Introduction

The Salar de Uyuni is the world’s largest salt desert, extending up to 10,000 square kilometres1. It is located 3,600 m above sea level2 in the southwestern part of the Potosí region in Bolivia (Fig. 1). The region has a distinct atmospheric cycle, giving the region dry and wet seasons3. The dry season is typically from April to November. During this period, the crust at the surface is almost pure halite (NaCl) and gypsum (CaSO4.2H2O)2. The dry phase presents a distinctive pattern of polygons (Fig. 2) that cover a pool of brines4, which are exceptionally rich in lithium5. The Salar de Uyuni becomes fully accessible during the dry season. To date, field surveys have been conducted to describe geochemical processes, and the results have been discussed in several papers4,6,7.

Fig. 1: Study area of the Salar de Uyuni (Bolivia), with six superimposed Sentinel-3 ground tracks.
figure 1

The base map is a true-colour composite (bands 4, 3, and 2) from the Sentinel-2 satellite acquired on 27 February 2024, which is used to show the geographical location of the study area.

Full size image
Fig. 2
figure 2

Image of the surface via the drone camera on 20 February 2024 showing a bright disc, which is a mirror reflection of the sun.

Full size image

Rain falls mostly during the wet season between December and March8. The Salar de Uyuni is a closed plateau exhibiting minimal surface topographic variation, typically within 1 metre9. The region receives water mostly from direct rainfall10. Owing to the negligible drainage of the impermeable bed, water is not absorbed into the ground but accumulates during heavy rain, causing the plateau to flood. As a consequence, the Salar de Uyuni transforms its appearance towards that of the surface of a lake a few centimetres deep4. The surface creates a mirror-like effect in which anything above it is reflected specularly11. To observe such an effect, the surface must be smooth enough to be within a fraction of the electromagnetic wavelength of visible light (380–750 nm) for reflection to occur. The only evidence is based on visitors’ photographs showing that the water surface is not disturbed by winds that are typically in the range of 4–5 ms−1 on average during the wet season12.

The Salar de Uyuni is speculated to be the world’s largest natural mirror. However, during the wet season, the area accessible to tourists is limited to the borders. The interior can be reached only with experienced local drivers and appropriate vehicles. Therefore, the only available information is from satellites; however, they have never been used for monitoring the smoothness of the surface. Radar altimeters were designed to measure heights13. Abileah and Vignudelli14 developed a metric to quantify the fraction of the incident electromagnetic energy that is reflected from a specular surface, i.e., one for which irregularities are small compared to a wavelength and reflecting in a well-defined way, without scattering, as opposed to a diffuse surface15.

The premise was to use the radar cross-section (RCS). The smoother the surface is, the higher the RCS. A major advantage of radar altimeters is that they do not require sunlight under cloud-free conditions, unlike satellites that use optical sensors. Moreover, they view the surface at nadir, unlike synthetic aperture radar (SAR) sensors, which have a slanted view, for which smoothed surfaces deflect radar pulses away from the spacecraft’s field of view. In this respect, satellite altimeter and SAR sensors behave in opposite ways while observing radar-smooth surfaces, as the former receives a powerful radar return, whereas the latter does not receive a return signal at all16.

Here, we use approximately three hundred ninety-two thousand radar altimeter bursts of Sentinel-3 satellite constellation data (S3A and S3B) collected along six ground tracks (Fig. 1) since 2016 within the Salar de Uyuni boundary to observe how the smoothness of the surface evolves spatially and temporally.

