Hydrology & Water Mapping

Your First Water Map: Lake Okeechobee in 12 Lines

Lake Okeechobee is the largest freshwater lake in Florida and a critical piece of the Everglades ecosystem. Let's map it in under 15 lines of code using the Modified Normalized Difference Water Index (MNDWI) from Sentinel-2. Let's see it in action.

// Map Lake Okeechobee using MNDWI from Sentinel-2
var lake = ee.Geometry.Point([-80.83, 26.95]);

var s2 = ee.ImageCollection('COPERNICUS/S2_SR')
  .filterBounds(lake)
  .filterDate('2023-03-01', '2023-03-31')
  .sort('CLOUDY_PIXEL_PERCENTAGE')
  .first()
  .divide(10000); // scale to reflectance

// MNDWI = (Green - SWIR) / (Green + SWIR)
var mndwi = s2.normalizedDifference(['B3', 'B11']).rename('MNDWI');

// Create a binary water mask using threshold > 0
var waterMask = mndwi.gt(0);

Map.centerObject(lake, 11);
Map.addLayer(mndwi, {min: -0.5, max: 0.8, palette: ['brown', 'white', 'cyan', 'blue']}, 'MNDWI');
Map.addLayer(waterMask.selfMask(), {palette: ['0000FF']}, 'Water Mask');

What Can We Do with a Water Map?

Think about it: flood responders need rapid maps of inundated areas to know where to send boats. Water resource managers track reservoir levels to plan irrigation and city water supply. Climate scientists measure how lakes and wetlands are shrinking or expanding over decades.

Traditional water monitoring relies on ground-based gauges and manual surveys. Both are expensive, and both only cover a tiny slice of the landscape. Satellite-based approaches give us synoptic, repeatable coverage of entire watersheds at no additional cost per observation. That's a game changer.

• Floods: Map inundation extent within hours of satellite overpass.
• Drought: Track shrinking reservoirs and disappearing lakes over time.
• Water resources: Monitor irrigation storage, wetland health, and river width.
• Climate change: Document long-term trends in surface water availability.

Two Indices, One Goal: NDWI vs. MNDWI

Both indices exploit the same physical fact: water absorbs strongly in the shortwave infrared (SWIR), creating contrast with visible bands. The key difference? Which visible band we pair with SWIR.

NDWI (Gao, 1996)

Formula: (NIR - SWIR) / (NIR + SWIR)
Sentinel-2 bands: (B8 - B11) / (B8 + B11)

NDWI was originally designed to measure vegetation water content. It works for detecting open water, but here's the catch: built-up surfaces (think rooftops and parking lots) also produce positive NDWI values. That can fool your water map.

MNDWI (Xu, 2006)

Formula: (Green - SWIR) / (Green + SWIR)
Sentinel-2 bands: (B3 - B11) / (B3 + B11)

MNDWI swaps NIR for Green, which suppresses built-up features and enhances open water contrast. For most water body mapping tasks, MNDWI outperforms NDWI. Think of it as the upgraded version.

// Compare NDWI and MNDWI side by side
var ndwi = s2.normalizedDifference(['B8', 'B11']).rename('NDWI');
var mndwi = s2.normalizedDifference(['B3', 'B11']).rename('MNDWI');

var waterPalette = {min: -0.5, max: 0.8, palette: ['brown', 'white', 'cyan', 'blue']};
Map.addLayer(ndwi, waterPalette, 'NDWI');
Map.addLayer(mndwi, waterPalette, 'MNDWI');

Drawing the Line: Thresholding for Water Detection

A spectral index gives us continuous values (typically from -1 to 1), but a water map is categorical: each pixel is either water or not water. Thresholding is how we draw that line. Think of it like setting a passing grade on an exam.

The simplest threshold for MNDWI is zero. Pixels above zero get classified as water; pixels below zero become land. This works well in most cases, but we can fine-tune the threshold to reduce errors.

// Basic threshold at 0
var waterBasic = mndwi.gt(0);
Map.addLayer(waterBasic.selfMask(), {palette: ['blue']}, 'Water (threshold = 0)');

// Refined threshold at 0.2 (reduces false positives from shadows)
var waterRefined = mndwi.gt(0.2);
Map.addLayer(waterRefined.selfMask(), {palette: ['darkblue']}, 'Water (threshold = 0.2)');

How Big Is That Lake? Calculating Water Area

Once we have a binary water mask, we can calculate the total water surface area. ee.Image.pixelArea() creates an image where each pixel's value equals its area in square meters. We multiply the pixel area image by our water mask, then sum the result over the region of interest. Let's see how big Lake Okeechobee really is.

// Calculate water surface area in square kilometers
var waterMask = mndwi.gt(0);

// Create an area image (each pixel = its area in m²)
var areaImage = ee.Image.pixelArea();

// Multiply by water mask so only water pixels have area values
var waterArea = areaImage.updateMask(waterMask);

// Sum the area within the region
var roi = lake.buffer(40000); // 40 km buffer around the point
var totalArea = waterArea.reduceRegion({
  reducer: ee.Reducer.sum(),
  geometry: roi,
  scale: 10,
  maxPixels: 1e9
});

// Convert m² to km²
var areaKm2 = ee.Number(totalArea.get('area')).divide(1e6);
print('Water surface area (km²):', areaKm2);

Where Spectral Water Detection Can Trick You

Spectral indices are powerful, but they're not perfect. Several common features can produce false positives (classified as water when they're not) or false negatives (water that gets missed). Don't worry, once you know these pitfalls, they're easy to watch for.

• Cloud shadows: Dark shadows on land produce high MNDWI values and frequently get misclassified as water. Always apply cloud masking before water detection.
• Terrain shadows: Mountain shadows in low-sun-angle imagery mimic water signatures.
• Dark surfaces: Lava flows, asphalt, and some dark soils can look water-like to certain indices.
• Ice and snow: Frozen water bodies may not be detected because ice has a different spectral profile than liquid water.
• Turbid water: Sediment-laden water reflects more in the visible bands, reducing MNDWI values and potentially causing false negatives.
• Small water bodies: Features smaller than the pixel size (10 m for Sentinel-2) simply can't be resolved.

Next Steps

• JRC Global Surface Water - Explore 40 years of global water occurrence, transitions, and seasonality data.