Your Flood Map Looks Precise. The Elevation Data Underneath May Not Be: DEM Quality and Flood-Risk Accuracy

Your Flood Map Looks Precise. The Elevation Data Underneath May Not Be: DEM Quality and Flood-Risk Accuracy

Your flood map indicates that the road does not flood. However, the residents know that it does flood each monsoon.

It is possible for both of them to be telling the truth; the map illustrates what its elevation model was able to see, and the residents show what takes place when the water actually arrives. Located between these two is a grid of elevation values, some structures that might be missing, and a number of assumptions that have not been included in the legend.

A flood map is based on a model, and the quality of that map is limited by the data it is using. In this article we examine the errors that can occur in elevation data, explain how culverts and bridges can affect the results, and describe how it is possible to compare the map with actual conditions before anyone carries out construction, takes out insurance, or evacuates as a result of it. This is the same point we made previously about buried pipes in our earlier post on mapped versus located utilities. Flood maps suffer from the same problem, but in this case it involves water.

A clean flood map appears more certain than it actually is.

A crisp edge is a grid cell, not a measurement

The boundary on a flood map is made up of cells, each of which contains one elevation and one calculated water depth; the clear line indicates the position where a calculated value exceeded a threshold, but it shows no information about where the water was actually measured.

Wrong does not always mean too small

It would be helpful if inaccurate elevation data always led to an underestimation of the flood extent. However, it does not do so. A study carried out in a mountainous city in China showed that coarser DEMs had the effect of altering the results in opposite ways depending on the type of flood. For urban flash floods, the extent and depth along the flow path were underestimated since the buildings were treated differently and the path changed. With river overbank floods, on the other hand, the modified channel shape caused the flood extent to be overestimated.

The same conclusion is reached by other tests. A study of global digital elevation models in a densely built-up area found that they tended to overestimate the area affected by flooding and underestimate the depth, since about 70 per cent of the flooded area had depths less than 0.3 m, which was nearly twice the value given by the reference model. In Tainan City, a digital elevation model with a spacing of 40 m produced a flooded area that was about 1.5 times greater than that obtained with a 1 m DEM.

As it is possible for the error to go either way, there is no correction factor that can be applied later on; you must carry out the test.

What can an elevation model overlook?

Resolution against the features that steer water

Water tends to follow small features such as an embankment one metre wide, a drainage ditch, a road crown, and a shallow depression located behind a wall. A cell with a size of 30 metres is able to average out all of these. The China study suggests using a DEM resolution and a cell size finer than the width of the river.

It doesn't follow that a finer resolution is always superior. A study conducted in the United States on flood maps based on HAND found no significant difference in map quality when comparing 3 m to 20 m, but did observe a clear decline at 60 m and 90 m. The important question to ask is whether the cell size you use matches the width of the feature you are interested in.

Surface or bare earth

Flood models require information about the ground; free global digital elevation models record what the sensor observed, such as tree canopies and rooftops. When five 30 metre global digital elevation models were compared with 65 airborne lidar surveys, it was found that land cover, particularly that of forests, was the main cause of vertical errors. FABDEM, which is a version of the Copernicus DEM with forests and buildings removed, came out first. A surface that is too high in forested and built-up areas can cause the modelled water to flow to a place where it would not actually go.

Why do culverts and bridges change the answer?

The structure the terrain doesn't show

In a terrain model, a road on an embankment appears as if it were a wall. Actually, water flows through culverts and under bridges. If the model is not aware of these openings, it might stop water from flowing through them or divert the water to places it never should go.

A study carried out in 2026 and published in Natural Hazards and Earth System Sciences points out that blocked culverts are frequently ignored when carrying out flood modelling. The study also examined official culvert datasets to assess their completeness and discovered that some culverts were missing. Thus the record which indicates to a model where water crosses the road may itself be incomplete.

Blocked is not the same as open

In the study, the blockages caused noticeable changes to the local flood hydrographs and extents, although the effect over the whole catchment remained small. That is advantageous. A few culverts have a great impact on the street next to them but affect the catchment only slightly. This shows exactly where fieldwork should be carried out: investigate the structures upstream of those areas where people live and drive.

