NDWI

The Normalized Difference Water Index (NDWI) is a spectral index used to detect and monitor water in satellite imagery. Confusingly, two distinct formulations share the NDWI name — one designed by McFeeters (1996) for mapping open water bodies, and another by Gao (1996) for estimating vegetation moisture content. Both produce values from −1 to +1 and use a normalized difference formula, but they target fundamentally different phenomena and use different spectral bands. Understanding which NDWI is appropriate for a given application is critical for accurate analysis. The index is one of the most widely used tools in flood mapping, drought monitoring, and hydrological studies.

McFeeters' NDWI (1996) uses green and near-infrared bands: NDWI = (Green − NIR) / (Green + NIR). Water strongly absorbs NIR radiation while reflecting more green light, so water pixels produce positive values while vegetation and soil produce negative values. A threshold of 0 is the standard starting point for water/non-water classification, though optimal thresholds vary by region and can range from −0.05 to 0.3 depending on water turbidity, depth, and surrounding land cover.

Gao's NDWI (1996) uses NIR and shortwave infrared (SWIR) bands: NDWI = (NIR − SWIR) / (NIR + SWIR). The SWIR region (1.57–1.65 μm) is sensitive to liquid water absorption within leaf tissue, so this formulation quantifies vegetation water content rather than surface water extent. This version is sometimes called NDMI (Normalized Difference Moisture Index) to reduce confusion.

Xu (2006) later developed the Modified NDWI (MNDWI), substituting the green band with SWIR to better suppress built-up area false positives: MNDWI = (Green − SWIR) / (Green + SWIR). This variant is often preferred in urban environments where buildings can be misclassified as water by the original McFeeters formulation.

McFeeters' NDWI can produce false positives in urban areas where built-up surfaces have spectral signatures similar to water, which is why MNDWI was developed. Turbid or shallow water bodies may fall below detection thresholds, leading to underestimation of water extent. Cloud cover and cloud shadows can be misclassified as water, requiring pre-processing with cloud masks. The Gao formulation is sensitive to soil moisture background in sparse vegetation, which can confuse the vegetation water content signal. All NDWI variants are sensitive to sun angle effects and atmospheric conditions, and performance varies across sensor types due to differing band positions and widths.

Both NDWI formulations were independently published in 1996. Stuart McFeeters introduced his water body detection index in the International Journal of Remote Sensing, motivated by the need for automated surface water extraction from Landsat imagery. Bo-Cai Gao published his vegetation liquid water index in Remote Sensing of Environment, focused on measuring canopy moisture from imaging spectrometer data. The coincidental naming created lasting confusion in the literature. Hanqiu Xu proposed the Modified NDWI (MNDWI) in 2006 to address McFeeters' NDWI limitations in urban settings. Today, the Joint Research Centre's Global Surface Water dataset and the European Space Agency's World Water Body products both rely on NDWI-derived methods for global water mapping.