SWOT Satellite Nodes as Virtual Stations During the 2024 Extreme Flood in Southern Brazil

Abstract

In 2024, Rio Grande do Sul (RS), Brazil, faced the most severe flood event in its recorded history, which compromised several ground-based hydrological gauges. The SWOT (Surface Water and Ocean Topography) satellite, capable of measuring water surface elevation (WSE) in continental waters, is a valuable tool for providing critical data. This study investigates whether node-level WSE data from the SWOT satellite can effectively function as virtual hydrological stations under such extreme conditions. The study was applied in all of RS state considering 100 in situ gauges and was subdivided into three sections: (i) an evaluation of the variation in SWOTʹs WSE data compared to the variation in in situ levels from telemetric gauges, considering subsequent cycles of passes between July 2023 and April 2025, yielding an MAE = 35 cm and an RMSE = 73 cm after outlier removal; (ii) an evaluation of the variation in SWOTʹs WSE data compared to the variation in telemetric level data, considering one window prior to and another during the extreme event, resulting an MAE = 26 cm and an RMSE = 34 cm; (iii) an analysis of SWOTʹs data availability during the extreme event, when in situ telemetric data were unavailable. The results demonstrate an agreement between the variation observed in SWOT data and that in telemetric gauges in RS, even during extreme events. Moreover, in the absence of in situ data, SWOT was still able to capture WSE data.

1. Introduction

Between 2000 and 2019, there were more than 3254 flood events worldwide, affecting approximately 1.65 billion people and resulting in more than 104,000 deaths, in addition to losses of over USD 651 billion [1]. Brazil is the most susceptible country to flooding in Latin America, ranking among the 15 countries worldwide with the largest population exposed to this risk. Between 2000 and 2019, the country experienced 70 flood-related disasters, affecting more than 70 million people [2].

In 2024, the state of Rio Grande do Sul (RS), located in southern Brazil, experienced the worst meteorological event in its history. Prolonged and intense rainfall between late April and early May caused precipitation levels that reached 40% of the annual average within just two weeks in some locations. This precipitation caused rapid and sharp rises in river levels that continued for weeks, resulting in one of the most significant floods and environmental disasters in the country. This extreme event affected 478 of the 497 municipalities in the state, harming more than 2 million people and causing at least 183 deaths [3,4].

This event occurred due to a combination of meteorological systems acting on different scales. El Niño favored above-average rainfall in southern Brazil. Additionally, warmer-than-average sea surface temperatures in the tropical Atlantic increased atmospheric moisture. On the synoptic scale, a blocking pattern in the mid-latitude Pacific led to atmospheric instability over the study region (RS). During the rainfall episodes, an intense low-level jet transported warm and humid air from the Amazon to RS. At mid-levels, a long-wave trough facilitated the development of surface low-pressure systems. The advance and interaction of cold fronts with moist and warm air masses contributed to the occurrence of severe storms, strong winds, tornadoes, hail, and frequent and intense rainfall [5].

Extreme events, such as the one that occurred in RS—with uninterrupted and intense rains and consequent floods throughout the state—are projected by the Intergovernmental Panel on Climate Change (IPCC) [6] to become increasingly frequent and intense worldwide due to the intensification of the adverse impacts of human-caused climate change. In [7], a global examination of recent changes in the magnitude, frequency, and probability of extreme river floods was carried out based on historical series of rivers, with global changes in 20-, 50-, and 100-year floods showing increases in the magnitude of floods in various parts of Brazil, including in the south, where the increase exceeds 50% in some locations. Additionally, Rio Grande do Sul is susceptible to rapid and intense floods, and it is possible to identify at least one flood in a large river per year in the historical records of the monitoring gauges of the national hydrometeorological network [8].

In scenarios of extreme events, there is an urgent need to implement preventive actions with the population, improve weather forecasting and disaster risk warning systems, develop innovative and accessible flood assessment tools, and implement urban planning strategies that promote more sustainable forms and structures [9,10]. The 2024 floods caused damage to, and in some cases, the loss of, many in situ hydrological monitoring gauges in RS due to river currents and flooding, resulting in gaps and failures in data acquisition. These failures significantly impaired real-time monitoring and hindered timely emergency response and alert issuance [8,11].

