Current approaches in UAV Operational Risk Assessment and Practical Considerations

The expansion of Unmanned Aircraft Systems (UAS) is creating new markets, particularly in Urban Air Mobility (UAM), which presents unique challenges. Beyond the risk of mid-air collisions, UAM services in urban areas introduce ground risks to buildings, traffic routes, and pedestrians. This research explores trends in Ground Risk models for different operation planning stages: strategic, pre-tactical, and tactical. It offers a logical pipeline for UAS operators, emphasizing detailed analysis and data source considerations, along with the importance of model interoperability. Using a Naples case study, the research provides practical steps for national authorities to streamline the UAS authorization process.


Introduction
Urban Air Mobility (UAM) is a rapidly growing and evolving transportation paradigm that uses unmanned aircraft systems (UAS) to provide on-demand, low-altitude transportation services in urban and metropolitan areas.UAM aims to reduce traffic congestion, commuting times, and offer more efficient transportation solutions.While drones offer many benefits, there are also significant safety challenges, particularly in terms of mid-air collisions and the risks they pose to people and infrastructure on the ground.These challenges must be addressed as the use of UAS technology continues to expand.The Joint Authorities for Rulemaking on Unmanned Systems (JARUS), supported by National Aviation Authorities (NAAs) and stakeholders, have developed the Specific Operational Risk Assessment (SORA) framework [1] to identify and assess safety risks for UAS operations.SORA is a simple basic tool for operators to use to guarantee safety in the air and on the ground.It is worth noting that even if a UAV is certified, it may still require additional operational approvals for specific activities, such as flying over populated areas or within certain distances from people, buildings, and infrastructure.In addition to SORA, there are various other risk assessment methods described in the literature, which consider the heterogeneous "world of models" underlying each term of the overall risk equation.These models require validation and ground truth data to be accepted by authorities.Operators need to know how to obtain and use data sources in this complex ecosystem.Additionally, each operator needs to understand which model best suits the application scenario under analysis in terms of pertinence and acceptability by authorities.They also need to know which relevant data are available and can be used.This work provides an overview of the current research trends on Ground Risk models, taking into consideration the different stages of operation planning, mission execution, and in-flight operations.It defines a workflow that an operator should follow when implementing detailed analysis, exploring the data sources they should gather, and the required level of interoperability among models and related sets of data.It also provides an instance of such a logic flow on a case study set in Naples, considering two different potential operations: from the airport to the seaport and across two different points of the seaport.From this reasoning, the work derives some key practical steps that national authorities could consider as a "to do list" to facilitate the authorization process for UAS operations.
2 Risk models for UAM UAM operations are not only about new technologies and new business, but also about aircraft safety and risk assessment.This is paramount for enabling such operations.Literature provides a great variety of risk assessment models for UAS operations.The main risk topics are: • Air risk, related to the risk of mid-air collision with other (manned) traffic, • Ground risk, related to the risk to people (and property) on the ground.Air risk has often been considered of primary importance for traditional aviation and in literature, it has also been addressed by several projects on UAS (e.g., SESAR); nevertheless, most UAM operations are expected to be held in atypical airspace or U-Space geographical zone, that offer mitigations for such issues; accordingly, ground risk is the most relevant driver for risk assessment of most uncrewed operations.Several risk models have been defined and used for UAS operations, especially in recent decades, in various contexts (civil-military).The Specific Operations Risk Assessment (SORA) and European Defence Agency (EDA) Risk Assessment Tool (RAT), and most other risk models, are based on the general assumption that the risk posed to overflown third parties is generally composed of three major terms, as expressed in the following equation for the expected number of fatalities: The terms of the equation behind assessment of Ground Risk model are: • (): the probability that the UAS enters a loss of control state per flight hour.This is related to aircraft integrity and reliability, and is the main driver of aircraft safety in traditional aviation; • (): the expected number of people the UA collides with during a loss of control event.This is the most relevant term for uncrewed aircraft, and relates to the number of people exposed to risk in case of a failure.Details will be provided in the following sections; • (|, ): the conditional probability that the UA will cause a fatality to an impacted person, given that the aircraft has failed and has collided with an individual.Mitigations such as extremely light UA or parachute recovery are commonly used to address this term.The equation ( 1) is further developed in SORA as in the following: Equation ( 2) is behind the tables in SORA assessing the intrinsic Ground Risk Class (iGRC).Once the iGRC is assessed, the operator can provide mitigations by setting up actions to reduce people exposure (mitigation type M1) or to reduce lethality in case of collision (mitigation type M2).Such actions modify the iGRC leading to the final GRC.If an operator does not completely comply with SORA iGRC table or intends to apply mitigations to the iGRC, the terms of the equation (2) need to be explored to evaluate the GRC "beyond the tables".Figure 1 clasiffies the models behind each term.It is here evident that people at risk are related to 3 major factors: 1. Population density in the ground risk area (  ): sources of data and acceptable methods for computing population density are paramount.This is the main topic addressed in this paper.2. Exposure factor (  ): this can be related to the effects of buildings or infrastructure that protect people (a sheltering factor to be used when computing people effectively at risk) or to the effects of obstacles that affect the impact location or the critical area (see SORA Annex F); 3. Critical area (  ): this is the main driver for mitigations according to JARUS (SORA Annex F) and EASA (MoC M2).
The capability to estimate these three factors is therefore crucial for risk assessment.An extensive overview of the models is provided in [2].

