Monitoring Water with Multi-Temporal NDWI in Google Earth Engine: A Pokhara Lekhnath Guide

Why monitor water with NDWI in Pokhara Lekhnath?

Pokhara Lekhnath Metropolitan City in Kaski district holds several lakes, including Phewa, Begnas and Rupa, along with rivers, wetlands and irrigated rice land, all shaped by a strong monsoon cycle. Water extent changes with season, sedimentation and land-use change, and much of that change is gradual enough to be missed without consistent records. Satellite water indices provide such a record at no data cost.

Bishal Neupane carried out a multi-temporal NDWI analysis of Pokhara Lekhnath using Sentinel-2 imagery in Google Earth Engine (GEE). As described on his methods page and CV, the analysis tracked seasonal water presence, identified seasonal pulses of water presence and persistent negative NDWI patterns, and produced maps that make water-body changes visible for decision-makers, with relevance to urban resilience, sustainable tourism and conservation. He summarised it in a LinkedIn post. This guide explains how to build that kind of analysis and how to read the results.

McFeeters NDWI vs Gao NDWI (NDMI) vs MNDWI

"NDWI" refers to more than one index, and mixing them up is the most common error in water studies.

McFeeters designed his index to delineate open water: water reflects more green than near-infrared light, so it gives positive values, while vegetation and soil reflect strongly in the NIR and give negative values. Gao's index answers a different question, namely how much water leaves hold, and is useful for crop water stress rather than lake mapping. Xu's modification swaps NIR for SWIR to reduce confusion between water and built-up surfaces, which is relevant in a growing city. Formulas and band mappings follow the Awesome Spectral Indices catalogue. Note that B11 is a 20 m band, so NDMI and MNDWI carry less spatial detail than McFeeters NDWI at 10 m.

How to interpret negative NDWI values

With McFeeters NDWI, negative values are the normal state of land. Vegetated fields, forest and dry soil are all negative, with dense green vegetation usually more strongly negative because of its high NIR reflectance. So a "persistent negative NDWI pattern" means an area that consistently behaves like land rather than open water across the time series. Whether that is ecologically interesting depends on context:

• Persistently negative pixels on a lake margin that were once positive can indicate shoreline change, sediment deposition or encroachment.
• Persistently negative pixels in cropland are expected and say nothing about soil moisture on their own; use NDMI or field data for moisture questions.
• Values near zero are ambiguous: mixed water–land pixels, turbid or shallow water, aquatic vegetation, or flooded rice fields early in the season.

A threshold of 0 is the textbook dividing line for open water, but turbidity, shadows from steep terrain and thin cloud can shift the best threshold. Checking a few known water and land pixels before fixing a threshold is worth the time.

Step 1: Build a clean NDWI collection in Google Earth Engine

The script below uses the harmonised Sentinel-2 surface reflectance collection, masks cloud, shadow and cirrus with the Scene Classification Layer, and computes both McFeeters NDWI and NDMI for every scene.

// aoi: your boundary, e.g. an uploaded Pokhara Lekhnath polygon asset function prep(img) { var scl = img.select('SCL'); var clear = scl.neq(3).and(scl.neq(8)).and(scl.neq(9)).and(scl.neq(10)); var r = img.select(['B3', 'B8', 'B11']).updateMask(clear).divide(10000); return r.normalizedDifference(['B3', 'B8']).rename('NDWI') .addBands(r.normalizedDifference(['B8', 'B11']).rename('NDMI')) .copyProperties(img, ['system:time_start']); } var s2 = ee.ImageCollection('COPERNICUS/S2_SR_HARMONIZED') .filterBounds(aoi) .filterDate('2019-01-01', '2025-12-31') .filter(ee.Filter.lt('CLOUDY_PIXEL_PERCENTAGE', 40)) .map(function (img) { return ee.Image(prep(img)); });

Step 2: Seasonal composites and water area

Single scenes are noisy and often cloudy, especially in the monsoon. Seasonal median composites smooth that out and give a fair comparison between periods. The example uses three seasons for one year; repeat for several years to separate a trend from normal seasonal variation.

// Seasonal median composites for one year function season(start, end, label) { return s2.filterDate(start, end).median().clip(aoi) .set('season', label).set('system:time_start', ee.Date(start).millis()); } var pre = season('2024-03-01', '2024-06-01', 'pre-monsoon'); var mon = season('2024-06-01', '2024-10-01', 'monsoon'); var post = season('2024-10-01', '2025-01-01', 'post-monsoon'); // Open-water area (NDWI > 0) in hectares function waterHa(img) { var water = img.select('NDWI').gt(0); var area = water.multiply(ee.Image.pixelArea()).divide(10000); return area.reduceRegion({reducer: ee.Reducer.sum(), geometry: aoi, scale: 10, maxPixels: 1e10}); } print('Pre-monsoon water (ha)', waterHa(pre)); print('Monsoon water (ha)', waterHa(mon)); print('Post-monsoon water (ha)', waterHa(post)); // Time series of mean NDWI across all clear scenes print(ui.Chart.image.series({imageCollection: s2.select('NDWI'), region: aoi, reducer: ee.Reducer.mean(), scale: 20})); Map.centerObject(aoi, 12); Map.addLayer(post.select('NDWI'), {min: -0.6, max: 0.4, palette: ['brown', 'white', 'blue']}, 'NDWI post-monsoon');

The water-area function counts pixels with NDWI above 0, multiplies by pixel area and converts to hectares. Comparing the same season across years shows long-term change; comparing seasons within a year shows the seasonal pulse.

Practical problems you will meet

• Monsoon clouds. June to September has few clear optical scenes. Expect gaps, widen the date window, or complement with Sentinel-1 radar, which sees through cloud.
• Terrain shadow. Steep slopes around the valley can produce dark pixels that resemble water. A slope or hillshade mask helps.
• Urban surfaces. Some roofs and roads can return positive McFeeters NDWI. MNDWI reduces this confusion.
• Turbid water. Sediment-laden water after heavy rain raises NIR reflectance and can lower NDWI towards zero.
• Validation. Compare the water mask with high-resolution imagery or field observations before reporting areas.

Why this matters for agriculture and planning

For farming around Pokhara, seasonal water maps show which irrigated and lowland areas flood, when ponds and canals hold water, and how lake margins used for cultivation change. For planners, consistent maps turn anecdotal observations about shrinking or silting lakes into evidence. The same index stack also feeds land-cover work; see the Kaski K-Means land-cover guide and the Sentinel-2 vegetation indices guide for NDVI, SAVI, GCI and PSRI.

Researchers working on water, wetlands or irrigated agriculture in Nepal who would like to collaborate can contact Bishal through the contact page.