SAR: Seeing Through Clouds with Radar

Your first radar image

Let's load a Sentinel-1 image over Houston, Texas, and see what the world looks like in radar. Fair warning: SAR images are grayscale and look "grainy" because of speckle noise. They won't look like the pretty color composites we're used to, but don't worry, we'll make sense of it.

// Load a Sentinel-1 GRD image over Houston, TX
var houston = ee.Geometry.Point([-95.37, 29.76]);

var s1 = ee.ImageCollection('COPERNICUS/S1_GRD')
  .filterBounds(houston)
  .filterDate('2023-06-01', '2023-06-30')
  .filter(ee.Filter.eq('instrumentMode', 'IW'))
  .filter(ee.Filter.listContains('transmitterReceiverPolarisation', 'VV'))
  .filter(ee.Filter.listContains('transmitterReceiverPolarisation', 'VH'))
  .filter(ee.Filter.eq('orbitProperties_pass', 'DESCENDING'))
  .first();

Map.centerObject(houston, 10);
Map.addLayer(s1.select('VV'), {min: -25, max: 0}, 'VV polarization');
Map.addLayer(s1.select('VH'), {min: -30, max: -5}, 'VH polarization');

What makes SAR different?

Think of it like this. All the sensors we've used so far (Landsat, Sentinel-2, MODIS) are passive sensors. They sit up in space and wait for the Sun to illuminate the ground, then record what bounces back. They only work during the day, and clouds block the signal.

SAR is an active sensor. It brings its own flashlight: a radar transmitter that sends microwave pulses toward the ground. The sensor then records the returning echoes (backscatter). Because microwaves pass right through clouds, SAR works in any weather and at any time of day. That's a game-changer.

Working with Sentinel-1 in Earth Engine

Sentinel-1 is a constellation of two C-band SAR satellites operated by the European Space Agency (ESA). In GEE, we access it through ee.ImageCollection('COPERNICUS/S1_GRD'). The data comes in decibels (dB), which is a logarithmic scale. Here's how to read it:

• High values (near 0 dB): Strong backscatter (rough surfaces, buildings, vegetation).
• Low values (below -20 dB): Weak backscatter (smooth surfaces, calm water).

The filters you always need

Here's something important: unlike Landsat, Sentinel-1 requires several metadata filters to get consistent results. Always apply all of these. Think of it as a checklist before takeoff:

// Standard Sentinel-1 filtering template
var s1Collection = ee.ImageCollection('COPERNICUS/S1_GRD')
  .filterBounds(myPoint)
  .filterDate('2023-01-01', '2023-12-31')
  // IW mode = standard land imaging mode
  .filter(ee.Filter.eq('instrumentMode', 'IW'))
  // Require both VV and VH polarizations
  .filter(ee.Filter.listContains('transmitterReceiverPolarisation', 'VV'))
  .filter(ee.Filter.listContains('transmitterReceiverPolarisation', 'VH'))
  // Pick one orbit direction for consistency
  .filter(ee.Filter.eq('orbitProperties_pass', 'DESCENDING'));

print('Number of images:', s1Collection.size());

VV vs. VH: what's the difference?

Sentinel-1 transmits radar pulses in the vertical (V) direction and records echoes in both vertical (V) and horizontal (H) directions. This gives us two polarization bands:

• VV (co-polarization): Transmitted V, received V. Sensitive to surface roughness and soil moisture. Water appears very dark.
• VH (cross-polarization): Transmitted V, received H. Sensitive to volume scattering (vegetation canopy structure). Forests appear brighter in VH than bare soil.

// SAR false-color composite
var vv = s1.select('VV');
var vh = s1.select('VH');
var ratio = vv.divide(vh).rename('VV_VH_ratio');

var sarComposite = vv.addBands(vh).addBands(ratio);
Map.addLayer(sarComposite, {
  min: [-25, -30, 0.3],
  max: [0, -5, 1.5]
}, 'SAR Composite (VV, VH, ratio)');

Dealing with that grainy texture

You've probably noticed that SAR images have a characteristic grainy texture called speckle. Here's the thing: speckle isn't random sensor noise. It's caused by the constructive and destructive interference of radar waves reflecting from many small scatterers within a single pixel. But for most analysis purposes, we need to smooth it out.

