived deformation signals (Figure 5).

The correlation results between GRACE-TWSA and reservoir storage in the Cantareira System, suggesting that regional variations in terrestrial water storage are reflected in the dynamics of the reservoirs that constitute the main storage component of the Cantareira System (Figure 7). This spatial pattern is particularly significant considering the operational structure of the system: Jaguari–Jacareí, Cachoeira, and Atibainha are located in the PCJ basin and constitute the main upstream storage component of the system, while Paiva Castro is located in the Upper Tietê basin and receives water transferred from the PCJ reservoirs through a system of tunnels and canals. These four reservoirs are responsible for flow regulation and water transfer from the PCJ basin to the Upper Tietê basin. In contrast, the Águas Claras reservoir (Figure 1), although physically connected to the system and located downstream of Paiva Castro, serves an operational function of storage and regulation before the water is transported by gravity to the Water Treatment Plant.

The EOF/PCA decomposition of the InSAR deformation fields reveals a hierarchical structure in both basins (Figure 8 and Figure 9). In the Upper Tietê basin, the first mode explains approximately 58% of the total variance and exhibits a spatially coherent pattern across most of the basin (Figure 8). The associated temporal component shows a smooth long-term evolution with superimposed oscillations. The second and third modes explain smaller fractions of the variance (~7% and ~4%, respectively) and present more heterogeneous spatial patterns, while their temporal components are dominated by oscillatory behavior without clear long-term trends. The fourth mode explains a marginal fraction (~3%) and is characterized by highly localized spatial features and irregular temporal variability.

In the PCJ basin, the first mode explains approximately 45% of the total variance and presents a relatively coherent spatial pattern, although with more pronounced internal contrasts (Figure 9). Its temporal component exhibits a clear long-term trend. The second and third modes (~9% and ~8%) display increasingly fragmented spatial patterns and temporal variability dominated by short-term oscillations. The fourth mode (~4%) accounts for a small portion of the variance and is characterized by localized spatial features and irregular temporal behavior.

4. Discussion

The results show a strong correlation between hydrological variability and crustal deformation, reflecting the sensitivity of the Cantareira System to both climatic forcing and anthropogenic influences. This coherence, however, manifests itself differently between the two basins analyzed, indicating that the integration of different geodetic observations, associated with hydrogeological features, provides a robust perspective for understanding the dynamics of droughts in the region.

These results indicate that the GRACE-TWSA signal is particularly consistent with the variability in storage at Jaguari–Jacareí and Cachoeira, suggesting that regional variations in terrestrial water storage are reflected in the dynamics of the reservoirs that constitute the main storage component of the Cantareira System. This spatial pattern is particularly significant considering the operational structure of the system: Jaguari–Jacareí, Cachoeira, and Atibainha are located in the PCJ basin and constitute the main upstream storage component of the system, while Paiva Castro is located in the Upper Tietê basin and receives water transferred from the PCJ reservoirs through a system of tunnels and canals. These four reservoirs are responsible for flow regulation and water transfer from the PCJ basin to the Upper Tietê basin. In contrast, the Águas Claras reservoir (Figure 1), although physically connected to the system and located downstream of Paiva Castro, serves an operational function of storage and regulation before the water is transported by gravity to the Water Treatment Plant.

The temporal relationship between GRACE-TWSA and reservoir storage reinforces this spatial interpretation. The one-month lag observed for the PCJ basin is physically consistent with the hydrological response of the basin, since changes in regional water storage associated with precipitation and surface runoff can take time to propagate through the hydrological system and reach the reservoirs. From a geodetic point of view, the strong correlation observed for the PCJ reservoirs, particularly Jaguari–Jacareí, is therefore especially relevant. The Cantareira System represents a large-scale interbasin water-transfer system, in which most of the water supplied to the São Paulo Metropolitan Region originates in the PCJ basin and is subsequently transferred to the Upper Tietê basin. According to the Brazilian National Water and Basic Sanitation Agency, of the approximately 33 m³ s⁻¹ produced by the system, only about 2 m³ s⁻¹ is produced in the Upper Tietê basin, while approximately 31 m³ s⁻¹ originates in the PCJ basin, with Jaguari–Jacareí contributing about 22 m³ s⁻¹. Therefore, the high GRACE-TWSA correlation with Jaguari–Jacareí, along with the one-month lag observed for the PCJ basin, provides independent geodetic evidence that satellite gravimetry can detect regional variations in water storage associated with the main upstream source of the Cantareira System. The different temporal behavior observed in the Upper Tietê basin, however, suggests a more complex relationship between regional terrestrial water storage and reservoir volume. The negative lag indicates that variations in reservoir storage may precede the corresponding GRACE-TWSA signal, which likely does not solely reflect the local hydrological response of the reservoir watershed. Instead, this behavior may be influenced by the operational management of the Cantareira System, particularly by the transfer of water from the PCJ basin to the Upper Tietê basin. Thus, storage in the Upper Tietê reservoirs is not determined exclusively by local water availability and natural hydrological inputs, but also by the managed redistribution of water through the interbasin transfer system. This distinction highlights the importance of considering both natural hydrological processes and water-management operations when interpreting GRACE-derived signals in highly regulated and interconnected watersheds.