In this context, satellite remote sensing offers a valuable alternative for acquiring hydrological data. Among these hydrological variables that can be remotely sensed, water level stands out. This can be obtained through remote sensing using radar altimetry missions (e.g., Sentinel-3 [12,13,14] and Jason-2/3 [13,14,15]) and laser altimetry missions (e.g., ICESat-2 [13,16,17]), thereby assisting in the water management of the location of interest. Laser altimetry satellites, although they provide accurate water surface elevation (WSE), obtain data at discontinuous points on the surface, have limited spatial coverage, and can be affected by scattering due to atmospheric conditions [18,19,20,21,22]. Traditional radar altimetry, on the other hand, provides broader spatial coverage along the satellite’s ground track; however, water elevation measurements are generally limited to the nadir direction, which also results in limited coverage of the Earth’s surface [18,19,20,21,22].

In December 2022, the Surface Water and Ocean Topography (SWOT) satellite was launched, specifically designed to monitor global surface water [23,24]. SWOT provides high-accuracy measurements of WSE in rivers, lakes, wetlands, oceans, and coastal areas through a novel Ka-band radar interferometer (KaRIn), which offers two-dimensional wide-swath coverage (~120 km with a ~20 km nadir gap), a significant improvement over traditional altimeter. The satellite orbits at 891 km altitude with a ~21-day repeat cycle [21,23,24].

As SWOT is a recent satellite, studies addressing its application in floods are still scarce. Some studies around the world compare WSE data from SWOT products with in situ data during floods [10,25], with data from other satellites in both critical and non-critical events [10,21,26], and with in situ and satellite data in wetlands [27]. None of these studies focus on assessing node data during floods. The studies by [10,25] also applied SWOT to the RS disaster in 2024; however, the former evaluated the raster product, and the latter focused on the pixel cloud product. Furthermore, both studies were conducted on a smaller scale, limited to the Guaíba basin, and primarily during the 2024 flood event, using data from two [10] and five [25] in situ gauges, respectively.

This study differs in that it focuses on the vector product of SWOT—specifically, the node data—and analyzes the entire state of RS, incorporating data from one hundred gauges over an extended period, from July 2023 to April 2025. The river vector data are in shapefile format, which is identified as L2_HR_RiverSP. This product is composed of two shapefiles—river reach ~10 km in length and nodes every ~200 m. The attributes from L2_HR_RiverSP include WSE, width/area, slope, derived discharged data, and other information [28].

Given the failure of ground-based monitoring during the 2024 floods and the potential of SWOT to provide wide-coverage water-level data, this study investigates the feasibility of using SWOT-derived node-level water surface elevation data to establish virtual hydrological monitoring stations. These virtual stations could serve as a complement to damaged or missing in situ gauges, particularly in disaster-prone or inaccessible areas.

2. Materials and Methods

Figure 1 outlines the methodological workflow, which is structured in three sequential stages: (i) data acquisition, (ii) data pre-processing, (iii) filtering and comparative analysis between SWOT WSE and telemetric gauge data. Initially (Figure 1a), water surface elevation data were acquired from the SWOT product L2_HR_RiverSP, focusing on nodes, which are points distributed every ~200 m along the river’s centerline. Additionally, in situ water-level data from telemetry gauges located in the rivers of Rio Grande do Sul were obtained for the same period covered by SWOT observations (Figure 1a). The objective was to compare the WSE measurements provided by the nodes with data from multiple nearby telemetric gauges (used as a water-level reference).

Before comparing the data, it was necessary to carry out the pre-analysis described in Figure 1b. The selection of study locations required the presence of nodes from SWOT and telemetric gauges. Only nodes located in proximity to existing telemetric gauges in RS were selected. Once the nodes and gauges were defined, the in situ water-level data were filtered according to the date and time of the SWOT satellite overpasses between July 2023 and April 2025. After this step, it was necessary to analyze the telemetry gauges to detect and remove potential measurement errors or failures in the in situ records. Additionally, the SWOT WSE data were analyzed to identify and filter out outliers. In addition to proximity, a correlation check was performed to ensure hydrological and morphological coherence between the node locations and telemetry gauges, considering factors such as dams, confluences, and bifurcations.

The telemetric gauges of the Brazilian hydrometeorological network lack a standardized vertical reference [29], while the WSEs of the nodes are associated with the Earth Gravitational Model 2008 (EGM2008). Thus, it was decided to evaluate both datasets (WSE and water-level reference) based on the variation in river levels recorded by the telemetry gauges and the variation in WSE measured by SWOT between two different dates, selected according to the subsequent cycles of SWOT passes (Figure 1b.1).