Results and discussion

The first-ever field survey in the interior of the Salar de Uyuni occurred from 16–20 February 2024 to check the validity of the interpretation of the radar altimeter observations. The photograph taken from a drone camera on 20 February 2016 at 10:17 UTC, which is exactly coincident with the Sentinel-3 satellite overpass, shows a uniform white disc-like spot that is the mirror image of the sun (Fig. 2). This finding indicates that surface imperfections do not substantially impact its specular reflective characteristics. With an optical tool, we measured zero vertical surface displacement within ±0.5 mm at a water depth of 1.8 cm. According to Fraunhofer’s criterion17, a water surface can be considered electromagnetically smooth at a radar altimeter wavelength of 2.2 cm if the root-mean-square (rms) of the surface height distribution variations in height is less than 0.69 mm. The measured RCS values were 120 ± 0.3 dBsm over 100 km along the track around the in situ site, which closely aligns with the theoretical maximum expected for specular bursts in the Salar de Uyuni (RCS = 119.76 dBsm; see the ‘Methods’). Observations also reveal that the wind decreased from an initial value of 4.5 ms−1 to 3.4 ms−1 (at the time of satellite overpass and optical measurements) and continued to decrease thereafter. During the four days prior to the satellite overpass, at this and other locations at similar water depths, a maximum wind speed of 5 ms−1 was recorded, and similar vertical surface displacements were calculated. The measured values match the expected February wind speed of less than 8 ms−118, as also reported in Schmidt7.

The statistical distribution of RCS values along the six Sentinel-3 tracks within the Salar de Uyuni boundary during the period from 2016 to 2024 is shown in Fig. 3. The red line represents the RCS shape as a continuous function of dBsm, with a mixture of two distributions that peak at 111.72 and 119.5 dBsm. The narrow distribution with a peak of an approximately 119.5 dBsm corresponds to specular wet surfaces, with the satellite crossing water areas that are not disturbed by wind. The RCS values are very high and close to the maximum theoretically expected value for specular echoes in the Salar de Uyuni; therefore, the surface is considered smooth at the radar altimeter wavelength.

Fig. 3: Distribution of the RCS values given in dBsm within the Salar de Uyuni boundary, as shown in Fig. 1.
figure 3

The dashed lines indicate the RCS peaks for specular and nonspecular bursts in the Salar de Uyuni.

Full size image

The wide distribution of the approximately 111.72 dBsm peak represents nonspecular surfaces. The RCS values vary in the broad range of 74–118 dBsm. They are probably due to extensive free-standing water that is rougher due to the windiest conditions during the wet season. We suggest that the highest values are also related to the surface characteristics during the dry season. Notably, surfaces that have no moisture content (which we call ‘dry’) can have varying RCS in the Ku-band depending on their material, geometry and smoothness. The Salar de Uyuni during the dry season is composed of a salt crust. It exhibits higher RCS values compared to other dry surfaces e.g. desert made of sand; pavement made of asphalt. With the evaporation of free water, the continuous film of water is replaced by a flat surface of salt that exhibits closely packed cauliflower structures19. Although it is quite rough at the subcentimetre scale, the surface appears to be a smooth reflector at a 2 cm wavelength20. The dominant effect is a quasi-specular reflection. Similar responses are found in the Utah Desert, which has geologic features comparable to those of the Salar de Uyuni21.

Figure 4 shows the 15-day fraction (%) of specular and nonspecular bursts relative to the monthly total, corresponding to the two previously identified RCS value distributions. The analysis reveals that approximately 11% of the total RCS values refer to specular surfaces, whereas the remaining RCS values are for nonspecular surfaces. The surface starts becoming radar smooth at the beginning of the wet season in December. The peak period is from late January to early March. In late February, visitors have the highest chance of seeing the mirror-like effect, as approximately 50% of the bursts are radar smooth. From late April to November, the surface becomes radar smooth only on very rare occasions.

Fig. 4
figure 4

Fraction (%) of specular and nonspecular bursts, computed every 15 days, relative to the monthly total of the year ((January to December).

