What is Synthetic Aperture Radar (SAR)? Bands, Data, Uses

Synthetic aperture radar (SAR) is an active imaging radar that sends microwave pulses from a satellite or aircraft and records the echoes to form an image of the ground. Microwaves pass through cloud and smoke and need no sunlight, so SAR images the ground at any hour. Each pixel stores backscatter, the share of energy the surface returns to the antenna. Complex SAR data also keeps the phase of the echo, the point in its wave cycle at which it returns. The European Union's Sentinel-1 mission publishes SAR data free, and analysts use it to map floods, detect ships, and measure subsidence. This article argues that water, buildings, and vegetation return distinct backscatter and coherence, a measure of how similar a pixel's echoes are on two passes. The harbor walkthrough below measures both values on surfaces labeled with Overture map data.

Key takeaways

• The synthetic aperture is a long virtual antenna, formed by combining echoes recorded along the flight path.
• Wavelength sets what SAR measures: X-band scatters from the top of a canopy, and longer C-band, L-band, and P-band waves penetrate deeper into vegetation.
• Interferometric SAR (InSAR) compares the phase of two passes to measure ground displacement of centimeters or less.
• At the Ports of Los Angeles and Long Beach, open water returns a median of -16.2 dB with a 12-day coherence of 0.03. Dense buildings return -5.8 dB and 0.57.

How synthetic aperture radar works

A SAR antenna transmits a microwave pulse to one side of the flight track and records the fraction the surface scatters back. Echo delay fixes each target's position in range, the direction across the track.

Resolution along the track, called azimuth, depends on antenna length. A real antenna fine enough for 10 m pixels from orbit would be kilometers long (NASA Earthdata SAR primer). SAR replaces that antenna with motion. Each target stays inside the beam while the satellite travels a stretch of its orbit. The echo frequency shifts as the satellite passes the target (the Doppler shift). A processor uses that shift to combine every echo from the stretch into one pixel. That stretch of orbit is the synthetic aperture. Moreira and colleagues' tutorial on synthetic aperture radar (2013) derives the math.

Side-looking geometry distorts terrain through foreshortening, layover, and shadow:

• Foreshortening. Slopes facing the radar appear compressed.
• Layover. The top of a mountain or tower returns its echo before the base, so it appears displaced toward the sensor.
• Shadow. Slopes facing away from the radar receive no signal and appear black.

Brightness depends on roughness, geometry, and moisture. Calm water reflects the pulse away from the antenna and appears dark. Fields and forests scatter part of it back and appear gray. Walls and the ground beside them form corner reflectors that return a double bounce, so cities, ships, and flooded forests appear bright. Interference between scatterers adds a grainy texture called speckle to every SAR image. Goodman (1976) showed that fully developed speckle has a contrast of one. Averaging M independent looks, separate estimates of the same pixel, reduces it by 1/√M. Lee (1980) introduced a filter that uses the local mean and variance to smooth uniform areas and keep edges.

SAR frequency bands

Longer wavelengths pass through leaves to branches, trunks, and the ground, and shorter ones scatter from the first surface they meet. The band ranges follow NASA Earthdata.

Band	Frequency	Wavelength	What it measures	Example missions
X	8 to 12 GHz	2.4 to 3.8 cm	Fine surface detail, ships, buildings, canopy top	TerraSAR-X, COSMO-SkyMed, ICEYE, Capella Space, Umbra
C	4 to 8 GHz	3.8 to 7.5 cm	Crops, soil moisture, sea ice, oil slicks	Sentinel-1, RADARSAT Constellation Mission
S	2 to 4 GHz	7.5 to 15 cm	Crops and moderate vegetation	NISAR (S-band)
L	1 to 2 GHz	15 to 30 cm	Forest structure, deformation under vegetation	NISAR (L-band), ALOS-2
P	0.3 to 1 GHz	30 to 100 cm	Forest biomass	ESA Biomass

L-band phase stays stable over vegetated ground for longer than C-band phase, so L-band InSAR works in forests where C-band coherence is lost.

SAR polarization

Polarization is the orientation of the radar wave's electric field, horizontal (H) or vertical (V). Each channel is named by two letters: the transmit orientation, then the receive orientation. HH and VV are co-polarized: the radar receives in the same orientation it sent. HV and VH are cross-polarized. Rough bare surfaces return most energy in VV and HH. A tree canopy scatters the wave many times and rotates part of it, so cross-pol returns rise with vegetation volume. Double bounce from buildings and flooded trunks is strongest in HH. Sentinel-1 acquires VV and VH over most land, and HH and HV over polar ice.