Comparisons with data variations—locations in Figure 2—were carried out in two stages: the first considered the period from 28 July 2023 to 28 April 2025 (Figure 1c.1), and the second focused on the flood event that occurred in Rio Grande do Sul in 2024 (Figure 1c.2). An evaluation was conducted of the data variation from the gauges and the corresponding SWOT nodes with these gauges, both before the event (24 March 2024 to 16 April 2024) and after the event began (14 April 2024 to 7 May 2024).

In addition, a third analysis was carried out (Figure 1c.3) focusing on locations where the time series of river telemetric gauge levels did not contain valid data during the SWOT overpasses (±720 min) from 14 April 2024 to 7 May 2024 (Figure 1b.2 and Figure 2). However, WSE values were effectively captured by SWOT and assigned to nodes, demonstrating SWOT’s capability for monitoring extreme events. The entire methodology is detailed in the following sections.

2.1. Study Site and Data from Telemetric Gauges

The state of Rio Grande do Sul, as shown in Figure 2, is located in the southern region of Brazil, with a population of 10,882,965 people according to the 2022 census. This state comprises 497 municipalities and has a total area of 281,707.150 km² [30]. RS contains two hydrographic regions (HRs) of the twelve presents in Brazil: the HR of Uruguay, which comprises the hydrographic basin of the Uruguay River located within Brazilian territory, and the South Atlantic HR, composed of the hydrographic basins of rivers that flow into the Atlantic—southern section [31]. The South Atlantic HR is subdivided into three hydrographic units (HUs): Guaíba, Coastal/RS, and Coastal/SC-PR. The first two are located within the state of Rio Grande do Sul and the third spans the states of Santa Catarina (SC) and Paraná (PR) [32].

Floods are frequent in both HRs [32]. In the Uruguay HR, floods affect populations along the main course and some of its tributaries and can occur at any time of the year in the lower, middle, and upper stretches of the river. In the South Atlantic HR, floods disproportionately impact low-income populations living in urban areas, primarily due to inadequate occupation of floodplains, lagoons, and riverbanks [32]. The most affected areas in the South Atlantic HR within Rio Grande do Sul include the Guaíba River region, Patos and Mirim Lagoons, where periodic flooding occurs in floodplains, surrounding lagoon systems, and along main watercourses, impacting both urban centers—Pelotas, Porto Alegre, and São Leopoldo—and rural areas [32].

The flooding events that occurred in RS between April and May 2024 had their most significant impact on the South Atlantic HR, particularly the Guaíba HU, where several major rivers reached critical levels, causing severe flooding. These included the Guaíba, Gravataí, Sinos, Taquari, Jacuí, and Caí rivers. During the same 2024 flood in the Uruguay HR, the Uruguay River also exceeded the alert level, causing significant impacts on border communities [33].

In this study, the time series of water levels from the monitoring gauges in Rio Grande do Sul were obtained from telemetric gauges, with data from the national hydrometeorological network of the Brazilian National Water and Sanitation Agency (ANA), available at https://telemetriaws1.ana.gov.br/ServiceANA.asmx (last access on 30 May 2025), spanning from 28 July 2023 to 28 April 2025. In automatic fluviometric (telemetric) gauges, water levels in rivers or reservoirs are measured using pressure, radar, or bubbler sensors, stored in the internal memory of the Data Collection Platforms, and transmitted telemetrically to the ANA corporate database via satellite, GPRS, or radio [29]. These data are collected at the gauges at intervals of 12 h, 1 h, 30 min, or 15 min [34].

At these gauges, the water column levels measured automatically determine the vertical position of the water surface in relation to a predefined reference plane (datum). The datum of the fluviometric gauge is a local zero-altitude mark. It is typically an arbitrary reference, meaning that there is no standardization of the vertical reference for the gauges in the Brazilian hydrometeorological network [29]. With rivers reaching unprecedented critical levels in the 2024 RS flood, several fluviometric gauges suffered significant equipment damage or were washed away by river currents [8,11].

2.2. Data from SWOT

The Surface Water and Ocean Topography (SWOT) mission was developed as a joint initiative between the National Aeronautics and Space Administration (NASA) [23] and the Centre National d’Etudes Spatiales (CNES) [24] with contributions from the Canadian Space Agency (CSA) [35] and the United Kingdom Space Agency [36].

