EGMS vs levelling in North Rhine-Westphalia: a check without GNSS

Question: does EGMS show the same subsidence as precise levelling? Levelling is a height measurement of fixed points on the ground, called benchmarks.

Data:

• EGMS L3 Ortho (vertical motion in cells of 100 × 100 m), release 2019–2023;
• levelling: the register of height benchmarks from Geobasis NRW. We use benchmarks measured twice inside the EGMS time window (mostly in May 2019 and May 2022).

Why it matters: levelling does not use GNSS, but EGMS is calibrated with a GNSS model. So this check does not depend on the calibration.

Method: for each benchmark we take the EGMS cell where it lies. We compare the height change between the same two days, in mm. We use every benchmark with two measurements and do not select them by EGMS. We set the pass/fail limits before the calculations.

1. Where EGMS sees motion

How to read the map: the red areas are subsidence bowls next to open-pit lignite mines (up to tens of mm per year). Blue means the ground goes up. Triangles are GNSS stations. Circles are the benchmarks shown over time below.

2. Does levelling confirm the motion from EGMS?

How to read the figure: each point is one benchmark. The horizontal axis shows the height change from levelling, and the vertical axis the change from EGMS, both in mm between the same two days. The diagonal line means perfect agreement. Orange points are benchmarks in subsidence bowls.

There are two bands around the line (±2σ):

• the narrow, darker band is EGMS noise. It comes from the rmse column in the EGMS files: the noise of one measurement around the EGMS model. A height change is the difference of two readings, so σ = √2 · rmse. This is the uncertainty that EGMS declares.
• the wide, lighter band is the usual difference between EGMS and levelling. We calculate it on benchmarks that do not move according to levelling: the spread of their speed differences, multiplied by the time between the measurements. It includes EGMS noise, levelling error, the difference between one point (the benchmark) and the average of a 100 × 100 m cell, and local EGMS errors.

What this means: only a small part of the benchmarks is inside the narrow band, but most are inside the wide band (see the table above). So the EGMS rmse alone is too optimistic compared with a measurement on the ground: the real difference is a few times larger than the noise the product declares (the ratio above the figure). We added the wide band after we saw the narrow one. Both bands are for information only and have no pass/fail limit.

The points lie along the line from 0 to about −140 mm. The whole cloud is a little lower than the line: everywhere, EGMS shows a little more subsidence than levelling. This is why more benchmarks are inside the bands after we remove this common offset.

3. Three benchmarks over time

How to read the figures: the blue line is EGMS, every 6–12 days. The blue band is EGMS noise (±2σ). The grey band is the usual difference between EGMS and levelling (±2σ, from stable benchmarks). The big yellow dots are the two levelling measurements; the dashed line only connects them, because there is no levelling between them. Both series start at zero on the day of the first measurement. A yellow dot inside the grey band means that EGMS and levelling agree as well as they usually do.

On the first two benchmarks, EGMS and levelling end at almost the same place. The third benchmark does not move, but EGMS shows a few mm of subsidence there. The next section explains why.

4. A stable benchmark: is a difference of a few mm a problem?

Levelling says that the benchmark does not move, but EGMS shows a few mm of subsidence in 3 years. We split this into two parts: a common offset (the same for all stable benchmarks, the dotted line) and uncertainty (the bands).

How to read the figure:

• Panel A (as measured): compare the blue EGMS line with the dotted line, the common speed of all stable benchmarks, on the day of the second levelling measurement (numbers above). There is no levelling before the first and after the second measurement, so those parts of the line do not show anything.
• Panel B (without the common speed): the same line after we remove the common offset. On the day of the second measurement, the difference EGMS − levelling is the value in the last point of the list above. The bands show if it is inside the uncertainty.
• Panel C (all stable benchmarks): how the EGMS changes are spread over all stable benchmarks, with this benchmark (orange line) and the median of the group. Criterion N3, set before the calculations, allows a median |Δh| of up to 5 mm.
• Scale: a few mm compared with −140 mm in the bowls is a few percent of the signal.

An example from everyday life: we measure how tall people are on a ferry. A tape measure that starts at the deck shows that nobody changes. A GPS that starts at sea level shows that everybody goes down by 1 mm per year, because the whole ferry slowly sinks. Both measurements are correct; they just have a different "zero". Levelling starts from the reference benchmarks of its network, and EGMS from the ETRF2000 reference frame, through GNSS.

What we know and what we only think: we measured that the offset is the same for all stable benchmarks and that it is within the limit. The different "zero" of the reference frames is the most likely reason for it, but we did not test this (Even et al. 2024 do not give the reason either).

5. Results for the criteria set before the calculations

How to read the table: ✔ = the criterion passes, ✘ = it fails, — = not calculated.

• The main variant compares the values directly, like Even et al. (2024). It fails N1, N2 and N4 because of a constant EGMS offset, not because of scatter.
• The relative variant (an extra check) first sets a common "zero" on benchmarks that do not move according to levelling. Then all calculated criteria pass. N3 is not calculated here, because these benchmarks are the reference.
• We added the relative variant after we saw the main result, so the main result does not change.

6. Subsidence bowls

How to read the table: in the bowls, EGMS and levelling agree on the direction of motion at every benchmark, and the size of the motion is almost 1:1. The bowls are a few kilometres wide, but the GNSS model used to calibrate EGMS has a grid of about 50 km. So the calibration could not "draw" the bowls: EGMS measured them by itself.

7. The largest differences

How to read the table: in all three cases the direction of motion is the same. Part of the "more subsidence" difference is the general EGMS offset (a few mm in 3 years). The benchmark with a large positive difference is a real outlier. A likely reason (not tested) is that the benchmark is at the edge of a bowl, where the 100 × 100 m cell mixes ground that sinks faster and slower.

Key numbers

Conclusion

• Levelling, a measurement that does not use GNSS, confirms the subsidence that EGMS shows: the bowls are in the same places, the direction of motion is the same and the size is almost the same.
• EGMS measures the motion by itself and does not only repeat the GNSS calibration: the bowls are much smaller than the grid of the calibration model.
• Be careful with absolute values: EGMS has an offset of about 1–2 mm per year. For motion below about 2 mm per year, and for absolute heights, you need a reference point measured on the ground.
• EGMS is an archive product, updated once a year. It does not detect sudden events.

What is not here: a comparison with GNSS in NRW (there is no subsiding station in the area, so the result is unclear), "time to detection" curves (the method needs a fix) and our own Sentinel-1 processing (not run yet). More details: EGMS_Monitoring.ipynb and docs/evidence/F10_NRW_niwelacja_EGMS.md.

Sources: EGMS (Copernicus Land Monitoring Service); levelling: Geobasis NRW; comparison with levelling in NRW: Even, Westerhaus, Kutterer (2024), PFG 92:253–270; limits: Koudogbo et al. (2025) EGMS Validation Report.

The full notebook and all the code are on GitHub: Showcase_2_Levelling_NRW.ipynb in the egms-check repository.