Full size image

A metric based on sidelobes in radar altimeter waveforms is also introduced to better refine the decision whether the water is smooth or alternatively not smooth or there is no water. Supplementary Fig. 1 shows the Sentinel-3 radar response during 1 March 2024. The lowest panel highlights that the average waveform is close to the theoretical one by applying Hamming window. In order to see the slight broadening of the main peak, no post-processing correction for the beat-frequency drift is applied14,22. The central panel displays the along track RCS, with an average value of around 120 dBsm. Supplementary Fig. 2 is from a Sentinel-3 passage 27 days later (28 March 2024). The average RCS has dropped to around 108 dBsm and there is a raised level on the sidelobes, presumably due to the roughness of the dry lake surface. We analyse 385 passes, which are then divided into two categories (specular and quasi-specular), based on the sidelobe levels. The water surface is assumed specular if the sidelobes raise less than 1 dB from the theoretical value. If sidelobes raise more than 3 dB the water surface is assumed quasi-specular. Supplementary Fig. 3 shows the probability density function (PDF) for the two categories. The majority of the data fall in the quasi-specular category. The specular cases sharply peak with a median of around 119.6 dBsm. Only 12 cases are identified in the wet category. There are two in January, six in February and four in March, confirming that February is the most likely month for smooth water.

The RCS values corresponding to radar-smooth surfaces reflect the rainfall cycle (Fig. 5) in the closest city of Uyuni (Fig. 1), where a meteorological station operates. These events are consequent to the rainfall events over the study area during the wet season. Pearson correlation coefficients were computed considering different temporal lags for rainfall accumulation and RCS representative values (see Supplementary Note 1 for methodological details). The best agreement (r = 0.74 ± 0.05) occurred when 35 days of accumulated rainfall and mean RCS values greater than 110 dBsm were considered.

Fig. 5
figure 5

Time series of the RCS in dBsm (blue for specular and yellow for nonspecular) and rainfall data in mm that accumulated during a day (red colour).

Full size image

The only important exception during the dry season was August 2018. The area reached 9.5 mm and 4.5 mm of water depth on 4th and 5th August 2018, respectively, which was enough rain to cover a surface of 1.4 cm, assuming uniform accumulation.

During the dry season, the westerly wind typically prevents easterly moisture from precipitating in the Altiplano3, where the Salar de Uyuni is located. We suggest that the exception is associated with a disruption of atmospheric conditions23, possibly due to large-scale climate anomalies24. This unexpected rainfall contributed to the temporary flooding of the Salar de Uyuni, and Sentinel-3 flew over the water layer on the 5 and 6 August 2018. There was a long range of peaks at approximately 120 dBsm (Fig. 6), a clear indication of the existing radar-smooth surface. A GPS-tagged photograph (retrieved from an on line source) taken on 13 August 2018 at a latitude of −20.414313° and longitude of −67.090105° confirms the presence of water inside the Salar de Uyuni southwest of Plaza de las Banderas, which is along the road leading west from Colchani (the gateway to the salt flat).

Fig. 6
figure 6

Distribution of RCS values in dBsm during the rare event observed in August 2018.

Full size image

Figure 7 shows the spatial distribution of bursts on a year-to-year basis during the wet season from late 2021 to 2024 when the Sentinel-3 satellites were fully operating. The RCS values decreased following lower smoothness as the colour changed from blue to yellow. In this figure, the focus is on the rainiest and driest years. From December 2021 to March 2022, the Salar de Uyuni received a large amount of rain (222 mm), which occurred mostly in December (102 mm) and that gradually decreased in January (84 mm), with a slow decline in February (32 mm) until a minimum was reached in March (4 mm). The rise in water during December led to a rapid RCS increase in some areas of the Salar de Uyuni. An increase was especially observed on the western edge. Smoothing remained stable in the middle (track 0167 of S3A) from January to March.

Fig. 7
figure 7

Distribution of RCS values annually during the wet season.

Full size image

The Salar de Uyuni had the driest period from December 2022 to March 2023, with a total rainfall of 87 mm. Rainfall started in December, with a value of only 25 mm. The lowest value was observed in January (8 mm), so visitors had fewer chances to see the mirror-like effect. Almost half of the total rain (42 mm) that fell in March (12 mm) was recorded in February. High RCS values were concentrated only in the middle of the Salar de Uyuni during December. The limited input of free water during January slightly decreased the smoothness of the Salar de Uyuni because of evaporation. The RCS values rapidly increased in response to the enhanced rainfall during February, with the Salar de Uyuni becoming fully filled and smooth in the central part. The RCS then quickly decreased in March following the reduced amount of rain.

