When a client sends over a site with simple request of "elevation data". You process the imagery from the drone, create a surface model, and you just hand it off. Two weeks later they're back on the phone — the cut/fill volumes are out by an order of magnitude and no one can figure why.
Most of the time, here is what went down: a client got a DSM when it needed a DTM. The roofs, the tree canopy, parked equipment on site — they all counted as "ground," and the volume calculation inflated. This is one of the most common and the most preventable issue when delivering drone data, and it boils down to one factor: understanding what surface you are actually passing on.
Let's get into it.
Two Surfaces, One Point Cloud — Defining DSM and DTM
A point cloud is the raw material from which every drone photogrammetry or LiDAR flight is born. How you deal with that point cloud ultimately dictates whether you get a dsm or dtm.
What a DSM Captures
The first thing you see when you look from the top is a DSM — Digital Surface Model That basically means it records all the information found on our planet at that very moment in time. It will show you the roof if there is a roof. If there is a tree and the camera saw clearly, it only shows the head of the tree but not something below on the ground. Everything is revealed, everything filtered — it's just a naked image of the naked surface. Which is one of the reasons why it is super simple to create a DSM directly from a point cloud (which we can consider as something like an enormous data-dense cloud).
What a DTM Strips Away
A Digital Terrain Model (DTM) is just the opposite, bare earth only. Buildings, vegetation, vehicles, part of the temporary site clutter — all left out and only the surface ground remains. This terrain is the "naked" version, and this bare-earth variant is the main product of most topographic survey work — if something relates directly to water movement or slope grading or amounts of material above/below a known reference elevation, it needs this naked (bare) version.
Another term that you may find it being thrown around like it's a synonym is "DEM". In reality, DEM is often a loose term that can refer to a DSM or DTM depending on context and the person using it. And that is why the distinction between these terms can be confusing. Now, allow me to add some clarity; let's take this a step further.
What You Really Need to Get from a Point Cloud to a DTM
How do you go from a raw point cloud that represents everything to a clean bare-earth surface?
The short answer: ground point classification. Software can use filtering algorithms such as Cloth Simulation Filter (CSF) and Progressive TIN densification to separate "ground" points from "non-ground" ones or a Simple Morphological Filter (SMRF) — the approach used by methods like OpenDroneMap— to extract ground surfaces. The concept of CSF is pretty simple: visualize it being some virtual cloth draping overtop of the point cloud you have. It settles down into the very lowest points it can find, and those become your floor. After classifying ground points, you interpolate those to create the DTM raster as continuous data.
Why Photogrammetry Struggles Under Canopy
The only drawback with photogrammetry cameras is that they have a blind spot for capturing ground images: they cannot see through trees. The inability to obtain a clear image of the ground under the dense vegetation is an enormous issue. Instead, they have to assume it with the knowledge of intelligent educated guessing relative to empty spaces in the trees. This means that when using photogrammetry to produce a DTM of a site with many trees, you need to be prudent. The output may not be entirely accurate, and you can sometimes validate the image using other sources like LiDAR to get a true representation of the terrain. Due to its ability to send pulses of light that can penetrate trees, LiDAR is generally much better at achieving accurate data in heavily forested areas. This is particularly crucial where information on the actual form of ground and its attributes are warranted, such as in surveying or mapping. In those instances, you have to use LiDAR if you want the best results. If you want to know more about how DTMs, DSMs and DEMs differ as well as how to choose the proper approach for your assignment check this DTMvsDSMvsDEM post.
Are DSMs Suitable For Cut / Fill Volumes?
For earthwork volume calculation by all means I hope you realize that ideally BUT (for those still in the 'not sure' crowd) it should be a digital terrain model not a digital surface model. This is due to the fact that DSM considers tree/structure heights existing on the site, which can dramatically influence your calculations. As an example, if the surface under investigation takes place with trees or buildings, a DSM based calculation will record those heights in the terrain surface and provide wrong numbers. This has happened in actual constructions, and getting it right is important. In our last post regarding contested cut/fill reports, we promised to delve deeper into issues of accuracy with using a DTM in preparing earthwork volumes — and there are no exceptions! Having a DTM helps ensure the accuracy of your calculations, and the reliability of your data — both critical to any construction project. So to sum it up — always use a DTM for earthwork volumes, never a DSM, especially if you have vegetation or structures on site.
How the wrong model spoils a deliverable
This matters not only in cut/fill, but also => The model you select has to be in line with the question you're answering.
