NDWI (Normalized Difference Water Index) is a water-sensitive vegetation/surface index based on the normalised difference between two reflectance bands. Two independent definitions share the same acronym, and confusing them is a common error. Both definitions produce values between −1 and +1.

The two core definitions

  • McFeeters NDWI (1996): (Green − NIR) / (Green + NIR). Used to extract surface water bodies (lakes, rivers, ponds). Water reflects the green band but strongly absorbs NIR, so the value tends to positive. Vegetation and soil reflect NIR, so they produce negative values.
  • Gao NDWI (1996): (NIR − SWIR) / (NIR + SWIR). Sensitive to leaf water content (Leaf Equivalent Water Thickness); used in agriculture and drought monitoring.

To reduce the confusion, MNDWI (Modified NDWI), proposed by Xu (2006), is widely adopted: (Green − SWIR) / (Green + SWIR). It outperforms the McFeeters formula at separating urban water surfaces from buildings.

Physical basis

The spectral reflectance of water falls rapidly with wavelength. In visible green (~560 nm) reflectance is weak but present; in near-infrared (NIR, 700–1300 nm) reflectance is near zero; in short-wave infrared (SWIR, 1400–2500 nm) absorption is again strong. Liquid water inside leaves has distinct absorption bands, primarily at 970, 1200, 1450 and 1940 nm, which form the basis of Gao NDWI and related indices (e.g. NDWI Gao, NMDI, NDII).

Sentinel-2 and satellite sources

The European Space Agency's Copernicus satellites Sentinel-2A (2015) and Sentinel-2B (2017), with a 5-day revisit and 13 spectral bands, are widely used for NDWI:

  • B3 (Green, 560 nm, 10 m): in McFeeters and MNDWI.
  • B8 (NIR, 842 nm, 10 m): in both McFeeters and Gao.
  • B8A (Narrow NIR, 865 nm, 20 m): less affected by atmospheric water vapour.
  • B11 (SWIR, 1610 nm, 20 m): for Gao and MNDWI.
  • B12 (SWIR, 2190 nm, 20 m): for soil-moisture and leaf-water indices.

Older systems (Landsat-5/7/8/9, MODIS) use different band centres; comparative studies require band calibration. Open commercial sources include PlanetScope and SkySat (3 m, daily).

Value ranges and interpretation

Typical interpretation of Gao NDWI:

  • −1 to −0.3: bare soil, structures, very dry vegetation.
  • −0.3 to 0: vegetation under mild water stress.
  • 0 to 0.3: moderate leaf water content, healthy vegetation.
  • 0.3 to 0.5+: high leaf water content; irrigated or naturally moist area.

For McFeeters NDWI a value of 0 is taken roughly as the water/land boundary; the threshold is often refined per region using automatic segmentation such as Otsu's method.

Applications

  • Water body mapping: tracking lake and reservoir surface area; flood extent mapping after events.
  • Crop water stress detection: moderate correlation with leaf water potential (r = 0.5–0.8, depending on species and season).
  • Drought monitoring: the year-on-year NDWI anomaly turns markedly negative in below-average rainfall seasons.
  • Wildfire risk analysis: NDWI and NDII (Normalized Difference Infrared Index) are used to estimate live and dead fuel moisture.
  • Wetland ecosystem monitoring: seasonal variation in water cover.
NDVI vs NDWI: NDVI measures chlorophyll density, NDWI measures water content. In the early stage of water stress a plant may continue to look green (NDVI is unchanged), while leaf water content has already started to fall (Gao NDWI dips first). The two indices are complementary and most monitoring systems use them together.

Limitations

Common factors that distort NDWI: cloud and shadow cover (cleaned by the Sentinel-2 SCL mask), atmospheric water vapour (corrected by Sen2Cor in Level-2A products), mixed-pixel effects (especially at the 20 m SWIR resolution on small parcels), and seasonal BRDF changes. Using a time series (e.g. a 30–60 day moving average) instead of a single image reduces noise.