When drones take to the skies – perhaps to deliver lifesaving blood, or to create digital maps of the ground beneath them - the operator will often have completed a Specific Operational Risk Assessment (SORA) before receiving approval to fly. It’s a detailed methodology used to classify risks, identify mitigations, and demonstrate compliance. We’ve come a long way since the first SORA was developed, and a new version 2.5 is now awaiting implementation by the European Aviation Safety Agency. It’s a huge improvement on the previous edition, but we think there is one more fix to be addressed. Egis’ Jan Cernan and Eric Denele, alongside FlyingBasket’s Thomas Markert* explain how and why in this new Insight. So, grab a coffee, and pay attention!
Back in 2022 Jan Cernan wrote about some of the practical problems and inconsistencies arising from assessing operational risk using the SORA 2.0 guidelines published by JARUS. The main point was to do with Step 2, the Intrinsic Ground Risk Class (iGRC), used to evaluate the potential impact of a drone operation on people and property. Part of that step involves assessing population density and it’s pleasing to see those problems addressed in the new version 2.5, which adds a quantitative approach to calculating population density. But it seems there is still more work to be done, because even using the new methodology could result in an artificially inflated risk level (Specific Assurance and Integrity Level (SAIL) score), which would make it difficult to obtain mission approvals. In this article, we explain some of the issues first highlighted by Thomas Markert, and then go on to present a proposed solution.
Population density along the flight path
Several of us at Egis have been supporting exploratory projects looking at how Copernicus data can contribute to the SORA assessment. In 2022, we assessed the possibility of using Urban Atlas data, which provides land use and land cover information, and can be used as a proxy to identify human presence. Last year, we looked at using GHSL (Global Human Settlement Layer) data, mainly the population density layer (GHS-POP), which is freely available and with consistent worldwide parameters. These efforts are feeding into Eurocae standardisation activities as part of the SORA working group (WG-105 SG6 SORA).
Experience shows us that challenges exist when it comes to working with population (raster) data, especially when planning missions with flight paths over so-called “heterogeneous iGRC footprints” with mixed population density. Historically, drone operators tended to average the population density, building on “average probability per flight hour” for their safety case. The JARUS guidelines on SORA (Annex F) explain that this approach is more appropriate for manned aviation. It considers the highest risk to passengers onboard who are continually exposed to the risk, rather than the risk to individuals on the ground. So, what is the right approach?
Figure 1: Potential dispersion based on altitude (SORA 2.5)
In a recent study, we explored how risk exposure time could be factored into calculating the iGRC score. We designed circumferences with a radius of 191 meters (see Figure 1), with centres aligned along the flight path. These circumferences represented the relationship between operating altitude and the area of potential dispersion (see Figure 2 below). We then analysed the population density within these circumferences in residential areas to determine where it peaks and for how long.
Figure 2: Flight path with intrinsic Ground Risk CLASS footprint overlayed with population density at 200 x 200m resolution
Figure 3: Dynamics of population density along the flight path
This kind of dynamic approach could offer a better understanding of when and where people are at risk along the flight path, allowing us to reshape flight routes toward areas with lower population density. Currently, we are integrating additional parameters and data to model not only residential profiles but also commercial and industrial areas. So, now to the specific problem with SORA 2.5.
Population over-estimation with SORA 2.5
When using population density maps, drone operators must pay special attention to raster resolutions. FlyingBasket is a European drone operator, and they identified an issue that could artificially inflate the population density for rasters with higher resolution, and therefore unnecessarily increase the SAIL score.
Once the drone operator has defined the operational area and identified the highest population density within the iGRC footprint, the intrinsic ground risk is determined using the table shown in Figure 4.
Figure 4: Intrinsic Ground Risk Class (iGRC) determination; source: JARUS guidelines on Specific Operations Risk Assessment v2.5
The table assumes the population density is measured in people per square kilometre. However, the population density maps recommended for operations below 150m (most operations today) have a grid size of 200 x 200m, see Figure 5 below. The advantage of the smaller grid size is the higher accuracy of the population distribution. Since many drone operations are conducted over a range of a few hundred meters or along corridors of a few dozen meters a grid size of 1,000 x 1,000m (or one square kilometre) would be inappropriate for measuring the ground risk, as it would account for population that it is up to 1,000m away from the iGRC footprint.
Figure 5: Suggested grid size for population density maps; source: JARUS guidelines on Specific Operations Risk Assessment v2.5
However, changing the grid size of population density maps also changes the data of the maps. Population density is the ratio of population per area, and therefore it depends on the grid size used to represent the area where the population is counted. For example, calculating the population density for 1,000 x 1,000m square containing one house with four people will result in a population density of 4 ppl/km2. However, when the grid size changes to 200 x 200m, the population density of the grid cell containing the house increases to 4/(0.2km x 0.2km) = 100ppl/km2 . This is problematic when it comes to the new guidelines, because SORA 2.5 uses not the average population density but the maximum population density within an iGRC footprint to determine the ground risk class.