Why Bare Earth is Necessary for Flood and Drainage Modeling
Water flows on the ground, not on tops. Flood/drainage models on DSMs etc, will see buildings that prevent the water flowing where it wouldn't flow normally at ground level. This is a raw-earth issue that you have to consider, so it requires every time a digital terrain model (DTM). Mistakes here are not just a minor matter - if you miscalculate the likelihood that an area will flood or how stable a slope is because you're using the wrong model, it can have severe safety and regulatory consequences. An example of this is that DSM vs. DTM will tend to overestimate flood extent, which can have catastrophic effects on communities in need. Just like the misestimation of slope stability has great safety implications. We should always make sure to use accurate models as we might get wrong results leading up to a disaster.
When You Actually Want A DSM Instead
This is not to say that DSMs are the "wrong" model, they are the right model for a different kind of question. Doing solar potential studies, RF line-of-sight planning, and obstruction analysis requires a DSM since a DTM would remove precisely the obstructions you want to assess (source).
Is DEM the Same as DTM?
This is where a lot of client confusion starts → not necessarily. A DEM is often used as a catch-all term to mean any elevation raster, so where a client asks for a "DEM", it could be either bare-earth surface or full surface model and they may not even know the difference. The safest bet is to always check directly: are they asking for the land or everything above it? This is especially critical once the data gets into GIS for more advanced analysis, because if we have an incorrectly identified elevation layer that propagates errors to every downstream map or model built up it. Having one clarifying conversation earlier on will prevent a whole reprocessing cycle later.
What This Looks Like on a Real Project
We had encountered similar situation this past summer while working on project in Naran, northern Pakistan. It was a topographic and bathymetric survey for a hydropower plant which was done under the Program by the World Bank. Pakhtunkhwa Energy Development Organization was executing the project, while Tractebel-ILF was supervising it. This survey covered a large area of around 80 hectares. We had to make maps, some of them detailed ones like digital terrain models, digital elevation models and contour maps. We needed to establish control points and benchmarks in order to maintain the accuracy of our work. It was a large enough initiative that we had to make some challenging decisions regarding how we deliver the results.
In this work, 403 aerial photographs were taken with the camera hanging approximately 99 m above ground level that was also annotated in the field using a total of 12 reference points. This allowed us to reach a total accuracy of 5.5cm over the complete site. Well, who cares that the photos are accurate when you have a wrong model? Then you need to build a bare-earth terrain model for your hydropower feasibility study channel geometry and grading. Final product to the engineering team that looked it over- so we did our best to nail down the digital terrain model and digital surface model early on. This was a key because the team needed to be able a fact based way number or stat that would help make an informed decision about the project. What was most important however, is by getting the models right, we were able to deliver something that the client could actually use.
Matching the request with both deliverable format and resolution
Once you know whether you're working with a DSM or DTM, you'll need to identify the format and resolution that your client requires - GeoTIFF, contour lines, coordinate reference system etc. Our GeoTIFF, Shapefile and KMZ comparison article goes into greater detail about this decision making process, so once you've got your head wrapped around the surface model itself- check it out. You can find it on our website, that should give you a clue of which file format actually best serves your client.
What Accuracy Do your Earthwork DTMs Need
When it comes to engineering-grade earthwork work, one common goal might be a vertical RMSE near 0.1m (~0.3ft) when referenced against surveyed ground control. It is not a concrete rule that applies to every project (a full feasibility grading study doesn't require the same exactness as a final construction cut/fill) but it is a solid reference point to make sure your deliverable passes the sniff test before going out the door. The density and distribution of your GCPs are just as important as what processing software you use, so plan accordingly and don't leave it to the last minute.
How to pick which model you want for your next project?
Thus the most important question you need to ask yourself before starting anything is: does the client require just the ground or also all what's above it? When they require the earth, you'll probably be digging into earthwork, drainage and topography analysis which means you will need to a Digital Terrain Model or DTM for short. Now, if they want everything that's sitting above grade: buildings, trees and other obstructions then you will be doing obstruction mapping, canopy studies and line-of-sight work, which means a DSM is what you actually need. Get that answer sorted before you start processing anything, and you'll save yourself the redo that comes from guessing wrong.
In our experience, this one clarifying question does more to prevent rework than any amount of processing precision after the fact. The model has to match the question you're trying to answer, or the numbers just won't hold up — no matter how clean your point cloud is.