December 2023 to March 2024 was the wettest part of the analysed period, with 265 mm of accumulated rain. The wet season started with approximately 25 mm in December, increasing to approximately 37 mm in January and 64 mm in February, and finally, an exceptional peak of 139 mm occurred in March; this pattern meant that visitors would have had a longer time to see the reflection. During the same period, the RCS values follow the rainfall variations, with the specularity starting in December, with peaks at approximately 120 dBsm when the rainfall is concentrated mostly in the northwestern part of the Salar de Uyuni. In January, the RCS increases and spreads to the east, with most of the Salar de Uyuni becoming completely specular in February and March 2024.

The evaporation rate is important to understand the short time between a specular and following quasi-specular pass. Borsa19 suggested that evaporation over Salar de Uyuni is spatially and temporally uniform at approximately 2 mm/day. The satellite passage on 10 March 2017 occurred after a 27-day period in which 7.5 cm of rainfall was recorded. Based on the previous evaporation rate, around 2 cm of surface water remained on the Salar de Uyuni at the time of the satellite. During the following 27 days, rainfall was negligible. At the time of the next satellite pass on 6 April 2017, quasi-specular reflections were observed, indicating a drying tendency. A similar pattern was observed in early 2019. Prior to 14 January, no rainfall was recorded. This was followed by several days of rainfall totalling about 8.5 cm. By 10 February, the satellite detected specular reflections over the Salar de Uyuni, with an estimated 3.1 cm of surface water. An additional 3.3 cm of rainfall occurred before the next satellite passage on 9 March, when about 1 cm of water remained. After that, rainfall ceased, and on 5 April 2019, the surface again exhibited quasi-specular reflections. These observations highlight the important role of evaporation during the relatively short transition from specular to quasi-specular conditions.

The rainfall intensity in the Salar de Uyuni has shown quite unpredictable schedules in recent years, which suggests a connection to climate change due to El Niño‒Southern Oscillation (ENSO) conditions that fluctuated from 2021–202425. The La Niña event began in September 2020, with pronounced effects from 2021 through 202226, resulting in increased precipitation in the Altiplano27. Then, La Nina shifted to a weaker status from 2022–2023, explaining the below-average rainfall over the region. WMO28 reported that La Niña ended and El Niño conditions began to develop in early 2023; however, during the latter part of 2023, the El Niño conditions weakened, which might explain the possible enhanced moisture transport to the Altiplano, causing above-average precipitation over the Salar de Uyuni.

To more effectively evaluate the relationship between rainfall and ENSO variability, cumulative rainfall during the wet season (December–February, DJF) was calculated using two complementary approaches. First, we extended the observational time series from the Uyuni station to cover the period 1976–2025, allowing for an analysis of long-term inter-annual variability in DJF rainfall (Fig. 8).

Fig. 8: Cumulative rainfall (mm) observed at the Uyuni station during the wet season (DJF) from 1976 to 2024. DJF stands for December, January and February months.
figure 8

The ENSO–rainfall relationship is shown with colours that indicate El Niño (brown), La Niña (blue) and neutral (grey) years according to the NOAA definition reported in the section ‘Climate data’.

Full size image

Second, we extracted recent DJF precipitation estimates (2021–2024) from the Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) satellite-based data, spatially averaged over the Salar de Uyuni Basin (Fig. 9). In both cases, the values were compared against the Oceanic Niño Index (ONI), using thresholds of ONI ≥ +0.5 to indicate El Niño conditions and ONI ≤ –0.5 to represent La Niña events.

Fig. 9: Cumulative rainfall (mm) DJF rainfall estimates derived from CHIRPS satellite-based observations, spatially averaged over the Salar de Uyuni from 2021 to 2024.
figure 9

The ENSO–rainfall relationship is shown as explained in Fig. 6.