Polarimetric decompositions split a pixel with all four channels into these scattering mechanisms. Cloude and Pottier (1997) classify pixels by two values computed from the channels' coherency matrix: entropy, how mixed the scattering is, and the mean scattering angle. Freeman and Durden (1998) fit surface, double-bounce, and volume components.

SAR data products

Processing levels, named as for Sentinel-1:

• Single Look Complex (SLC). Focused data in slant range, the sensor's line-of-sight geometry. Each pixel holds amplitude and phase. SLC is the input for interferometry.
• Ground Range Detected (GRD). Amplitude only, averaged over several looks and projected to ground range. GRD files are smaller and serve flood mapping and ship detection.
• Radiometric terrain corrected (RTC) backscatter. Calibrated, terrain-corrected with a digital elevation model (DEM), and placed on a map grid as gamma nought (γ⁰). Gamma nought is backscatter normalized by the illuminated area projected perpendicular to the beam, following David Small's terrain flattening (2011). The Committee on Earth Observation Satellites (CEOS) Analysis Ready Data specification calls this Normalised Radar Backscatter.
• InSAR products. An interferogram is the phase difference between two SLC images of the same place. Coherence, from 0 to 1, measures how similar the two signals are and drops wherever the surface changed.

Sentinel-1 measurement layers ship as GeoTIFF, and cloud archives repackage them as Cloud Optimized GeoTIFF. The Alaska Satellite Facility runs HyP3, an on-demand service that produces RTC and InSAR products from Sentinel-1.

SAR vs optical imagery

SAR and optical imagery measure different physical properties, so Earth observation programs such as Copernicus use both.

Attribute	SAR	Optical
Energy source	Own microwave pulse (active)	Reflected sunlight (passive)
Night and cloud	Images through both	Needs daylight and clear sky
Pixel value	Backscatter: roughness, structure, moisture	Reflectance by color band
Geometry	Side-looking, slant range	Near-vertical view
Noise	Speckle	Haze, cloud, shadow
Interpretation	Requires training	Close to a photograph

Optical sensors such as Landsat, Sentinel-2, and hyperspectral instruments identify materials by spectrum. SAR supplies images on cloudy dates, and its backscatter and phase measure structure and ground motion. Wherobots' Strait of Hormuz analysis counted hulls in Sentinel-2 and reported that "Traffic through the surveyed corridor fell 95%". The post cites a separate study that combined Sentinel-1 radar with AIS, the positions ships broadcast by radio.

Synthetic aperture radar examples

Ship detection, threshold flood mapping, and sea ice charting need one scene. Other uses compare two dates by phase (deformation and elevation), coherence (damage), or backscatter (change detection).

• Landers earthquake deformation, 1992. Massonnet and colleagues (Nature, 1993) mapped the ground displacement of the Landers, California earthquake from ERS-1 (European Remote Sensing satellite) interferograms. That result made InSAR a standard geophysics tool.
• Gorkha earthquake damage, 2015. Yun and colleagues (2015) mapped likely damage from the Nepal earthquake within days, using coherence loss in COSMO-SkyMed and ALOS-2 (Advanced Land Observing Satellite 2) image pairs.
• Flood mapping. Floodwater is dark against land, so analysts map it with a backscatter threshold. Twele and colleagues (2016) built a fully automated Sentinel-1 flood processing chain. The Copernicus Emergency Management Service runs a Sentinel-1 Global Flood Monitoring service. Flood risk data combines that observed extent with modeled hazard.
• Dark vessel detection. Steel hulls return bright points on a dark sea, and detections without a matching AIS broadcast are dark vessels. The xView3-SAR benchmark (NeurIPS 2022) labels vessels in nearly 1,000 Sentinel-1 scenes for this task.
• Global elevation. The TanDEM-X radar formation mapped the topography behind the Copernicus DEM GLO-30.
• Sea ice. National ice services chart sea ice from SAR through the polar night.