The product used in this work was the SWOT Level 2 River Single-Pass Vector Data Product (L2_HR_RiverSP Version C), specifically the node dataset. The nodes are predefined from a static Prior River Database (PRD), which for SWOT is the SWOT River Database (SWORD) [28]. Since the practical application and consistency of the SWOT vector products depend on a predefined global river network database, SWORD serves as the foundation for these products. It integrates several global datasets, including the Global River Widths from Landsat, MERIT Hydro, HydroBASINS, and the Global River Obstruction Database, providing a standardized framework for rivers worldwide with widths over 30 m [37]. This prior definition of SWOT nodes enables the attribution and association of SWOT observations to these predefined, static nodes.

Each of these nodes in SWOT has attributes such as water surface elevation (WSE), river width (width), water surface area (area_total), quality flags, uncertainty estimates, and other parameters. The measurements assigned to the nodes are calculated using an algorithm that determines which pixels from the pixel cloud product (L2_HR_PIXC) should be assigned to each SWORD node and then aggregates all these pixel measurements appropriately across each node [28,38].

To define the nodes to be analyzed, vector data were acquired using the earthaccess v.0.14.0 Python library. After nodes’ selection, the time series for these nodes, covering the period from 28 July 2023 to 28 April 2025, were extracted using the HydroCORN tool, an API that repackages the SWOT hydrology datasets into CSV or GeoJSON formats [39].

2.3. Pre-Analysis

Given the SWOT satellite’s orbital cycle and the higher temporal resolution of telemetric gauges, it was necessary to restrict gauge data to time windows closely aligned with each SWOT overpass. Between July 2023 and April 2025, SWOT had 29 cycles (~21-day repeat orbit) utilized in this study, with 13 distinct passes over Rio Grande do Sul (Figure 2). These passes were grouped based on similar and consecutive acquisition dates, which naturally resulted in a separation by acquisition mode—ascending or descending—due to the average 11-day interval between them. This grouping led to a total of 58 SWOT coverages across the study period in RS.

Telemetric gauge data are subject to errors related to instrumental failures, environmental factors, maintenance issues, and data transmission problems [40,41,42,43]. The analysis of the RS time series (Figure 1b) revealed the following inconsistencies and corresponding treatments:

(i)

Constant values (≥24h), indicating sensor failure, were removed.

(ii)

Datum shifts were corrected by identifying step changes via k-means (k = 2) [44]; if the smaller cluster (>5%) was continuous, an offset aligned the baselines.

(iii)

Anomalies like spikes were removed by detecting extreme local deviations using sliding windows and local standard deviation.

To link SWOT nodes to gauges, the nearest node within a 1 km buffer was selected per gauge, along with four upstream and four downstream nodes for hydrological analysis. SWOT’s node data can be highly noisy, as noted by previous accuracy assessments [45]. Thus, first we filtered using the bitflag attribute, retaining only nodes with bitflag values below 8,388,608, which exclude observations flagged as outliers (wse_outlier) [28]. Despite the SWOT bitflag, a high level of noise remained (Figure 2). To address this, we applied two filters to the node time series [46]: the interquartile range (IQR) to remove extreme outliers (Equation (1)), and the modified Z-score filter to detect subtler anomalies. Values with an absolute Z-score > 3.5 were considered outliers and removed (Equation (2)).

$$\left[{\mathrm{Q}}_1-1.5·\mathrm{I}\mathrm{Q}\mathrm{R},{\mathrm{Q}}_3+1.5·\mathrm{I}\mathrm{Q}\mathrm{R}\right]$$

(1)

$${\mathrm{M}}_{\mathrm{i}}={\displaystyle \frac{0.675·\left({\mathrm{W}\mathrm{S}\mathrm{E}}_{\mathrm{S}\mathrm{W}\mathrm{O}\mathrm{T}}\left[\mathrm{i}\right]-\overline{{\mathrm{W}\mathrm{S}\mathrm{E}}_{\mathrm{S}\mathrm{W}\mathrm{O}\mathrm{T}}}\right)}{\mathrm{M}\mathrm{A}\mathrm{D}}},\mathrm{E}\left(\mathrm{M}\mathrm{A}\mathrm{D}\right)=0.675·\mathsf{\sigma }$$

(2)

where ${\mathrm{W}\mathrm{S}\mathrm{E}}_{\mathrm{S}\mathrm{W}\mathrm{O}\mathrm{T}}$ is the water surface elevation from SWOT; $\mathrm{M}\mathrm{A}\mathrm{D}$ is the median absolute deviation value; and ${\mathrm{M}}_{\mathrm{i}}$ represents the modified Z-score value.

