Comparing RTK Drone Elevation Maps to ATV/SXS Collected Topo



At Green Aero Tech we often get our data compared to other methods of collection. Be it public LiDAR, manned photogrammetry, manned survey methods, or driving collection methods like Tractor/Truck/Quad.  All of these have their place, and done right, they can ALL be accurate.  Our goal has been to give everyone accurate data at an extremely low price. Due to operational efficiencies over the years, Green Aero Tech has been able to offer accurate elevation so economically priced, it can be hard to believe it’s accurate.  Our goal is to dispel those myths. 

This specific comparison is between our UAV (drone) based data, using commercial grade RTK equipped UAVs with Green Aero Tech’s best practices for both data collection and processing, and comparing it to data collected using the contractor’s Quad on the ground.

So let’s dive in!

The process involved bringing in contractors shapefile as collected from driving the field, creating a surface grid from the points, and then comparing it to the drone digital surface model (DSM).  Because the shapefile doesn’t have visual imagery, we couldn’t really synchronize it by X/Y, however we were able to offset our elevation to match very closely to the contractors generated surface.


Topographic comparison


When compared, this next image shows the differences between the two layers.  Green = ±5cm or less difference (good), and Red = greater than +5cm difference (ground layer higher elevation than drone layer), Blue =  greater than -5cm difference (ground layer lower than drone layer).  It’s obvious that the ground layer would indicate blue (lower) elevation at the house yards, since there’s no data and it’s interpolated, whereas the drone shows actual surface height much higher.  Since the difference layer is primarily green, this indicates that the two layers differed by ±5cm or less in the majority of areas, which is very good and expected.


Looking closer at the ground points (dots) overlaid on the difference layer shows that the spacing in between the two swath lines of ground data is where the differences tend to arise, so it’s the interpolation effect that is creating most of the red/blue, because where the points actually are the drone data agrees with the measurement.

By looking at a profile along the ground topo lines, we can see the drone data lines up very well (green in profile chart) to the interpolated ground line (red in profile chart).

To get a better idea of exactly how accurate we are to the same points as the ground collection, we took the elevation data from the drone and sampled it at the same point locations as where the ground topo has data.  Then in Excel we compared these two values (the original elevation as indicated from SVT vs drone-stated elevation at the same coordinate) and came up with some statistics.  Across the 11,000 ground points checked, the average was ±2.5cm difference between the two, which is better than our normal stated accuracy of ±5.0cm Z.  The median value was actually ±2.1cm Z, with only a few % being outside our expected range, so very positive results.  The attached XLS has all the data used in this comparison.

Profiling between ground lines is where we see some divergence between the output layers, particularly where there are small features in the topography.  These differences aren’t likely to be too significant in most cases.

But if we look cross-profile perpendicular to the ground lines, we can see where the interpolation (red line) of wide ground swaths (black dots) misses smaller features and shows a significant difference in elevation compared to the drone layer (green).


Drainage model comparison


For the most part, this may be fairly inconsequential to the overall drainage report unless looking purely at the surface drainage.  But an interesting thing we noted on this project is that the (apparently) identified outlet on the west side of the field (red arrow) appears to have a culvert that doesn’t actually drain out the right direction, or is otherwise too high in elevation to drain off the field on that west side.  In the screenshot below, the left side shows the Green Aero Portal drainage report based on the ground surface file (created from the ground points), as you can see there are 7-8 relatively small “sink” areas highlighted, because it’s assumed (from the extent of data available) that water will drain out the west side easily.  However, when looking at the drone layer, which has a measurement on the other side of the road (and other end of the culvert), it definitely appears that work will be needed on that ditch to get flow moving in the right direction otherwise this will be a bottleneck and “sink” water in the southwest corner.  Of course, since this is a surface model from the drone, it doesn’t take into account hidden culverts or other terrain features that would allow water to drain past these areas.  Or if there’s a waterway going west through the corn field, this may be sufficient to drain without using the ditch going north/south.


By modelling in the culvert to a depth that would allow full flow to the west of the field, re-running the drainage report (Culvert Full Flow map on Portal) shows a much more similar set of flow paths and sink areas, but with our expanded data we see additional areas that may still be of concern (northwest corner, south center) because the topography past those areas (not captured by the ground data) indicates elevation may be too high to allow free flow off the field.  Upon closer inspection of the visual imagery, it’s likely showing this way due to the standing corn south of the field, so wouldn’t likely be a problem.


Of course, this is all thinking about surface drainage in general, and may not be overly relevant for tile installation, but as we all know, getting the surface drainage under control in as many areas as possible will save hugely on tile installation costs.


Here’s an additional note of interest from the comparison data.  Where the land was evidently too wet for the ground equipment to drive/survey across safely, the interpolation of the surface diverges almost 6 inches (17.8cm) from what the drone was able to accurately capture from the air.  Green = interpolated surface from ground capture, Blue dashed line = drone elevation surface, Black dots are the capture points/lines from the ground equipment.  Depending on the area and if this is being used for an outlet, it could make a significant difference in a tile main line depth requirement.




The conclusion is that our drone aerial capture is very accurate in agreement with the ground-collected points.  The differences start to arise when comparing between lines of ground capture, as interpolation must be done and this shows a smoothing effect that can hide small topographic features.  Since we capture so many points per square meter of ground covered, we don’t see these interpolation errors in the same magnitude as with the ground capture.  We also have the added benefit of visual imagery spatially and temporally relevant to every point of elevation, so if an anomaly is noted in the elevation data, it can be looked at visually to help determine the cause.  When looking at the full drainage model of a field, the extended coverage of the drone beyond just the cultivated acres can help to alert planners to issues with outlets, particularly in surface drainage operations.