Off-Nadir Delta

Monitoring Water Surface Changes with NDWI and MNDWI Time Series

Kazushi Motomura1 April 2026(Updated: 26 October 2026)8 min read
Monitoring Water Surface Changes with NDWI and MNDWI Time Series

Quick Answer: NDWI (Green−NIR)/(Green+NIR), proposed by McFeeters in 1996, and MNDWI (Green−SWIR)/(Green+SWIR), proposed by Xu in 2006, highlight open water in satellite imagery. Monitoring these indices over a polygon covering a lake, reservoir, or floodplain reveals seasonal cycles in water surface extent, drought-driven drawdown, and flooding events. MNDWI was designed to suppress the built-up land noise that makes NDWI overestimate water near cities, so it is the safer choice where water bodies sit among buildings. They measure surface extent, not volume, and clouds and tides need separate handling.

Water from Space: Why Index Monitoring Matters

Water bodies are among the most dynamic features on Earth's surface. Lakes rise and fall with seasons and droughts. Reservoirs fill and drain with demand and precipitation. Floodplains inundate and recede with storms and snowmelt. Wetlands expand in wet years and contract in dry ones.

Tracking all this variability with field measurements is expensive and geographically limited. Satellite-based water indices add area-wide coverage: Sentinel-2 revisits every 5 days at the Equator with its two-satellite constellation (currently Sentinel-2B and 2C, with 2A in an extension campaign), at 10 m for the green and NIR bands and 20 m for SWIR (Sentinel-2 mission, SentiWiki).

NDWI vs. MNDWI: Choosing the Right Index

NDWI — Normalized Difference Water Index

The original water index, proposed by McFeeters (McFeeters, 1996), uses green and NIR bands — on Sentinel-2, B3 (~560 nm) and B8 (~833 nm) (SentiWiki):

NDWI = (Green − NIR) / (Green + NIR)

Water reflects more in green wavelengths than in NIR, which it absorbs strongly, so water bodies show positive NDWI while vegetation (high NIR reflectance) shows much lower values.

Value interpretation (from the Sentinel Hub script documentation for Sentinel-2 NDWI, Sentinel Hub):

NDWITypical Surface
> 0.5Usually water bodies
~0 – 0.2Built-up areas
Much lowerVegetation

Treat these as starting points. Turbid or shallow water, sun glint and mixed shoreline pixels all move the values, so the threshold that separates water from land should be checked against imagery for each site.

Limitation: In urban areas, buildings and roads can have positive NDWI values that contaminate water detection — Xu found that NDWI often overestimates water area for this reason (Xu, 2006). Near vegetation-fringed water bodies, the vegetation signal may suppress NDWI.

MNDWI — Modified NDWI

MNDWI, proposed by Xu (Xu, 2006), replaces NIR with a shortwave-infrared band — on Sentinel-2, B11 (~1610 nm):

MNDWI = (Green − SWIR) / (Green + SWIR)

Water absorbs SWIR strongly as well. Xu designed the index to enhance open water while suppressing — and in many cases removing — the built-up land noise that affects NDWI (Xu, 2006). This means:

  • Water stays positive
  • Built-up surfaces drop to lower values than they have in NDWI
  • Vegetation and soil are also suppressed

MNDWI is preferred when:

  • Monitoring urban water bodies surrounded by buildings
  • Coastal monitoring where saltmarsh and tidal flats are present
  • Areas with mixed land use near the water body

For simple open-water lake monitoring in rural areas, NDWI and MNDWI typically give similar results.

NDMI — Vegetation Moisture Index

While not strictly a water index, NDMI (Normalized Difference Moisture Index) responds to water content in vegetation canopies. The same NIR–SWIR formula was published by Gao as an index of vegetation liquid water — confusingly, also under the name NDWI (Gao, 1996):

NDMI = (NIR − SWIR) / (NIR + SWIR)

Because NDMI follows canopy water rather than greenness, it carries information NDVI does not. Whether it shows drought stress earlier than NDVI depends on the vegetation and how the stress develops, so compare the two over your own site before relying on either as an early warning.

What Water Index Time Series Reveals

Reservoir Storage Levels

Draw a polygon over a reservoir and monitor MNDWI over time. As water levels drop, exposed shores and mudflats appear within the polygon — lowering the average MNDWI value. This gives you a proxy for reservoir storage level.

Comparing year-to-year MNDWI seasonal cycles reveals:

  • Whether drought years show earlier drawdown
  • Whether downstream water use is increasing
  • Whether the filling pattern after monsoon or snowmelt is changing

Important caveat: MNDWI measures water surface area within the polygon, not volume. For reservoirs with sloped sides, smaller surface area means much less volume — the relationship is non-linear and site-specific.

