Flood Risk Mapping with GIS and Remote Sensing: Methods, Data and Best Practice

In fact, floods are the most frequent natural disaster worldwide. In Pakistan, the 2022 monsoon floods affected more than 33 million people and submerged large parts of Sindh and Balochistan for weeks. Consequently, events on that scale make the case for flood mapping plainly: you cannot plan defences, warn communities or target relief without knowing where water goes.

"Flood map" can mean several very different products, and confusing them causes real problems. This guide explains the three main types, the data behind each, the methods we use, how to check whether a map is any good, and how the results feed into decisions.

Flood extent, susceptibility and risk

In general, the usual framework is simple to state: risk is a function of hazard, exposure and vulnerability. However, the work is in measuring each part well.

Data used in flood mapping

• Digital elevation models. The foundation of any terrain-based analysis. For example, slope, flow direction, flow accumulation and height above the nearest drainage channel (HAND) are all derived from it.
• River and drainage networks. Mapped rivers, canals and drains, plus channels extracted from the elevation model.
• Rainfall. Gauge records where available, and satellite products such as CHIRPS or GPM IMERG for long, consistent coverage.
• Land cover and soils. Built-up surfaces and clay soils generate more runoff, whereas forests and sandy soils absorb more.
• Satellite imagery. Sentinel-1 radar for flood extent in cloudy conditions, Sentinel-2 and Landsat for clear-sky events and historical floods.
• Exposure data. Population grids such as WorldPop, building footprints, roads, schools, health facilities and cropland.
• Historical records. Past flood extents from agencies, the Copernicus Emergency Management Service or UN-SPIDER activations, plus local reports.

Common flood mapping methods

Satellite flood extent mapping with radar

For most monsoon floods, therefore, Sentinel-1 is the first choice. The workflow we use in Google Earth Engine:

1. Select Sentinel-1 ground range detected images in VV and VH polarisation from before and during the flood, on the same orbit track.
2. Then apply speckle filtering to reduce the grainy noise typical of radar.
3. Afterwards, identify water as pixels with very low backscatter, using a threshold set from the image histogram (Otsu's method works well) or from known water samples.
4. Also subtract permanent water, taken from the pre-flood image or the JRC Global Surface Water dataset, to isolate newly flooded areas.
5. Similarly, remove false water on steep slopes and in radar shadow using the elevation model and a HAND mask.
6. Summarise flooded area by district, and overlay population and cropland to estimate who and what the flood affected.

Radar has known blind spots. Flooded forest and tall crops can appear bright rather than dark, and flooded streets between buildings are hard to detect. Therefore, these areas need optical imagery, local reports or careful interpretation.

Optical water indices

When skies are clear, Sentinel-2 and Landsat give sharper, easier-to-read results. MNDWI, calculated from green and shortwave infrared bands, separates water from built-up land better than the original NDWI. Optical archives are also the best way to build a history of past floods.

Machine learning classification

In addition, Random Forest or deep learning models trained on labelled water and non-water samples can combine radar, optical and terrain inputs and generalise across many events. Our automated flood water detection project uses this approach to reduce manual thresholding.

Multi-criteria susceptibility analysis

Where there is no hydraulic model, however, susceptibility mapping gives a practical first picture. To begin with, the analyst turns each factor into a raster layer, rescales it to a common score, then weights and combines the layers:

The weights above are illustrative. On a real project they come from stakeholder workshops, the Analytic Hierarchy Process or statistical methods such as frequency ratio, and you should test them against observed floods. Our flood susceptibility mapping for a river catchment followed this approach.

Hydrological and hydraulic modelling

When the question is "how deep will the water be in a 1-in-100-year flood", you need a model. Hydrological models estimate how much water reaches the river; hydraulic models such as HEC-RAS route that water across the terrain and calculate depth and velocity. Meanwhile, GIS prepares the terrain, land roughness and boundary data and presents the results.

Why elevation data matters so much

In flat floodplains, such as the Indus plains, a difference of one or two metres in elevation can separate land that floods every year from land that rarely floods. Also, free global elevation models have vertical errors of a few metres and include buildings and trees in places. However, FABDEM removes much of the forest and building bias from Copernicus GLO-30 and is a better starting point for flood work. For town-scale or engineering studies, therefore, LiDAR or detailed survey data is worth the cost.

Validating flood maps

After all, a flood map that nobody has tested against real floods is a hypothesis. We validate in three ways:

• Against observed extents. Specifically, compare predicted high-susceptibility zones with flood extents mapped from satellite imagery for past events. For example, a common summary is the area under the ROC curve; values above 0.8 indicate the map discriminates well between flooded and non-flooded areas.
• Against independent samples. Check a random sample of mapped water and non-water pixels against high-resolution imagery for extent maps, and report overall, producer's and user's accuracy.
• With local knowledge. Finally, walk through results with people who know the area: district officers, irrigation engineers, community representatives. They often spot errors no dataset reveals.

How flood maps support decisions

• Land-use planning. Steering new housing and critical facilities away from high-risk zones.
• Infrastructure investment. Prioritising drains, embankments, culverts and retention areas where they protect the most people and assets.
• Early warning and preparedness. Identifying communities, roads and shelters likely to be affected so agencies can plan evacuation routes and stockpiles.
• Rapid response. Producing extent maps within hours of an image so agencies can direct relief where water actually is.
• Insurance and finance. Pricing risk, designing parametric insurance and screening investments.
• Recovery. Tracking how fast water recedes and which fields farmers can replant, using repeat imagery.

Getting started

The right product depends on the decision. For an active event, start with a radar-based extent map. However, for district planning a well-validated susceptibility map is often enough. For designing a bridge or setting floor levels, commission a hydraulic model. If you are not sure which you need, our environment and climate and remote sensing teams can review your data and recommend a workflow.