Population Density Estimation
The considerations reported in the previous section support the fact that the estimation of population density is a key element for the calculation of the GRC.Equally important is the ability to estimate this density at various stages of a UAS mission: strategic (before the flight day), pre-tactical (up to 2 hours before), and tactical (during the flight).In this context planners may be aided by several kinds of estimation models.An important distinction is between "static" and "dynamic" models.Static models estimate density using fixed data related to residents, often enhanced with additional information.They are valuable in the strategic and pre-tactical phases and can be improved with spatio-temporal modeling, considering local factors.Dynamic models focus on estimating density based on human mobility and real-time data, making them more suitable for the tactical phase.
In any case, obtaining flight authorization necessitates accurate analysis and therefore our main focus will be on the strategic phase.Evaluating population density along the planned flight path is critical for defining the operational volume and adapting it to avoid high-risk areas.Operators must use various data sources to assess the risks associated with the flight operation's footprint.In general, data maps intended for ground risk evaluation must meet several crucial requirements to be considered valid by flight authorities.Specifically, they should be generated by combining multiple layers of ancillary data, be of high resolution, and pertain to recent time periods.Furthermore, they should be produced by organizations that offer comprehensive information about their methodology, validation, and accuracy.These requirements aim to ensure that population density maps are reliable, up-to-date, and suitable for making informed flight planning decisions and support operational authorizations.

Data Sources
Several data sources may be consulted to help the assessment of ground risk: • Census.The census database provides access to a wide range of statistical data, including demographic data, population data, housing information and other information collected during censuses.These data are updated regularly but are inferred via a population sample.• Real-estate Registry.Cadastral cartography allows understanding land use and ownership.
Cadastral maps typically show parcel boundaries, land use types, building footprints, road networks, and other features.
• Specific High Resolution Map Products.There are free mission planning map products that meet the abovementioned requirements but differ in grid resolution and epoch.The next section will describe data from the Global Human Settlement Layer (GHSL) of the European Commission.• Data from Mobile Network Operators.Mobile phone historical data may be useful to estimate the population distribution at various times in different areas of cities. Real-time mobile phone data may also be used for the tactical phase in order to take into account temporal variations.• Other Data.Additional sources for estimating population distribution include: data from public authorities regarding local events (including crowd estimates and traffic disruption details), and data on expected traffic disruptions due to road closures, construction, or other factors.

GHSL -Global Human Settlement Layer
The Joint Research Centre (JRC) and the DG for Regional and Urban Policy of the European Commission together with the international partnership GEO Human Planet Initiative carry out the Global Human Settlement Layer (GHSL) project that provides open and free data and tools for assessing the human presence on the planet [3].
The GHSL uses remote sensing, satellite imagery, and various geospatial data sources to create highresolution mapping of human settlements, population distribution and built-up areas, employing advanced image processing, census data, and crowd-sourced geographic information.
Useful spatial raster datasets are: • GHS-POP.This depicts the baseline distribution of population, expressed as the number of people per cell.It is basically based on census data.• GHS-BUILT-S.This depicts the distribution of built-up surfaces.It gives an estimate of the presence of buildings and shelters and allows to evaluate the built fraction in each grid cell.It covers both residential and non-residential areas.It is based on image processing of satellite imagery.• GHS-BUILT-H.This depicts the spatial distribution of the (average) building heights per cell.
It is useful for determining the obstacle effect for critical area.• ENACT-POP.This depicts the seasonal night-time and daytime population grids, expressed as the number of people per cell.Although the current available data are referred to 2011, they may still give indication on the day and night increment/decrement of the population with respect to the baseline given by GHS-POP.These products have different resolutions depending on the reference epoch.This is due to the ancillary data available at the time of interpolation.