Flood Extent Mapping

During and after major flooding events, NDWI/MNDWI spikes sharply upward as floodwaters cover normally dry land. The time series shows:

  1. Pre-flood baseline (normal dry-land values)
  2. Flood peak (high NDWI/MNDWI across the polygon)
  3. Recession (gradual return to baseline over days to weeks)

The rate of recession indicates drainage speed, which is relevant for agricultural loss assessment (crops flooded for longer suffer more damage) and disaster response planning.

Wetland Dynamics

Seasonal wetlands — marshes, playas, and ephemeral lakes — show strong NDWI/MNDWI cycles tied to precipitation. Monitoring these systems reveals:

  • Whether wet season flooding extent is increasing or decreasing
  • The duration of inundation within annual cycles
  • Long-term drying trends associated with climate change or upstream diversions

Coastal and Tidal Zones

Tidal flats are alternately exposed and inundated — once a day where tides are diurnal, twice a day where they are semidiurnal or mixed (NOAA) — making single-image assessment highly dependent on tidal timing. Time series monitoring averages over many tidal states to reveal the mean intertidal zone and detect long-term changes from sea level rise, sediment dynamics, or land reclamation.

Practical Monitoring Setup

For Lake or Reservoir Monitoring

Draw your polygon to cover the entire water body including some buffer around the typical shoreline. This ensures that as water levels drop and the shoreline retreats, the exposed area remains within your polygon and suppresses the average MNDWI.

Set the start date to at least 12–24 months back to capture at least one full annual cycle.

For Flood Monitoring

For proactive flood monitoring, set up polygons over:

  • Floodplains in flood-prone river basins
  • Low-lying coastal areas
  • Urban areas with known flooding history

Use SAR (CR or VV) monitoring alongside NDWI/MNDWI — SAR observations are independent of daylight and cloud, and smooth open water returns almost nothing to the antenna, so it appears dark (UN-SPIDER).

For Drought Monitoring

Combine:

  • NDWI/MNDWI over the reservoir or lake to track storage
  • NDMI over surrounding vegetation to track soil and plant moisture stress
  • NDVI over agricultural areas to track crop condition

This multi-index approach gives a comprehensive picture of drought progression from water supply (reservoir) through soil moisture (NDMI) to plant stress (NDVI).

Interpreting Seasonal Patterns

Water index time series often show strong seasonal patterns that should not be confused with anomalies:

Monsoonal climates: Sharp NDWI rise during monsoon season followed by gradual drawdown. This is completely normal — the anomaly would be a smaller-than-normal monsoon peak.

Snowmelt-fed systems: Spring NDWI peak from snowmelt, followed by summer drawdown. Later or earlier peak timing reflects climate variation.

Semi-arid ephemeral lakes: These may be completely dry for most of the year with only brief inundation after major rainfall events. The time series shows long periods near zero with occasional spikes.

Cloud Contamination and Data Quality

Clouds over water bodies create a special challenge: clouds are also highly reflective, and poorly masked clouds can produce false high NDWI values. To check:

  • Look at the true-color imagery for any data point with unexpectedly high NDWI
  • Use SAR VV (which drops over smooth open floodwater and is unaffected by cloud) as a cross-check

A sudden NDWI spike not matched by a SAR VV drop is more likely cloud contamination than actual flooding — with one caveat: SAR has its own difficulties with flooded vegetation and with floods in built-up areas, where backscatter from houses and other objects limits accuracy (UN-SPIDER), so check the land cover before treating a missing VV drop as proof.

Applications

Water resource management: Track reservoir storage trends for hydropower planning, irrigation water allocation, and municipal supply forecasting.

Flood disaster response: Map flooding extent from the first clear optical or SAR pass after an event and track inundation duration for damage assessment.

Wetland conservation: Monitor protected wetland extent over years, documenting whether conservation measures are working.

Climate change research: Track long-term changes in lake extent, wetland coverage, and seasonal water cycle timing.

Agriculture: Monitor irrigation canal and field flooding patterns for water use auditing.

Summary

NDWI and MNDWI time series monitoring provides continuous tracking of water surface extent for lakes, reservoirs, floodplains, and coastal wetlands. The time series reveals seasonal storage cycles, drought-driven drawdown, and flood events — changes that require months or years of data to distinguish normal variability from genuine long-term change. MNDWI is generally more accurate than NDWI near urban areas and coastlines, while NDMI provides complementary information on vegetation water stress. For flood monitoring specifically, pairing optical water indices with SAR ensures cloud-free detection even during the storm conditions that typically accompany floods.

Track water surface changes continuously with the Monitoring Dashboard, or for immediate flood extent mapping after an event, see SAR Flood Mapping.

Kazushi Motomura
Kazushi Motomura

Remote sensing specialist with 10+ years in satellite data processing and AI. Founder of Off-Nadir Lab. Ph.D. in Informatics (Yokohama National University, 2026). Master's in Earth System Science and Technology (Kyushu University). Co-author, Remote Sensing Encyclopedia. More about the author →

From headline to satellite evidence

One connected intelligence workflow across four surfaces — free to start, no GIS software or remote-sensing background required.