In contrast to the Upper Tietê basin, where reservoir storage is shaped by both natural hydrological inputs and interbasin water-management operations, the PCJ basin, where crystalline rocks and fractured aquifers predominate, exhibits a response that is mostly elastic and strongly controlled by regional climatic variability. This indicates a high degree of coherence between InSAR and the regional GRACE-derived TWSA signal, with inverse response and near-zero delay, suggesting an efficient coupling between regional water storage variations and locally resolved crustal deformation. The SPBP station, located in this basin, differs from the other GPS stations used in this study by presenting a structural control dominated by a fractured aquifer system, which results in a slightly negative linear trend and the largest phase difference observed between the stations analyzed.

Agricultural activity places a significant demand on the PCJ basin [40]. Water extraction is cyclical and linked to irrigation periods, resulting in an annual response of InSAR deformations through seasonal soil contraction and expansion. Hydrologically, this region exhibits a pronounced interannual behavior of TWSA (strong negative trend; see Figure 4 for more details), consistent with the regional-scale water storage deficits associated with multi-year droughts documented by GRACE-based studies elsewhere (e.g., [41,42]); while GRACE alone cannot isolate this deficit as specific to the PCJ sub-basin, the negative trend is corroborated at the sub-basin scale by the InSAR and GPS deformation records discussed above. This scenario in the PCJ basin is consistent with a prolonged water storage deficit, characterized by persistent negative TWSA and sustained groundwater decline. However, the six-year GRACE record available for this study is insufficient to establish whether this deficit represents an early stage of a longer-term “water bankruptcy” condition [43] or falls within the range of normal interannual-to-decadal hydrological variability. Longer GRACE time series and continued piezometric monitoring will be required to determine whether the system is failing to recover to pre-drought storage levels.

In the Upper Tietê basin, in turn, the hydrological response is conditioned by intense urbanization and continuous groundwater exploitation. Outside the Northeast Brazil (semi-arid), the state of São Paulo has the highest density of tubular wells for groundwater extraction in Brazil [44], which may intensify these hydrogeodetic responses. Located in the São Paulo Sedimentary Basin (surrounded by crystalline environments such as the Cantareira and Coastal montain ranges), the POLI and EACH stations show greater seasonal amplitudes, consistent with the greater storage capacity of the porous media. However, the positive linear trends of InSAR deformations (PC1; see Figure 8a and Figure 9a for more details), indicative of water loss in the medium and long term [45], show magnitudes lower than those observed in the PCJ basin (Figure 9), suggesting a partial decoupling between hydrological variations and deformational response in highly urbanized environments.

Soil impermeability, the presence of underground infrastructure, and the artificial management of water flows tend to reduce infiltration and the effective variability in storage, attenuating the expression of the geodetic signal [46]. In this context, although the system is also subject to water deficit, its deformational response is dampened. EOF analysis reinforces that, despite the dominance of a common regional signal, higher-order modes capture the spatial heterogeneity associated with local controls, including geological structure and water use [13]. In fact, the higher spatial resolution of InSAR observations allows for the additional detection of signals that occur at very short spatial wavelengths, such as deformations corresponding to variations in water storage in shallow aquifers [47].

Analyzing the responses of the different datasets to the occurrence of droughts in the Cantareira System, we observed that the record of severe drought (Figure 5) is characterized by the sensors used, with the intensity of this drought reflected in the amplitude of these observations. With the onset of the drought in mid-2020, the following can be consistently observed: a decline in well levels in the PCJ basin, crustal uplift (InSAR and GPS) in both basins, and a negative TWS anomaly.

The temporal behavior of PC3 is consistent with the intensity of the 2021/2022 drought in the Upper Tietê Basin, where the integrated InSAR signal shows a statistically significant correlation with the drought indicator ($r=0.64$). This relationship is not statistically significant in the PCJ Basin, suggesting that the InSAR decomposition captures basin-dependent responses to drought-related hydrological variability.