SAR missions and data access

Public missions supply SAR data free:

• Sentinel-1. The Copernicus C-band mission. Sentinel-1A launched in April 2014. Sentinel-1B followed in 2016, Sentinel-1C in December 2024, and Sentinel-1D in November 2025. Two satellites repeat each orbit every 6 days, and one satellite every 12. Over land, the Interferometric Wide swath mode covers 250 km at 5 m by 20 m resolution. Torres and colleagues (2012) describe the mission design. The Copernicus Data Space Ecosystem distributes the archive, and NASA's ASF DAAC, the Alaska Satellite Facility's Distributed Active Archive Center, mirrors it.
• NISAR. The NASA-ISRO SAR mission is a joint project of NASA and the Indian Space Research Organisation (ISRO), launched in July 2025. One satellite carries an L-band radar from NASA and an S-band radar from ISRO. It images global land and ice every 12 days. The ASF DAAC archives NISAR data openly, and ISRO releases the S-band products through its Bhoonidhi portal.
• National missions. Canada's RADARSAT Constellation Mission (C-band), Japan's ALOS-2 (L-band), Germany's TerraSAR-X, and Italy's COSMO-SkyMed (X-band) serve government programs and license data commercially. The European Space Agency's Biomass satellite, launched in April 2025, is the first P-band SAR in orbit.

Commercial operators such as ICEYE, Capella Space, and Umbra fly constellations of small X-band SAR satellites and sell tasked spotlight images at sub-meter resolution.

SAR analysis in Wherobots at the Ports of Los Angeles and Long Beach

The header image above shows the ports in Sentinel-2 and Sentinel-1. Kellndorfer and colleagues' Global seasonal Sentinel-1 interferometric coherence and backscatter data set (Scientific Data, 2022) processed about 205,000 Sentinel-1 Interferometric Wide scenes from December 2019 to November 2020. The output is seasonal layers in 1 by 1 degree tiles at 3 arcsecond pixel spacing. Each season has mean RTC backscatter in VV and VH and median VV coherence for repeat intervals of 6, 12, 18, 24, 36, and 48 days. It also has the parameters of an exponential model of coherence decay. WherobotsDB reads the tiles in place from a public AWS bucket with RS_FromPath as out-db rasters, rasters left in their files in storage.

The analysis joined these sources:

Source	Format	Size	What the queries read
Kellndorfer et al. (2022) Sentinel-1 coherence and backscatter, tile N34W119, summer 2020	Cloud Optimized GeoTIFF in a public AWS S3 bucket, read as out-db rasters	1 by 1 degree tile, 1,200 by 1,200 pixels per layer	64,512 pixels in the test area: VV backscatter and 12-day VV coherence, plus VH backscatter and 6, 24, and 48-day coherence for the decay table
Overture buildings (buildings_building)	Havasu table (Apache Iceberg) in Wherobots open data	2,533,842,612 rows on 9 October 2026	Footprints in the test area, for each pixel's building cover
Overture water (base_water)	Havasu table (Apache Iceberg) in Wherobots open data	66,204,219 rows on 9 October 2026	Polygons of class ocean, bay, water, or dock, for the open-water label
Overture land cover (base_land_cover)	Havasu table (Apache Iceberg) in Wherobots open data	Not counted	Forest, shrub, and grass areas, for the vegetation label
Overture land use (base_land_use)	Havasu table (Apache Iceberg) in Wherobots open data	56,097,254 rows on 9 October 2026	Areas of class industrial, for the industrial yards label

The test area, 118.36°W to 118.08°W and 33.68°N to 33.84°N in tile N34W119, holds 64,512 pixels of the summer 2020 layers (June to August). One WherobotsDB query reads both raster files and joins their pixels to two Overture tables. Raster and vector in one engine; most stacks pick one. In the query, RS_BandAsArray and posexplode turn each raster into one row per pixel. ST_PolygonFromEnvelope builds each pixel's footprint. Spatial joins to Overture data then assign a surface type. A pixel's building share is the area of Overture building footprints inside it (ST_Intersection), divided by the pixel's area. A pixel counts as open water when its center falls in an Overture water polygon of class ocean, bay, water, or dock and no building touches it.