META's Data for Good
As part of the Data for Good program, Meta provides free access to high-resolution population density maps with resolution 30 meters and reference epoch year 2020.According to [4] the overall process to obtain these data is similar to the GHSL and therefore, regardless the fact they are not from institutional bodies, they comply with the requirements described above and could be used for drone flight planning.

Crowd Counting
Analyzing urban images to detect crowds provides valuable data for monitoring population density changes in time and space.This data is essential for UAS flight paths planning and also enables realtime adjustments in the tactical phase.Crowd counting, which estimates the number of people in images or videos, is a key aspect, and there are multiple techniques available for this task.There are five main crowd counting methods [5]: counting by detection, by regression, by density estimation, by clustering, and by CNN, the latter being the most accurate but computationally expensive.All of these methods rely on machine learning techniques.In this context, the concepts of acceptability, trustworthiness, and explainability are of paramount importance and will be paramount in future research.Acceptability involves issues of transparency and consent regarding both input and output data.Trustworthiness relates to data quality and the system's ability to deliver reliable performance in EASN-2023 Journal of Physics: Conference Series 2716 (2024) 012055 real-world applications.Explainability is concerned with the interpretability of crowd counting methods, especially when employing complex AI models as per CNN, which is essential for user understanding and regulatory compliance.

Sheltering
The sheltering factor is a measure of the protection from UAS falls provided by natural or artificial elements in an area, such as buildings, roofs, and trees.These elements can be used to reduce people at exposed to risk and as a way to absorb kinetic energy and shield people from debris during a crash.Impact energy can vary significantly, from minimal shelter to strong shelter providing high levels of protection for people.
To estimate the probability of fatality given exposure (|) a model [6] is used based on kinetic energy at impact (Eimp), aircraft mass, and sheltering, represented by three parameters (α, β, and ps).
is the sheltering parameter.It determines how exposed is the population to an impact.The α parameter is the impact energy required for a fatality probability of 50% with   = 0.5 and the β parameter is the impact energy threshold required to cause a fatality as   goes to 0. Consequently, sheltering implies that the operator is aware of the types of obstacles the vehicle may encounter in the event of a failure.

GRC Evaluation Pipeline
This section presents a workflow for assessing unmitigated and mitigated GRC in the strategic phase.Similar considerations can be made for the pre-tactical and tactical phases.The design of this pipeline adheres to specific requirements, including leveraging referenced data, using understandable models, maintaining a simple and verifiable process, ensuring flexibility with data and models, and adopting a conservative approach.

Workflow
The strategic phase pipeline consists of several steps.
• Step 1: Analysis of the mission area.In this initial step, a comprehensive analysis of the mission area is conducted.This involves a detailed population density assessment using high-resolution and up-to-date data from various sources.The dispersion area where the UAS might potentially fall is calculated and a ground risk area is established along the path.Aggregation centers within this buffer, such as parks and stadiums, are evaluated, and the population density is recalculated, accounting for spatial and temporal variations.The critical area is assessed based on existing literature models.Ultimately, an estimate of the "raw" Ground Risk Class according to SORA guidelines can be derived.• Step 2: Sheltering.Using data products and models [7] it is possible to assess the exposure factor that may intrinsically reduce the ground risk, since it decreases the "exposed" population.This is done considering the following equation: where %built is the built fraction of the considered area.However, depending on the type of structures, the built fraction could be reduced as it may not resist the impact.Various literature models are available to assist in making these assessments, provided that the building characteristics are known.• Step 3: Temporal Model.Some models [8] allow to make considerations regarding population density variations linked to specific studied behavioral patterns.These patterns help estimate the time people spend within various locations (home, office, vehicles, outdoors, etc.).This, in turn, allows to modulate the local population density value with respect to the baseline density.
• Step 4: Considerations on P(fatality).The evaluation of sheltering in Step 2 allows also to estimate the probability of fatality given the exposure using the model described in the previous section.It is paramount to consider the kinetic energy of the UA impact since it is an input to the model.Once the probability is determined, the GRC can be recalculated.• Step 5: Considerations on parachute mitigations.Utilizing a parachute to reduce impact kinetic energy can potentially mitigate risk.Nevertheless, this approach necessitates a thorough assessment of dispersion areas, coupled with a reassessment of the critical area.Once these steps are completed, a comprehensive understanding of ground risk is attainable, leading to a more precise calculation of the Ground Risk Class (GRC) following the SORA approach.