A temporal window of ±720 min—based on the SWOT overpass time and gauge intervals—was used to match SWOT observations with the nearest filtered telemetric data. Spearman correlation (ρ) was applied to assess agreement, and only nodes showing at least strong correlation (ρ ≥ 0.8) [47] were retained.

Since this study aims to evaluate the potential of SWOT mission nodes as virtual stations, it was necessary to focus on both proximity and similarity in behavior between the two datasets. This step of Spearman correlation was intended to identify factors that may cause spatial and morphological mismatches between SWOT nodes and the locations of telemetric gauges, such as hydrological features (e.g., confluences, bifurcations) and anthropogenic structures (e.g., dams) in the rivers of Rio Grande do Sul.

Due to the absence of a standardized vertical datum for gauges—often based on arbitrary local references, direct comparisons with SWOT’s EGM2008-referenced WSE are not feasible. Therefore, this study focused on water-level variations over time, which remain consistent between both datasets regardless of vertical reference. Thus, these variations (Figure 1b.1) were calculated based on successive cycles from each specific pass shown in Figure 2, ensuring comparable spatial and temporal consistency as well as hydrological conditions.

Additionally, to assess SWOT’s performance during the 2024 flood event, WSE data were extracted between 4 and 7 May 2024, focusing on locations where telemetric gauge data were unavailable in a window of ±720 min (Figure 1b.2). This analysis yielded 131 nodes linked to 36 gauges, demonstrating situations in which SWOT provided valid measurements despite the absence of ground data. For these nodes, variations were calculated based on the previous cycle to understand the flood’s magnitude (Figure 1c.3). Of the 36 matched gauges, only 8 had variations calculated, given the need to integrate two data of SWOT.

2.4. Comparison

We assessed the accuracy of SWOT’s WSE measurements by comparing their variation with the water levels recorded by telemetric gauges, which were considered as the reference. This was performed by calculating the difference between the variations of both datasets for each time window (Equation (3)).

$$∆={∆\mathrm{W}\mathrm{S}\mathrm{E}}_{\mathrm{S}\mathrm{W}\mathrm{O}\mathrm{T}}-∆{\mathrm{W}\mathrm{L}}_{\mathrm{G}\mathrm{a}\mathrm{u}\mathrm{g}\mathrm{e}}$$

(3)

where ${∆\mathrm{W}\mathrm{S}\mathrm{E}}_{\mathrm{S}\mathrm{W}\mathrm{O}\mathrm{T}}$ is the variation in water surface elevation derived from SWOT between two cycles; $∆{\mathrm{W}\mathrm{L}}_{\mathrm{G}\mathrm{a}\mathrm{u}\mathrm{g}\mathrm{e}}$ is the variation in water level obtained from the telemetric gauge for the same period; and $∆$ represents the difference between both of these variations. This approach allows the elimination of vertical reference inconsistencies, focusing exclusively on the consistency of hydrological behavior captured by the datasets.

In addition to this individual metric, two metrics were also used to assess the accuracy of the dataset. These metrics are the mean absolute error (MAE), as defined in Equation (2), and the root mean squared error (RMSE), as defined in Equation (5). MAE and RMSE express errors in the variation in the WSE in relation to the variation in water levels; however, RMSE is more sensitive to discrepant values. We also applied a linear regression analysis between the two variations to analyze the coefficient of determination (R²).

$$\mathrm{M}\mathrm{A}\mathrm{E}={\displaystyle \frac{1}{\mathrm{n}}}\sum _{\mathrm{i}=1}^{\mathrm{n}}\left|∆{\mathrm{W}\mathrm{L}}_{\mathrm{G}\mathrm{a}\mathrm{u}\mathrm{g}\mathrm{e}}\left[\mathrm{i}\right]-{∆\mathrm{W}\mathrm{S}\mathrm{E}}_{\mathrm{S}\mathrm{W}\mathrm{O}\mathrm{T}}\left[\mathrm{i}\right]\right|$$

(4)

$$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=\sqrt{{\displaystyle \frac{1}{\mathrm{n}}}\sum _{\mathrm{i}=1}^{\mathrm{n}}{\left(∆{\mathrm{W}\mathrm{L}}_{\mathrm{G}\mathrm{a}\mathrm{u}\mathrm{g}\mathrm{e}}\left[\mathrm{i}\right]-{∆\mathrm{W}\mathrm{S}\mathrm{E}}_{\mathrm{S}\mathrm{W}\mathrm{O}\mathrm{T}}\left[\mathrm{i}\right]\right)}^2}$$

(5)

The comparisons detailed above were made for two sets of data. The first (Figure 1c.1) aimed to verify the overall consistency between SWOT nodes and telemetry gauges across Rio Grande do Sul. This analysis considered the complete set of 5382 water-level variations obtained over the study period, resulting in a single MAE and RMSE value that reflects the general accuracy of the SWOT measurements relative to the gauge data.