The example shown in Figure 6 illustrates this problem using a population density map provided by Copernicus and resampled on a 200 x 200m grid. The maximum population density in the iGRC footprint is in the lower left yellow grid cell with a single house and five people, resulting in 125 ppl/km2 (or the 4th row with <500 ppl/km2 in Figure 4).
Figure 6: Example iGRC footprint in red with a population density map with 200x200m grid overlayed
Contrast that with the qualitative description of population density provided by SORA 2.5 in Figure 7 the where the 3rd row “Lightly populated” (<50 ppl/km2) would match the area in the example.
Figure 7: Qualitative descriptors for population density estimation; source: JARUS guidelines on Specific Operations Risk Assessment v2.5
Using Figure 4, the intrinsic ground risk class (iGRC) is chosen based on the maximum population density. The iGRC is an input to the SAIL score, and a higher iGRC results in a higher SAIL score. In the example above, the SAIL would be IV for a Beyond Visual Line Of Sight operation of a drone in the 3m class and Air Risk Class “b” without any additional mitigations. It would doubtless be harder to get mission approvals with that score. Whereas with SORA 2.0, the SAIL for the same operation would be SAIL III, and mission approvals would be much easier to obtain.
How can this be fixed?
The issue is caused by reducing the grid size. With a typically uneven population distribution a reduction of grid size always yields higher maximum population density. Section 3.9 in Annex F of SORA 2.5 acknowledges the impact of resolution of population density maps in iGRC, but it does not offer a solution that could be directly implemented by the operator. Similarly, the mitigation of time of exposure discussed in section 3.3 to address heterogenous population densities in iGRC footprints does not solve the problem as it could imply limitations on recurring flights within the same SORA approval. Although the text in the Annex provides helpful guidance for assessing drone flights in areas with different population densities, it does not address the issue of dealing with population density maps, which are very often raster maps, rather than specific areas of population density.
Sticking with the approach of SORA 2.5 a potential resolution could be to add an optional correction step after determining the grid cell with the maximum population density. The idea is to assess the average adjacent population density of all grid cells that fall into the highest population density class of the iGRC footprint. The average adjacent population density of these cells is calculated as the average of the cell and its eight adjacent cells. If the adjacent average population density is at least one class lower the iGRC can be determined at one class lower than the maximum population density. The correction step essentially mitigates the rise in population density that occurs solely due to reducing the grid size in areas with uneven population distribution.
This may sound complicated, but it isn’t. With the help of QGIS and the function ‘r.neighbors’ the population density map can be recalculated, where every grid cell is based on the average of the cell and its neighbouring grid cells. Using this smoothed map, we can verify if the cells with maximum population density class have a one class lower average adjacent population density. Returning to our earlier example, we now see in Figure 8 and Figure 9 that the proposed correction step yields the expected result of classifying the iGRC footprint as “Lightly populated” with <50 ppl/km2.
Figure 8: Example iGRC footprint with 200x200m population density map overlayed. Maximum population density is <500 ppl/km2 (yellow)
Figure 9: example iGRC footprint with 200x200m averaged adjacent population density map overlayed. The maximum average adjacent population density is <50 ppl/km2 (green)
Important! The smoothing step should be optional, depending on the actual planned trajectory, because there will be instances when smoothing artificially and negatively impacts the population density value (and therefore the iGRC score), as shown in the example below (Figure 10 and Figure 11).
Figure 10: 2nd example iGRC footprint with 200x200m population density map overlayed. Maximum population density is in the class <500 ppl/km2 (yellow)
Figure 11: 2nd example iGRC footprint with 200x200m averaged adjacent population density map overlayed. The maximum average adjacent population density is in the class <5000 ppl/km2 (orange)
Time to act
There’s no doubt that, overall, SORA 2.5 is a significant improvement on SORA 2.0, enhancing the precision and consistency of risk evaluations. It provides much better guidance for drone operators who must carry out appropriate and consistent risk assessments. If the nuances we have described here can be addressed by the European Aviation Safety Agency’s Acceptable Means of Compliance (AMC), drone operators will also be able to use SORA 2.5 in areas of scattered and low-density population, without facing an increase of ground risk and consequently SAIL compared to SORA 2.0.
If you have come across this issue, do get in touch! There may still be time to adjust the AMC to better cover important use cases.
-------------------------
*At the time of writing, Thomas Markert was working for FlyingBasket. He joined AirHub Consulting in April 2025.
Drone image credit: FlyingBasket