Full size image

The extended La Niña from late 2020 to 2022 coincided with a notably wet DJF in 2021–2022, which recorded 222 mm at the Uyuni station and was associated with early and persistent radar-smooth surface conditions. However, in 2022–2023, despite the continuation of La Niña (ONI = –0.7), DJF rainfall decreased sharply to 87 mm, and specular reflections were minimal. This breakdown in the expected ENSO–rainfall relationship underscores the importance of examining longer records and suggests a potential weakening of the ENSO teleconnection or the influence of additional drivers. The long-term comparison between ONI phases and DJF precipitation from 1976 to 2024 reveals that La Niña phases often align with wetter summers and El Niño events with suppressed rainfall, but these patterns are not consistent every year. Such discrepancies highlight the complexity of ENSO’s influence on the Altiplano’s hydrology and justify the need to incorporate multi-decadal data in climate-rainfall analyses.

Rainfall records from Potosí Los Pinos, a representative station in the southern Altiplano, indicate that approximately 77% of the annual rainfall occurs between December and March24. Future climate projections suggest that global warming could intensify ENSO variability, potentially leading to more extreme El Niño and La Niña events29. Such changes may disrupt current precipitation patterns in the Altiplano, altering the seasonal dynamics of the Salar de Uyuni. Also, additional modes of variability, such as the Atlantic Meridional Mode (AMM) and the North Atlantic Oscillation (NAO) may also modulate regional moisture fluxes and contribute to observed anomalies24. One example is the off-season rainfall event in August 2018, during a moderate La Niña, which led to the temporary flooding of the Salar de Uyuni.

Our results indicate that during the wet season, water typically sits on the surface, but its smoothness evolves spatially and temporally. The Salar de Uyuni is not a vast uniform mirror for the radar altimeter. Therefore, it is also likely that it is not a large mirror for optical wavelengths30, as suggested in the literature. Satellites indicate how the Salar de Uyuni transforms into a smooth wet surface early on, thereby helping visitors to better decide when to travel to see the mirror effect. As radar echoes from a smoothed water surface tend to be very strong, the required radar power can be considerably reduced with a paradigm change away from large spacecraft towards a swarm of small satellites that would provide better coverage with more frequent revisits.

We anticipate a parallel study to explain why no waves were generated initially, despite considerable winds. According to Munk31, there is a threshold wind speed below which waves do not form and the flow of water is laminar. Although several theoretical and experimental studies have been conducted32, some discrepancies are found in the reported thresholds33. A wave model will be used to establish a threshold for wave generation, given the depth and viscosity of the water samples collected at the survey site.

Methods

Radar altimeter data

Sentinel-3 is one of the satellite missions of the European Commission’s Copernicus program. Two satellites, Sentinel-3A (S3A) and Sentinel-3B (S3B), are presently in operation. S3A was launched on 16 February 2016 and orbits at an altitude greater than 800 km, having accumulated more than eight years of data at the time of this study. S3B was launched on 25 April 2018. The orbit of S3B is identical to that of S3A, but it flies +/−140° out of phase with S3A. Both satellites carry one nadir-looking altimeter radar called the SAR Radar ALtimeter (SRAL) at 13.575 GHz (Ku band). S3A and S3B are on repeat orbits, i.e., the satellite revisits the same locations every 27 days. This orbit configuration results in regular sampling over the Salar de Uyuni. The Sentinel-3 radar altimeter transmits chirped pulses and receives the reflected echoes, which are split into two components: the original beat called “I” (in-phase) and the same beat but shifted by 90° called “Q” (in quadrature). Together, these components form a complex signal (I, Q), which preserves the amplitude and phase information of the echo for Doppler processing. The complex signals (I, Q) are then grouped in bursts and downlinked to the ground stations. Each burst contains 64 individual echoes (IEs). It corresponds to approximately 25 m on the ground. Bursts are spaced approximately 89 m apart. Major details about the Sentinel-3 radar altimeter process can be found in Donlon et al.34. A notable feature of Sentinel-3 satellites is that IEs over the whole Earth are archived via a Level-1A product that is used as input for processing. Briefly, the processing consists of two steps: (1) fast fourier transform (FFT) of the fast time dimension of the (I, Q) data into a radargram; (2) scaling of the radargram into dBsm with the scaling factor reported in Dinardo and Lucas35. The code implementation is described in Vignudelli36. When this analysis was conducted, radar altimeter data were available for the period from 2016 to 2024.