WITH r AS (
  SELECT RS_BandAsArray(RS_FromPath('s3://sentinel-1-global-coherence-earthbigdata/data/tiles/N34W119/N34W119_summer_vv_AMP.tif'), 1) AS vv,
         RS_BandAsArray(RS_FromPath('s3://sentinel-1-global-coherence-earthbigdata/data/tiles/N34W119/N34W119_summer_vv_COH12.tif'), 1) AS coh
),
pix AS (  -- one row per pixel, with its footprint as a polygon
  SELECT pos, z.vv, z.coh,
         ST_PolygonFromEnvelope(-119 + (pos % 1200) / 1200.0, 34 - (pos DIV 1200 + 1) / 1200.0,
                                -119 + (pos % 1200 + 1) / 1200.0, 34 - (pos DIV 1200) / 1200.0) AS cell
  FROM r LATERAL VIEW posexplode(arrays_zip(vv, coh)) t AS pos, z
  WHERE pos DIV 1200 BETWEEN 192 AND 383 AND pos % 1200 BETWEEN 768 AND 1103
),
bld AS (  -- share of each pixel covered by Overture building footprints
  SELECT p.pos, SUM(ST_Area(ST_Intersection(p.cell, b.geometry))) / MAX(ST_Area(p.cell)) AS frac
  FROM pix p JOIN wherobots_open_data.overture_maps_foundation.buildings_building b
    ON ST_Intersects(p.cell, b.geometry)
  WHERE b.bbox.xmax > -118.36 AND b.bbox.xmin < -118.08 AND b.bbox.ymax > 33.68 AND b.bbox.ymin < 33.84
  GROUP BY p.pos
),
water AS (  -- Overture water, clipped to the test area
  SELECT ST_Intersection(geometry, ST_PolygonFromEnvelope(-118.36, 33.68, -118.08, 33.84)) AS geom
  FROM wherobots_open_data.overture_maps_foundation.base_water
  WHERE class IN ('ocean', 'bay', 'water', 'dock')
    AND bbox.xmax > -118.36 AND bbox.xmin < -118.08 AND bbox.ymax > 33.68 AND bbox.ymin < 33.84
),
wat AS (
  SELECT DISTINCT p.pos
  FROM pix p JOIN water w ON ST_Contains(w.geom, ST_Centroid(p.cell))
)
SELECT CASE WHEN w.pos IS NOT NULL AND COALESCE(b.frac, 0) = 0 THEN 'open water'
            WHEN b.frac >= 0.25 THEN 'dense buildings'
            ELSE 'other land' END AS class,
       COUNT(*) AS pixels,
       ROUND(20 * LOG10(percentile_approx(p.vv, 0.5)) - 83, 1) AS vv_db_median,
       ROUND(percentile_approx(p.coh, 0.5) / 100, 2) AS coh12_median,
       ROUND(AVG(CASE WHEN 20 * LOG10(p.vv) - 83 < -15 THEN 1 ELSE 0 END), 3) AS share_below_minus15db
FROM pix p LEFT JOIN bld b ON p.pos = b.pos LEFT JOIN wat w ON p.pos = w.pos
GROUP BY 1
ORDER BY pixels DESC

The data set stores both layers as scaled integers, called digital numbers (DN). The query converts them: 20 log10(DN) minus 83 gives γ⁰ in dB, and DN / 100 gives coherence. The query returned:

Surface (from Overture)	Pixels	VV γ⁰ median	12-day coherence	Share below -15 dB
Open water	23,637	-16.2 dB	0.03	85.8%
Dense buildings	16,126	-5.8 dB	0.57	0%
Other land	24,749	-7.2 dB	0.50	2.0%

Coherence decay by surface type

Joins to Overture land cover and land use split the other land class. The 6, 24, and 48 day layers show coherence decay over time:

• Vegetation: the pixel center falls in Overture land cover of subtype forest, shrub, or grass (see land cover classification), with under 2% building cover.
• Industrial yards: the pixel center falls in Overture land use of class industrial, with under 10% building cover. Around the harbor this includes the container terminals.

Medians per class:

Surface	Pixels	VV γ⁰	VH γ⁰	Coherence, 6 days	12 days	24 days	48 days
Open water	23,637	-16.2 dB	-27.3 dB	0.03	0.03	0.03	0.03
Dense buildings	16,126	-5.8 dB	-13.9 dB	0.59	0.57	0.53	0.51
Industrial yards	3,516	-2.5 dB	-13.4 dB	0.42	0.38	0.32	0.27
Vegetation	2,552	-10.2 dB	-16.5 dB	0.39	0.35	0.27	0.20
Other land	18,681	-7.1 dB	-14.7 dB	0.54	0.52	0.47	0.43

Each class matches scattering physics:

• Buildings keep a stable phase. Dense buildings return -5.8 dB from double bounce, and coherence falls only from 0.59 at 6 days to 0.51 at 48.
• Port terminals are brightest and change fastest. Industrial yards return the highest VV at -2.5 dB, consistent with stacked containers and cranes acting as corner reflectors. Their coherence decays from 0.42 to 0.27, consistent with containers moving between passes.
• Vegetation scatters in volume. VH sits 6.3 dB below VV, the smallest gap of any class. The gap is 8.1 dB for dense buildings and 11.1 dB for water. Canopy scattering rotates the polarization of part of the wave, which narrows the gap. Coherence falls fastest, to 0.20 after 48 days.
• Water is dark and incoherent. Calm water reflects the pulse away, and its surface changes between passes, so coherence is 0.03 at every interval.

One threshold for water and land

A second query bins VV backscatter in 1 dB steps for water and land. Of 23,637 open-water pixels, 20,291 fall below -15 dB. Of 40,875 land pixels, 507 (1.2%) do. The 448 water pixels at -5 dB or brighter are consistent with the ships at anchor in San Pedro Bay and the harbor structures.

That split is the starting point of SAR flood mapping. On a post-event scene, RS_MapAlgebra applies the same threshold to mark new water. RS_ZonalStats summarizes that water mask under each Overture transportation segment. The SpatioTemporal Asset Catalog (STAC) is a standard for imagery catalogs. The STAC reader loads scenes from an imagery catalog, one row per scene. The AWS Glue Data Catalog announcement applies this to supply chain disruption. It joins Sentinel-1 imagery and National Oceanic and Atmospheric Administration (NOAA) HRRR (High-Resolution Rapid Refresh) weather forecasts to the Overture road network to score each corridor. RasterFlow (Public Preview) builds mosaics from imagery, runs computer vision inference on them, and writes the results as vectorized outputs on the same platform.

Language models were trained on text, documents, databases, and the internet, which hold no record of last week's flood. The Model Context Protocol (MCP) is a protocol that lets AI agents call data tools. Through the Wherobots MCP Server, an AI agent can run a query like the one above and return the closed road segments.

Synthetic aperture radar research

From Doppler beam sharpening to Seasat

Carl Wiley of Goodyear proposed in 1951 that the Doppler shifts of radar echoes could sharpen a side-looking radar's resolution along its track. He later described the principle in "Synthetic aperture radars: a paradigm for technology evolution" (IEEE Transactions on Aerospace and Electronic Systems, 1985). The first civilian SAR in orbit flew on NASA's Seasat in 1978. Its L-band radar imaged the oceans for about three months before a power failure.

Interferometry and coherence

Graham proposed SAR interferometry for topographic mapping in 1974 (Proceedings of the IEEE). Goldstein, Zebker, and Werner solved two-dimensional phase unwrapping (Radio Science, 1988), the step that turns wrapped phase, which repeats every wave cycle, into height. Gabriel, Goldstein, and Zebker (1989) introduced differential interferometry, which removes topography and leaves surface motion of a centimeter or less. Rosen and colleagues (Proceedings of the IEEE, 2000) wrote the standard review.

Every InSAR method needs coherence. Zebker and Villasenor (1992) separated the sources of decorrelation: baseline geometry, thermal noise, and change in the surface between passes. The port table measures the last of those directly.

Persistent scatterers and small baselines

Single interferograms carry atmospheric noise. Ferretti, Prati, and Rocca (2001) picked out permanent scatterers, points such as building corners that stay coherent for years, and estimated millimeter-scale motion at each one. Berardino, Fornaro, Lanari, and Sansosti (2002) built the small baseline subset (SBAS) method, which combines many interferograms with short orbital separations to recover deformation over distributed targets. Both depend on stable urban phase like the 0.51 coherence the port buildings keep after 48 days.

Change detection, damage, and floods

Backscatter change detection compares two dates with a log ratio. Speckle multiplies the signal, and the logarithm turns it into addition. Bazi, Bruzzone, and Melgani (2005) set the threshold on the log-ratio image automatically from a generalized Gaussian model. For floods, Wagner and colleagues (Remote Sensing of Environment, 2026) review the scientific challenges of the fully automatic Copernicus Global Flood Monitoring service. Hard cases for any threshold include floods under vegetation and in cities, where double bounce brightens floodwater, and dry sand, which can be as dark as water.