Bridges can also be a source of problems. If debris gets stuck upstream of a bridge, it can cause backwater effects, a phenomenon that a bare terrain model will not display.

Does elevation make up the entire flood model?

What else the model needs

The model is told by the elevation data where water can flow; it doesn't indicate how much water reaches each area, when it reaches it, or how much the ground takes in. The amount of rainfall or the flow of rivers, the boundary conditions at the edges of the model, infiltration, and surface roughness all affect the outcome. For instance, a HAND map is calculated solely from the terrain model and can show which cells sit low relative to the nearest channel, but it contains no information about the storm. These assumptions are part of the map and should be placed next to it.

Hazard is not risk

A hazard map shows where and how deep the water is. Risk adds exposure, which refers to who or what is present there, and vulnerability, which refers to the extent of their impact. The American Geosciences Institute gives a simple example in this case: two areas which have a similar hazard can have very different levels of risk.

Elevation errors are also carried over into the risk figures. In the study carried out in Kuala Lumpur, the damage estimates at the building level based on a 5 m DEM were almost twice as high as those based on a 12.5 m DEM and almost four times as high as those from the 30 m SRTM. The same study points out that coarser DEMs remain useful in areas where data are scarce and for large-scale assessments. A small variation in the terrain can still result in a big difference in the amount that someone includes in a budget.

How do you check if a flood map is correct?

Compare it with what actually happened

Since satellite radar can penetrate cloud cover, it is a practical method for verification. When a 2024 flood in China was validated by comparing a 2D model with extents derived from SAR, a critical success index of 72.7 percent was achieved. Regarding the 2022 floods in Pakistan, a Sentinel-1 analysis found that a total of 30,492 km² had been flooded at least once, with the greatest extent occurring on 30 August.

Satellite imagery also has its limitations, with timing being one of them. For example, in a study of the 1994 Piedmont flood in Italy, a radar image taken three days after the flood had only shown about 20 per cent of the land that in fact had flooded. Another limitation is vegetation: optical sensors are unable to see through clouds or beneath vegetation. The Copernicus emergency mapping service mentions the same limitations and states that floods may not be visible in the streets of densely populated urban areas.

Field evidence provides information that satellites cannot. The researchers are currently preparing ground-truth flood maps based on high-water marks recorded during field surveys, since these marks show the highest level that the flood reached after the event. Likewise, accounts given by residents of where the water went should be included in that pile.

On its own, the matching of extent is a weak test since a model can still get the depths wrong even if it covers about the right area, depth being what damage and risk estimates are based on.

A practical workflow

Here is the order we would follow:

  1. Begin with the decision; a plan for closing a road requires a different level of accuracy from a regional screening map.
  2. Match the DEM with the various features and determine whether it is a surface model or a bare-earth model, as well as check its vertical accuracy and cell size in relation to the channel width and embankment size.
  3. Survey the features that control the flow of water. This includes culverts, bridge openings, the crests of embankments, and the cross-section of the channel, which is why a topographic survey is worthwhile. For example, during the Batakundi-Naran hydropower survey we combined a bathymetric survey of the river with drone-generated DTM and DEM products so that the river bed was measured rather than estimated from the banks.
  4. Carry out sensitivity analyses. Apply the same storm scenario with different DEMs and with the culverts both left open and blocked. If the results change, you have established how far the map can be trusted.
  5. Check this against a real-world event by using the extents derived from satellite remote sensing, those recorded at high water marks, and the accounts given by residents.
  6. Present the exposure data by overlaying buildings, roads and other important facilities on the GIS output and note the assumptions next to the map.

Final thoughts

The value of a flood map depends largely on the terrain and the structures it is based on. Since elevation errors do not tend to occur in a single direction, it is better to carry out tests rather than attempt to correct them. You should also verify the map using actual events, checking both the depth and the extent of the flooding, and make sure that hazard and risk are kept separate.

When a flood map is provided for a project, my advice is straightforward: find out what elevation data it was based on, whether the culverts were included in the model, and against which flood it was checked. If no one is able to give you these details, then you have gained some insight into the map.

Back to blog

Leave a comment