The second dataset (Figure 1c.2) focused on evaluating the capability of SWOT to function as a virtual station under extreme hydrological conditions, specifically the 2024 flood event in RS. Two time windows were analyzed: the pre-event period (24 March 2024 to 16 April 2024) and the period during the event (14 April 2024 to 7 May 2024). This approach enabled a comparison between normal hydrological conditions and flood conditions. For consistency, the analysis included only 10 gauges and 25 nodes with valid data in both time windows.

3. Results

The findings reveal a consistent alignment between variations in SWOT data and those observed in telemetric gauges across RS, even during critical events. Figure 2 presents the distribution of the 100 locations where simultaneous data from SWOT nodes and telemetric gauges were analyzed, focusing on the variations between the WSE captured by SWOT and reference water levels from in situ gauges. The Coastal/RS HU, located in the southern part of the state, had only four of these locations. In contrast, the Uruguay HR and the Guaíba HU accounted for most of the analyzed data.

In Figure 2, the locations with both datasets (before and during the event) are shown in red. The histogram of variation before the event is shown in Figure 3b—window from 24 March 2024 to 16 April 2024—and during the event in Figure 3c—window from 14 April 2024 to 7 May 2024. Although many of the data points in the general analysis—black points in Figure 2—are concentrated in the Guaíba HU, when considering only the overlapping data for these two time windows, the number of locations decreases. In total, only one variation in the Guaíba HU was filtered during the event within the selected timeframe and matched with SWOT overpasses (±720 min). The Guaíba basin was the most affected by the 2024 floods, recording the highest damage observed during the event, which compromised measurements at the telemetric gauges [11,32].

In Figure 3a, the data analysis for the entire period—from July 28, 2023, to April 28, 2025—is presented through a density plot. This graph shows an unbiased distribution, with both positive and negative differences across the 5382 selected datasets. Additionally, in approximately 56% of comparisons, the differences were less than 20 cm between the WSE variation from SWOT and the variation in reference water levels from telemetric gauges. Although only about 6% of the data variation shows absolute differences greater than 100 cm.

The severity of the event is also reflected in the increased discrepancies between SWOT and in situ data, as shown in Figure 3b, which highlights a deterioration in data agreement compared to Figure 3c. When focusing on the 10 locations with data available in both datasets during the extreme event, the histogram (Figure 3b) shows that, before the event, the highest frequency of differences was between −20 cm to 20 cm, with 16 occurrences, while significant differences were isolated. After the event (Figure 3c), the number of occurrences within the −20 cm and 20 cm range dropped to 12, while the remaining data became more widely dispersed, including slightly higher difference values. In the window during the event, there was a greater presence of cases where the variation observed in the in situ gauges was larger than the variation in SWOT data.

Figure 4 presents scatterplots of data variations with a linear regression line, showing the R² as well as the error metrics MAE and RMSE. This analysis covers the entire period—from 28 July 2023 to 28 April 2025—and the two time windows from the second stage: before the event—24 March 2024 to 16 April 2024—and after the event started—14 April 2024 to 7 May 2024. In Figure 4a, for the whole period with 5382 data, the highest densities are concentrated near the line of perfect agreement. An R² value of 0.70 was achieved, indicating a moderate linear relationship between SWOT-derived and in situ variations in WSE measurements. This is supported by the MAE of 35 cm and RMSE of 73 cm.

In the second stage (Figure 4b,c), which focuses on the floods that occurred in Rio Grande do Sul between April and May, the scatterplot in Figure 4b shows the distribution of the 10 datasets—25 nodes matched with 10 gauges—before the event started. In this window, the error is lower than in the full-period analysis, with an MAE of 18 cm, an RMSE of 23 cm, and an R² of 0.94.

After the start of the event (Figure 4c), it is noteworthy that increases in water variation were observed in the field at all locations studied, as expected since 478 of the 497 municipalities were affected by the extreme event. As observed in Figure 3b,c, there is a greater difference between the two datasets, leading to a worsening in the MAE, at 26 cm, and in the RMSE, at 34 cm, with an R² of 0.96 (Figure 4c). However, even with this increase in error, the results still demonstrate a good overall accuracy for the relationship between the variations of SWOT WSE observations and in situ water-level data.