Earlier satellite radar altimetry missions (e.g., TOPEX/Poseidon, Jason-1/2/3, ERS-1/2) cannot be used to extend the existing eight-year RCS time series. The main reason for this is that those missions did not implement the on-board calibration in power that is required to establish the radar equation budget. Also, they were not designed to downlink both the in-phase (I) and quadrature (Q) components for each received echo. Envisat was the first satellite radar altimetry mission to collect an experimental data set of in-phase (I) and quadrature (Q) components from radar echoes37. However, these sequences were recorded only every 3 min, therefore, they were only available for a small percentage of the Earth’s surface. CryoSat-2 mission, designed for polar region, has a drifting orbit making sporadic passages over the Salar de Uyuni. Morever, the design constraints are that the low resolution mode is available over most of the earth surface and the Full Bit Rate (FBR) mode, enabling the acquisition of the individual echoes, is limited to few areas that do not include the Salar de Uyuni38. Unfortunately, Sentinel-6 (the follow-on mission of Jason-3), which would acquire individual echoes, does not cross the Salar de Uyuni. It is also important to highlight that two identical satellites (S3C and S3D) will be launched sequentially into the same reference orbit of S3A and S3B34. The Sentinel-3 Next Generation Topography (S3NG-TOPO) mission is dedicated to ensuring the baseline continuity of existing Copernicus Sentinel-3 nadir-altimeter observations from 2030 to 2050, thereby supporting climate-scale monitoring over Salar de Uyuni.

It is also useful to recall the experimental Surface Water Ocean Topography (SWOT) mission that moves from point-wise along-tracks to pixel-based maps. The SWOT satellite is equipped with a Ka-band Radar Interferometer (KaRIn) sensing water surfaces within two 50 km wide swaths left and right of the satellite, separated by a 20 km wide nadir gap39. The SWOT satellite is operating in non-synchronous 21-day repeat orbit since July 2023 for regular data acquisition. SWOT mission’s Version C science data products became publicly available in October 2024. The SWOT satellite overpassed the region on 20 February 2024, almost coinciding with both the field survey and the Sentinel-3 passage. We use here the highest available resolution data product called High Rate (HR) PIXel Cloud (PIXC) described in JPL D-10953239. This product has the best spatial resolution from 10 to 60 meters in the range direction depending on the position within the swath and around 22 meters in the along-track direction40.

When the surface is wet and smooth, no imaging of specular waters is possible, other than on the nadir track. This behaviour was already anticipated with experiments—e.g. Nouvel and Souyris41 showed the case of backscatter of a smooth pond at low incident angles. It is also expected since ‘water detection over land assumes that water is more reflective than the surrounding land’, as reported in JPL D-10953239. But now it can also be proved with acquired SWOT data, as the satellite crossed the Salar de Uyuni during the same day (20 February 2024) of the field survey and Sentinel-3 passage. During that day, we know that the entire Salar de Uyuni was covered with water and that Sentinel-3 RCS was constant at maximum level from the north to the south edges, which confirms the water surface was smooth. Supplementary Fig. 4 shows the backscatter (sigma-zero) in dB from the water surface as measured by the KaRIn instrument. The image (left swath) shows the specularity at the nadir, during the brief nadir passage on the edge of the Salar de Uyuni. There is a signal at the nadir as the KaRIn instrument acquires data over a window in range, and that window can extend all the way to the nadir. In all other off-nadir locations, the backscatter is at or near the noise floor of the KaRIn instrument. At the field survey site, the mean intensity (within 1 km) is 0.39 dB (with standard deviation (std) of 2.88 dB) much lower than dry soil (mean of 6.84 dB and std 8.15 dB). This is the scenario when water cannot be distinguished from land, as water is not much more reflective than land39. There is a demonstration of the limitations of SWOT with regard to specular water targets. It could only see those targets near nadir just like all previous altimeters.