Workflow example: use case in Naples area
In this section we present an example of the application of the GRC evaluation pipeline, applied to the planning of a UAS mission in the suburban/urban area of Naples, Italy.The considered drone is a fixedwing cargo aircraft with MOTM 500 kg, wingspan 11 m and length 6 m.The mission planners foresee two possible business trajectories: (A) between the airport and the port flown at altitude 500 m, (B) across the harbour at 130 m.The UA has a parachute system.According to the pipeline, the first step is to analyse the mission area.This analysis involves two key elements: population density data and the determination of dispersion areas.The data sources utilized are GHSL products.However, given the presence of the parachute system, an understanding of the dispersion areas is needed before processing population data.To this end, using appropriate simulation models and data coming from the COSMO [9] numerical weather prediction model we can use historical wind data from various hours and seasons to assess the maximum and the average parachute dispersion areas along the intended paths.
Figure 2 shows the areas of interest for paths A and B together with the municipality area of Naples.Table 1 reports the numerical values for such areas.Using GHSL data and GIS software, one can estimate the population density in the areas potentially affected by drone impacts.Additionally, GHSL offers map products that enable the assessment of population variations during the day and night in these areas.For instance, in the municipality area of Naples there is a population increase of about 14% during the day and a decrease of about 0.4% during the night with respect to the baseline data.
In order to calculate the "raw" GRC it is also necessary to determine the critical area for the drone.This can be done using models and empirical formulas reported in [9].This calculation must be repeated when considering the drone under the parachute.The same considerations holds for the of the impact kinetic energy.Average speed at impact for fixed-winged is around 25 m/s while usually the final speed when the drone is under a parachute is around 6 m/s.The built fraction from GHSL allows to determine the exposure factor in the areas.For simplicity, in this example, we consider that the buildings offer 100% sheltering.In addition, the impact kinetic energy allows the evaluation of P(fatality).Finally, the use of the temporal model [8] also allows to make an estimation of the actual population likely to be outside at the time of flight.Table 2 reports the acquired data for the four areas considered.Given these data, an example of the calculation of the ground risk class at strategic level for a fixed wing drone mission is shown in Table 3. Table 3 demonstrates how the application of models to existing hi-res data facilitates informed mission considerations based on local characteristics and the use of mitigation techniques.This enables the reduction of the ground risk class according to SORA, thereby fostering safer mission planning.

Open Issues and Future Actions
Working groups in JARUS, EASA, EUROCAE and others are paving the way to new operations and UAM Conops, but UAS operators are still loaded with the excessive burden of proving all evidences within the risk analysis (SORA).In this work new methodologies are proposed to improve and make risk analyses more effective.They are mainly based on state of the art in literature and research programs, but are not yet common practice for UAS operators nor acceptable means for NAA.An example of workflow to be followed by an operator is sketched out to share the uncertainties that it has to face and provide the authorities with a risk assessment process as simple as possible.Regarding the identified steps in the Pipeline, the following open questions arise: EASN-2023 Journal of Physics: Conference Series 2716 (2024) 012055 • Provided that an empirical model is chosen by the operator, it is not clear if it is applicable to all types of vehicles and what specific evidence is required by the Civil Aviation Authority.• Specific requirements for wind condition analysis, including frequency, granularity, and accuracy standards for the associated wind dataset, need to be established.• Strategies for incorporating the influence of geographical location and other factors, such as seasonal fluctuations, daily variations, and so on, require further elaboration.• Specific guidelines on how to correctly consider the envelope of the dispersion areas should be designed.
• The impact of mobile data statistics on the precision of population density estimates and the acceptability of these data for strategic mission planning need to be thoroughly investigated.• The acceptability of real-time data for non-strategic mitigation considerations and is still unknown.From the operator perspective, these elements represent a barrier to an easy and fast authorization process.In order to reduce such a barrier, further steps should be done: a subset of models/approaches could be appointed, cities should be characterized with respect to such needs and the usage of new type of data (e.g., mobile data) should enable a more tactical situation management.

Figure 1 .
Figure 1.Factors contributing to the ground risk model

Figure 2 .
Figure 2. Naples Municipality, the planned paths A and B and their dispersion areas.

Table 1 .
Average and maximum dispersion areas for the two scenarios

Table 2 .
Data derived from models and data sources.

Table 3 .
GRC calculated making different assumptions for a drone with MTOM = 500kg