4. Discussion

The results obtained in this study demonstrate that the variations in WSE observed by the SWOT satellite shows agreement with the variations in water levels measured by telemetric gauges in the state of Rio Grande do Sul. During the studied period, from July 2023 to April 2025, favorable relative overall errors and a moderate linear agreement were observed between the two datasets, with an MAE of 35 cm, an RMSE of 73 cm, and an R² of 0.7 (Figure 4a). During the flood event, the MAE was 26 cm and the RMSE 34 cm, with an R² of 0.96 (Figure 4c). These values reflect the relatively good performance of SWOT and highlight its potential for continuous hydrological monitoring.

These findings are consistent with the existing literature on the use of SWOT data during disasters in Rio Grande do Sul. The authors of [25] applied the SWOT Level 2 Water Mask Raster product to this disaster, focusing on the Jacuí–Guaíba–Patos system (180,000 km²) between 14 April and 27 May 2024. In their analysis of differences between in situ level data from five gauges and SWOT WSE data, a correlation of R² = 0.99 was found, considering changes in water levels during the event, with an RMSE of 63 cm and a bias of 14 cm. In [10], the SWOT Level 2 Water Mask Pixel Cloud was used in the metropolitan region of the Guaíba River between January and July 2024. A comparison with in situ observations from two gauges showed an average error of 20 cm and a Pearson correlation coefficient of 98%. Both studies confirm the satellite’s reliability under extreme flood conditions and underscore its value as a decision-support tool in emergencies.

The pixel cloud product may be unwieldy for large-scale or multitemporal analyses due to its complexity, while the raster product lacks information about river network topology. In contrast, the vector product is more suitable for such applications, as it enables multitemporal analysis of river nodes and reaches that consistently represent the same river sections [37]. This capability is particularly valuable for detecting changes in water bodies over time, such as those caused by flood events in RS.

Studies evaluating WSE from SWOT nodes against in situ water levels under non-extreme conditions—such as those also included in the present study—have demonstrated the satellite’s capability to monitor inland water bodies. For example, the authors of [48] focused on 15 in situ gauges in the Middle and Lower Reaches of the Yangtze River in China, where they found that, in some locations, both datasets aligned closely, with R² values exceeding 0.9 and RMSE values of around 25 cm. However, in other areas, the R² values were below 0.8 and the RMSE exceeded 100 cm. In another example, the authors of [49] evaluated 17 nodes along the Chao Phraya River in Thailand, finding an RMSE of 35 cm and an MAE of 33 cm when compared to in situ gauges.

During the extreme flood event in May 2024, SWOT continued to provide WSE data even in locations where telemetric gauges failed. Between May 4 and 8, out of 84 locations with valid SWOT data, 36 had no available records from telemetric gauges in a temporal window of ±720 min of SWOT overpass. In these locations, the satellite provided data for 131 distinct nodes effectively functioning as an alternative monitoring network. This is particularly relevant in regions such as the Guaíba Hydrographic Unit, which experienced the highest accumulated precipitation—exceeding 600 mm in some municipalities—and the most severe flood impacts [5,11,33]. Rivers such as the Jacuí, Taquari, Caí, Sinos, and Gravataí had WSE values recorded by SWOT in areas where ground-based gauges were damaged or destroyed (Figure 5) [11]. In Figure 5, it is possible to observe the maximum variation between the pre-event and after the event starts. Although 36 matched gauges were included, variations could be calculated for only 8 due to the need to integrate two cycles.

Nevertheless, the data revealed an increase in discrepancies between SWOT measurements and telemetric gauge data in the affected regions (Figure 3b,c and Figure 4c). The severity of rainfall and the extreme nature of the event may have directly impacted on the measurement quality. Although SWOT is theoretically resilient to various weather conditions, previous studies have shown that heavy rainfall can affect data quality [38,50,51].

The flags assigned to the data by SWOT themselves support this, as during the event, there was a significant decrease in the proportion of measurements flagged as reliable (flag 1) from 48% to 4%, and a corresponding increase in degraded-quality flags (flag 2) from 36% to 80%. Despite the reduction in data quality under adverse conditions such as heavy rainfall and flooding, the average discrepancies—MAE and RMSE—during the flood period were less than 11 cm compared to those observed before the event (Figure 4b,c).

These findings underscore the robustness and operational potential of SWOT-based analysis for monitoring inland waters given its spatiotemporal resolution—even under extreme hydrometeorological events, as demonstrated both in this study and in previous works [10,21,25,26,27]—making it a valuable tool for flood assessment.