Satellite radar altimetry processing and calculation of the Sentinel-3 RCS

The RCS is a quantity explained by Abileah and Vignudelli14 following investigations in Abileah et al.37. It is computed for each burst and summarizes in a single number the extent to which the surface target captures energy from the radar and reradiates it back. The unit of measurement is decibels relative to a square metre (dBsm, where 0 dBsm represents an RCS of  1 square metre). Initially, we used a very basic formulation, considering the unknown designed parameters of the radar. In this study, RCS values were more accurately estimated using the specific radar equation for Sentinel-3 data provided by the European Organisation for the Exploitation of Meteorological Satellites, EUMETSAT35.

The Salar de Uyuni can be assumed to be a plate when the surface is covered by water and is specular (i.e., the rms of the distribution of the surface height is less than 0.69 mm). It is exceptionally large, effectively approximated by a 100 × 100 km disk, and is best represented as an “effectively infinite” surface in terms of radar response. Also, there is no ground clutter (i.e., received signals from unwanted surface targets) interfering with the direct echoes. While Dinardo and Lucas35 report a theoretical maximum RCS of 132 dBsm, the correct value for the Salar de Uyuni is 119.76 dBsm, or approximately 120 dBsm. Dinardo and Lucas35 derived the maximum RCS for a 180 m diameter disk surface (the first Fresnel zone) under assumptions that do not represent the Salar de Uyuni’s scenario. Those assumptions are a far-field condition, a flat Earth, a water reflectivity of 1 and neglecting atmospheric attenuation. The RCS of the Salar de Uyuni invokes the near-field theory explained in Pouliguen et al.42.. Using Sentinel-3 system parameters, the RCS is 129 dBsm under a flat Earth approximation. When Earth’s curvature is included, the maximum RCS slightly decreases to 128 dBsm. A further loss of 6 dBsm must be considered for the radar pulse response to an ‘infinite’ Salar de Uyuni basin43. Finally, when accounting for a realistic water reflectivity ( ~ 2.1 dBsm loss for a salt desert in Ulaby et al.21.) and a nominal loss in the atmosphere of 0.14 dBsm35, the correct theoretical maximum RCS for specular bursts in the Salar de Uyuni is 119.76 dBsm. This value is consistent with the ~120 dBsm measured by Sentinel-3 during the 20 February 2024 field survey with the surface water-covered and exhibited specular reflection.

The RCS analysis revealed that cycle 21 (29 January 2019) of S3B track no. 231 is missing because of instrument malfunction. The RCS values along S3A track no. 174 from 14 June 2016 to 20 June 2017 were not used, as the radar was not tracking the surface properly because of Open Loop Tracking Command settings. Approximately 4.5% of the bursts have RCS values > 119.76 dBsm, with 96.21% in the range of 119.76–121.76 dBsm, 3.70% within 121.76–125.76 dBsm and only 0.09% exceeding 125.76 dBsm. These observed RCS values differ from those predicted by theory. Their spatial distribution is not localized to specific areas in Salar de Uyuni but instead appears to be random. Moreover, they predominantly occur during the wet season