The drought recorded in the second half of 2024 can also be observed in the geodetic data, although with less intensity and smaller ranges of variation. However, due to the shorter period of occurrence, the response of the observations tends to merge with the response of the drought indicated in 2025, suggesting that both are part of the same event and that the rainy period (water storage recharge) was again insufficient to remove the system from a critical state.

From a methodological perspective, the integration of GPS, GRACE, and InSAR data proves particularly effective for capturing hydrological processes at different spatial and temporal scales. While GRACE provides information on regional variations in total water storage, InSAR allows for the identification of patterns. Detailed spatial data on surface deformation is available, and GPS offers continuous and independent time series for validation. This integrated approach expands the capacity for drought monitoring and assessment of changes in the hydrological system.

Nevertheless, some limitations must be considered, such as the low spatial resolution of GRACE, the susceptibility of InSAR to atmospheric noise and decorrelation problems, as well as the limited coverage of groundwater data, since all four monitoring wells are located within the PCJ basin. The groundwater comparisons presented here therefore validate the PCJ basin exclusively and cannot be extended to the inferred water-storage variations in the Upper Tietê basin. Despite this, the consistency between the different datasets reinforces the robustness of the interpretations and highlights the potential of integrated geodetic techniques to advance the understanding of complex hydrological systems.

5. Conclusions

By integrating geodetic observations and hydrological data (GNSS, InSAR, GRACE/GRACE-FO, groundwater level records, and meteorological indicators), this paper examines the dynamics of drought in the Cantareira Water Supply System. The evidence supporting these conclusions operates at three distinct levels, which we distinguish explicitly below.

At the regional scale (~300 km), the GRACE/GRACE-FO product shows a coherent pattern of terrestrial water storage variability shared by the PCJ and Upper Tietê basins, with pronounced negative anomalies during 2021–2022 and 2024–2025 and a partial recovery in 2023. This regional signal is consistent with documented drought episodes in southeastern Brazil, but GRACE alone cannot resolve whether the two basins respond differently to this common forcing.

At the local scale, independent observations from GNSS, InSAR, and, in the PCJ basin, groundwater wells do resolve sub-basin differences that GRACE cannot. Vertical deformation from GNSS and InSAR shows an inverse, near-zero-lag relationship with the regional TWSA signal, and this coupling is markedly stronger in the PCJ basin than in the Upper Tietê basin, where the response is comparatively muted. Piezometric records, available only for PCJ, corroborate a sustained groundwater decline over the study period consistent with the local deformation and GRACE signals.

The correlation analysis between GRACE-TWSA and reservoir storage in the Cantareira System reinforces this local-scale evidence, revealing a consistent relationship between regional water storage variability and reservoir dynamics, particularly for Jaguari–Jacareí and Cachoeira. These findings suggest that GRACE-TWSA provides an important regional-scale perspective on water storage variability that can complement conventional reservoir monitoring in the assessment of hydrological conditions and drought evolution. At the interpretive level, we attribute the contrast between basins to differences in hydrogeological setting and anthropogenic pressure: the fractured-aquifer, less urbanized PCJ basin appears to respond directly and predominantly elastically to hydrological forcing, whereas the more urbanized Upper Tietê basin shows a muted, partially decoupled response plausibly associated with reduced infiltration and sustained groundwater extraction. These mechanistic explanations are interpretations of the local-scale evidence above, not conclusions independently supported by GRACE.

The PCA/EOF decomposition of InSAR time series separates the dominant regional hydrological signal from higher-order modes associated with more localized processes, evidence that it likewise operates at the local and interpretive levels described above, not at the regional GRACE level. The presented integrated multi-sensor assessment effectively captures the temporal evolution and intensity of drought events during the 2020–2025 period at the regional scale, with the local-scale geodetic and groundwater records indicating signs of incomplete hydrological recovery, particularly in the PCJ basin.

The findings point to a rising vulnerability of the Cantareira System to persistent drought conditions. Regional GRACE data carry spatial-resolution limitations, InSAR is subject to atmospheric noise, and independent groundwater observations remain scarce and geographically restricted to PCJ; nonetheless, the consistency between the regional and local-scale datasets, kept analytically distinct as described above, reinforces the robustness of the overall interpretation. The multi-sensor approach described here presents a robust strategy for drought assessment and water resource monitoring in regions lacking conventional hydrological data. This work emphasizes the potential for remote sensing and geodetic techniques to enhance understanding of complex hydrological systems and support water resource decision-making under increasing climatic and anthropogenic pressures.