Another possible source of error may stem from the telemetric gauges themselves. During extreme events, the presence of debris, pressure surges, or structural issues can impair the operation of pressure, bubble, or radar sensors [29], thereby compromising the reliability of the reference data. Thus, part of the discrepancies may not be solely attributable to the satellite but also to the instability of ground-based sensors in disaster scenarios.

Additionally, uncertainties may arise from pixel misclassification in SWOT products. Since the vector product (nodes) is derived from the pixel cloud, the identification of water areas can be affected by factors such as saturated soil, surface roughness, excessive moisture, or wind interference—conditions often associated with strong backscatter in the Ka-band [38,52,53]. Such features were observed in many flooded areas of RS, especially in agricultural zones near major rivers.

SWOT was initially designed to provide accurate measurements over continental water bodies larger than 250 × 250 m or rivers wider than 100 m [23,24,38]. However, this study, in line with others [27,54], demonstrates that the satellite can capture valid altimetric information from rivers as narrow as 30 m—the threshold adopted by the SWORD database on which the node product is based [37]. This capability significantly expands SWOT’s applicability in smaller hydrographic systems, such as those commonly found in urban and rural watersheds in Brazil.

It is expected that the water masks associated with SWOT will yield WSE estimates with an accuracy of approximately 10 cm [55,56]. However, this accuracy may be influenced by the water body’s morphology, including its shape, extent, and width. Narrow and sinuous rivers, as observed in specific locations in RS, tend to produce greater uncertainties which may lead to divergent variations in water between SWOT and in situ gauges.

Despite its strong spatial coverage, SWOT’s main limitation remains its temporal resolution. The ~21-day repeat orbit hinders its ability to detect short-duration or rapidly evolving events [25,57,58,59], as occurred during the 2024 flood in RS. The satellite was unable to capture the onset of the flood in much of the state, which limits its usefulness for early warning systems. Nonetheless, SWOT proved effective in characterizing the evolution and persistence of the event, serving as a complementary tool to conventional monitoring.

Finally, the fact that SWOT nodes are static and have a unique identifier makes them comparable to virtual stations. This allows for detailed and regular spatiotemporal analyses, even in remote or inaccessible locations. Given the structural failures of ground-based monitoring networks during disasters, the use of remote sensing—particularly SWOT data—represents a significant advancement in strengthening water security and risk management.

5. Conclusions

This study demonstrated the altimetric capability of the SWOT mission for hydrological monitoring, focusing on water surface elevation derived from the vector product nodes, which were used as virtual stations. The analysis encompassed both an extended time series—from July 2023 to April 2025—and the extreme flood event that occurred in the state of Rio Grande do Sul, Brazil, in early 2024—the worst meteorological disaster in the state’s history, which affected nearly the entire territory, resulted in hundreds of deaths, displaced thousands, and severely damaged the in situ hydrological monitoring system.

In the first scenario, covering over a year and a half of data, the comparison between the SWOT WSE variations and variations in water levels recorded by telemetric gauges showed a consistent performance, with an MAE of 35 cm and an RMSE of 73 cm after outlier removal, based on subsequent cycles of SWOT passes. These results indicate a moderate correlation and reliability of the satellite data, reinforcing its viability as a complementary alternative for continuous monitoring of river water levels.

In the second scenario, focused on the 2024 extreme event, we assessed windows before and after the flood’s onset. Before the event, the SWOT data showed an MAE of 18 cm and an RMSE of 23 cm relative to the telemetric gauges. After the onset of the flood, these values shifted to 26 cm (mean absolute error) and 34 cm (root mean squared error). Despite the degradation of metrics during the flood peak, the overall errors remained moderate, supporting the robustness of SWOT in critical scenarios.

During the flood period, 36 telemetric gauges located at sites with valid SWOT nodes experienced data acquisition failures in the timeframe window proposed. SWOT was able to record WSE data in these areas, highlighting its relevance as an operational tool when conventional systems fail.

Given the increase in extreme hydrometeorological events, such as those observed in 2024, the use of remote-sensing systems like SWOT becomes essential to enhance the coverage, redundancy, and reliability of hydrological monitoring. The vector node product is only one of several possibilities provided by the mission. Other products, such as the pixel cloud and flood masks, combined with variables from SWOT, including flooded area and discharge estimates, should be explored in future studies to enhance our understanding and response to extreme events.