We believe that in these cases the theoretical results do not predict real-world effects. For example, it is assumed that the propagation of electromagnetic waves occurs across an homogeneous atmosphere. The electrical properties of the medium contribute together with the surface geometric characteristics to determine the intensity of the radar returns. Those electrical properties are unknown in the Salar de Uyuni at the time when Sentinel-3 was flying. The value of the 2.1 dBsm loss that we used may in fact be even lower, e.g., a loss of 0.4 dBsm (corresponding to a coefficient of reflectivity of 0.9) was reported in Livingston44. There might also be scattering mechanisms that were not considered in the theoretical approach. For example, during the dry season, the surface is formed of cauliflower-like structures of repeating clusters of florets with reentrant cavities, with concordant high points forming a flat plain. When their size is comparable to the wavelength of the incident radar wave, the interference of front-face specular waves with rear-side diffracting waves might cause a resonant region. This means that those structures contribute to the RCS overall (called the bulk RCS, Uluisik et al.45). Notably, resonance regions can markedly increase the RCS46. Resonance effects have already been reported in the literature in similar regions47. Following the previous considerations, all RCS data with RCS > 121.76 were removed from the analyses.

On the basis of the RCS distribution, a rule set was defined to sort bursts into two categories of scientific interest: specular (RCSs ≥118 dBsm) and nonspecular (RCSs <118 dBsm). A land mask was applied to discard RCS values at the boundaries of the Salar de Uyuni to avoid possible contamination from outside the Salar de Uyuni.

Optical method to compute the smoothness of the water surface

The optical survey procedure (Supplementary Fig. 5) consists of videos to record fine-scale details of surface displacements. The white and black panels are located underwater and are pressed against the surface of the Salar de Uyuni, which is 1.8 cm deep. The waterproof panels are 203 mm × 250 mm in size; they are laminated photographic balancing sheets. A tripod was equipped with a camera oriented to view the panels from the top. A second camera was fixed to provide a nearly horizontal view. This configuration allows videos of the water surface to be taken from two directions. Wooden and plastic beads (diameter ~13 mm) were floated on the water surface and video recorded by the two cameras, one imaging from above, and the second horizontally. The beads were of various colours to be sure that with some there was high contrast with the background. In the vertical videography the drifting beads measured the speed and direction of the current. In the horizontal videography beads measured vertical displacement. The vertical position rms was <0.5 mm, which is better than the Fraunhofer criterion for smooth surface in Ku band.

Ground-based measurements

The rain gauge data used in the study are from the Bolivian Weather Service (Servicio Nacional de Meteorología e Hidrología; SENAMHI). Rainfall intensity is measured once per day as 24-hour accumulated rainfall (in mm) at 8 am local time48. One millimetre is equivalent to 1 litre per square metre49. SENAMHI offers daily and monthly aggregated rainfall. The Uyuni station (Fig. 1) was selected on the basis of its proximity to the Salar de Uyuni12 and the availability of rainfall measurements, which should cover the period of Sentinel-3 satellites (2016–2024). During the field survey in the Salar de Uyuni, the wind speed was measured using a hot-wire anemometer (Model BA30W; Trotec GmbH, Germany) for precise determination (0.01 ms−1 resolution) of even low-flow conditions. The sensor was positioned approximately 4 m from the water surface. The data were logged at 1-sec intervals via the MultiMeasure mobile app. Recording sessions were repeated during the survey period to confirm the reproducibility of the responses.

Climate data

We use the ONI derived from the 3-month running mean of the National Oceanic and Atmospheric Administration (NOAA) Extended Reconstructed (ER) Sea Surface Temperatures (SST) version 5 (ERSSTv5) data available from the National Centre for Climate Prediction. Major details are found in Huang et al.50. The identification of El Niño, La Niña and neutral phases is based on ONI. El Niño events are defined when the ONI reaches the threshold of +0.5 °C for at least 5 consecutive 3-month means. An ONI value of −0.5 or lower indicates La Niña event. Neutral status means that neither El Niño nor La Niña conditions are present. As a complementary source of rainfall data over the Salar de Uyuni, we use the CHIRPS product that provides gridded estimates spaced ~5 km apart. It is a blended product combining multiple sources, including satellite-based observations from thermal infrared sensors51. CHIRPS has been shown to accurately reproduce in-situ observations in the region8,52.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.