The simplest approach is spatial averaging using a focal (neighborhood) filter. Think of it like blurring a photo slightly to remove grain. A common choice is a 3x3 or 5x5 pixel mean or median filter.

// Simple speckle filtering using a focal mean
var vvSmooth = s1.select('VV').focal_mean({
  radius: 50,
  kernelType: 'circle',
  units: 'meters'
});

Map.addLayer(s1.select('VV'), {min: -25, max: 0}, 'VV (original)');
Map.addLayer(vvSmooth, {min: -25, max: 0}, 'VV (speckle filtered)');

Another approach is temporal averaging: take the median of multiple SAR images from the same orbit over several weeks. This reduces speckle without blurring spatial details, which is pretty slick.

// Temporal speckle reduction using multi-image median
var s1Median = s1Collection
  .select(['VV', 'VH'])
  .median();

Map.addLayer(s1Median.select('VV'), {min: -25, max: 0}, 'VV (temporal median)');

Mapping floods when clouds won't move

Here's where SAR gets really powerful. Flood mapping is the most widely used SAR application, and the logic is beautifully simple: water is smooth, smooth surfaces produce very low backscatter, so flooded areas appear much darker than surrounding land. By comparing a "before" image with a "during flood" image, we can pinpoint exactly what got inundated.

Case Study: Hurricane Harvey hits Houston

Hurricane Harvey dumped over 60 inches of rain on Houston in August 2017, causing catastrophic flooding. Optical satellites couldn't see a thing through the storm clouds, but Sentinel-1 mapped the flood extent clearly. Let's recreate that analysis:

// Flood mapping: Hurricane Harvey, Houston TX, August 2017
var houston = ee.Geometry.Point([-95.37, 29.76]);

// Standard Sentinel-1 filters
var s1Filters = function(col) {
  return col
    .filter(ee.Filter.eq('instrumentMode', 'IW'))
    .filter(ee.Filter.listContains('transmitterReceiverPolarisation', 'VV'))
    .filter(ee.Filter.eq('orbitProperties_pass', 'DESCENDING'));
};

// Before: June 2017 (dry baseline, temporal median to reduce speckle)
var before = s1Filters(ee.ImageCollection('COPERNICUS/S1_GRD')
  .filterBounds(houston)
  .filterDate('2017-06-01', '2017-06-30'))
  .select('VV')
  .median();

// During: August 30, 2017 (peak flooding, single image)
var during = s1Filters(ee.ImageCollection('COPERNICUS/S1_GRD')
  .filterBounds(houston)
  .filterDate('2017-08-29', '2017-09-05'))
  .select('VV')
  .mosaic();

// Change detection: pixels where backscatter dropped significantly
var difference = before.subtract(during);

// Threshold: areas where backscatter dropped by more than 3 dB are flooded
var floodMask = difference.gt(3);

Map.centerObject(houston, 10);
Map.addLayer(before, {min: -25, max: 0}, 'Before (June 2017)');
Map.addLayer(during, {min: -25, max: 0}, 'During flood (Aug 2017)');
Map.addLayer(floodMask.selfMask(), {palette: ['cyan']}, 'Flood extent');

// Calculate flooded area in km²
var floodArea = floodMask.multiply(ee.Image.pixelArea()).reduceRegion({
  reducer: ee.Reducer.sum(),
  geometry: houston.buffer(50000),
  scale: 10,
  maxPixels: 1e9
});
var areaKm2 = ee.Number(floodArea.get('VV')).divide(1e6);
print('Estimated flood extent (km²):', areaKm2);

What SAR can't do

SAR is powerful, but let's be honest about its limitations. Every tool has trade-offs:

• Speckle noise: All SAR images require noise reduction before analysis. It's just part of the deal.
• Geometric distortion: Radar images are acquired at an angle, causing foreshortening, layover, and shadow effects in mountainous terrain.
• Interpretation takes practice: SAR images look very different from optical images. Bright does not mean "light colored"; it means "rough" or "structured." It takes some getting used to.
• Wind roughening: Ocean and lake surfaces roughened by wind produce higher backscatter and can be confused with land.
• No spectral indices: We can't calculate NDVI, NDWI, or other spectral indices from SAR data. SAR and optical data complement each other.

Next Steps

• Lab 27: SAR Flood Mapping - Apply SAR change detection to independently map a flood event.
• GEE Best Practices - Optimize your scripts for large-scale SAR processing.