Machine learning on SAR

Zhu and colleagues (2021) review deep learning for SAR and the pitfalls that speckle, side-looking geometry, and complex values create for methods designed for optical images. Labeled data sets set the benchmarks: Sen1Floods11 (Bonafilia and colleagues, 2020) for floods, and xView3-SAR (Paolo and colleagues, 2022) for dark vessels. SSL4EO-S12 (self-supervised learning for Earth observation with Sentinel-1 and Sentinel-2; Wang and colleagues, 2023) pairs Sentinel-1 and Sentinel-2 patches from 251,079 locations for self-supervised pretraining. It is a basis for geospatial AI foundation models that read radar and optical data together.

What scale changes

A SAR scene is a raster, and the decisions it supports are about vectors: which buildings sit in the flood, which road segments closed, which terminals slowed down. Answering them takes a distributed spatial join of millions of radar pixels to millions of features. Jia Yu, Jinxuan Wu, and Mohamed Sarwat introduced GeoSpark (ACM SIGSPATIAL 2015) for spatial data processing on Apache Spark. Yu, Zongsi Zhang, and Sarwat described its partitioning, indexing, and join design in Spatial data management in Apache Spark (GeoInformatica, 2019). GeoSpark became Apache Sedona, which runs the pixel-to-polygon joins in the port analysis. WherobotsDB is 100% code compatible with Sedona across all spatial functions.

Analysis-ready formats make those joins repeatable. CEOS Normalised Radar Backscatter fixes the calibration. The STAC SAR extension records polarizations, frequency band, and instrument mode, so a query can filter scenes on catalog metadata before reading any pixel.

Open problems

• Atmosphere and unwrapping. Water vapor delays the signal, and unwrapping errors propagate across a scene. Both limit deformation accuracy over large areas.
• Geometry. Layover and shadow hide parts of cities and mountains from any one orbit, so complete coverage needs ascending and descending passes.

Conclusions

At the ports, SAR backscatter and coherence separated water, buildings, industrial yards, and vegetation.

1. SAR backscatter gives a starting point for water and flood extent: a VV threshold near -15 dB. At the ports it caught 85.8% of 23,637 open-water pixels and misclassified 1.2% of 40,875 land pixels. Ships and harbor structures make some water pixels bright, so check bright water against a vessel or infrastructure layer.
2. SAR coherence measures change between passes, so read it beside backscatter. Industrial yards were brightest at -2.5 dB VV, and their coherence fell from 0.42 to 0.27. Dense buildings held 0.59 to 0.51. For C-band subsidence monitoring, pick targets that keep a stable phase like the port buildings. For activity monitoring, use coherence loss.
3. In SAR data, vegetation shows a small cross-pol gap and short coherence. VH sat 6.3 dB below VV, and coherence fell to 0.20 at 48 days. For InSAR over forest, plan on L-band.
4. SAR pixels need vector context before interpretation. A dark pixel could be water, pavement, or dry sand. The classes above came from joining 64,512 pixel footprints to Overture buildings, water, land cover, and land use, and the vector labels identified the water. The 18,681 pixels outside the labeled classes stayed in an "other land" group.
5. These SAR medians come from summer 2020 composites (mean backscatter, median coherence) at one harbor. Averaging scenes reduced the speckle a single acquisition carries, so a threshold on a single scene needs a wider margin.

Related terms

• Earth observation data: the wider imagery family
• Change detection: comparing scenes across dates
• Flood risk data: SAR flood extents in exposure analysis
• Digital elevation model: the terrain input for SAR correction
• LiDAR: light detection and ranging, the laser counterpart
• Geospatial AI: models that detect ships and floods
• NDVI: the optical vegetation index (normalized difference vegetation index)
• Sentinel-2: the optical Copernicus mission
• Raster vs. vector data: SAR scenes are rasters; extracted ships and floods are vectors
• Satellite imagery analysis: the wider set of methods

Read more from Wherobots

• Wherobots now federates with the AWS Glue Data Catalog
• Measuring the Strait of Hormuz shutdown with Sentinel-2, WherobotsDB and Overture Maps
• Raster processing at scale: the out-of-database architecture behind WherobotsDB
• RasterFlow is now available in Public Preview

Join SAR flood extents and ship detections to roads and buildings with a Wherobots free trial at cloud.wherobots.com.