4. Measuring and Improving the Resilience of Outdoor Open Areas Against Terrorist Acts: A Behavioural Design Approach

As introduced in previous chapters, the risk assessment and mitigation of the terrorist threat in an outdoor Open Area (OA) are parts of a complex matter, since they are based on the joint analysis of features of the OA itself, perpetrator behaviours, and user behaviours and response to emergency conditions [1‐4]. Moreover, supporting local authorities and their technicians to manage such issues is affected by the level of detail on available information and data, as well as of knowledge on the matter by safety designers. For this reason, methods should both pursue a qualitative and rapid standpoint, but also a quantitative and simulation-based approach, to ensure a complete understanding of possible risk scenarios and effects on the hosted crowd [5]. This chapter hence shows different methods for the creation of risk scenarios, assessment and mitigation effectiveness analysis, correlated to the elements that affect the phenomenology of terrorist acts.

In particular, a risk assessment method to provide possible attack points is defined depending on the effective features of the analysed OA (Sect. 4.2). Such method also supports the development of scenario creation concerning the desirability of the perpetrators in respect to the different specific areas and intended uses composing the OAs.

Then, in view of the dynamics of such a public open space [6], the time-dependent assessment of variations in the OAs use is discussed to evaluate users’ exposure and vulnerability depending on the day and hour of the day in which a terrorist event can be performed (Sect. 4.3). In fact, OAs are typical soft-target for terrorist acts [7], and the probability, characteristics and modus operandi of a terrorist act, as well as its effects on the hosted users, strictly depend on the use of OAs spaces over time in view of dynamics at both the macro (urban) and micro (single OA) scales [6, 8‐10]. Specific user-oriented key performance indicators (KPIs) are defined to this end.

Simulation-based approaches can exploit the results of such analysis to have a deep view of the emergency process, including the representation of evacuation behaviours and of the effects of the attack on the crowd, according to the model proposed in Chap. 3, Sect. 3.​4. In particular, behavioural and simulation-based KPIs are herein defined and rules for the comparisons of scenarios (both in pre-retrofit conditions and in pre- versus post-retrofit conditions) are discussed (Sect. 4.4).

Finally, a rapid summary of Risk-Mitigation and Reduction Strategies (RMRSs) are also offered to support the decision-makers and designers’ actions against terrorist acts in view of the user and OA-related features (Sect. 4.5).

The parametrization of the phenomenon characterizing terrorist threat in the OAs and the identification of boundary conditions can support the risk interpretation and resolution with smart approaches [11‐13].

Such approaches require to be supported by qualitative and quantitative details related to the parameters involved [14, 15]. The parametrization process is already discussed in Sect. 2.​3 and highlighted the major relationships among risk determinants (Hazard H, Vulnerability V, Exposure E) and prevalent features that describe OAs and the phenomenon. However, starting from previous study [16] the same parameters can be detailed towards the identification of logical and mathematic rules functional in solving the risk assessment formulation.

Specifically, as already discussed in [16], a system of indexes and parameters is setup in order to identify the global risk of real OAs. The risk formulation is based on three main assumptions at the basis of the structure in Fig. 4.1:

A diagram of a pyramid depicts 3 risks. The exposure risk includes the crowding, reaction, and attack indexes. The vulnerability risk includes the obstacle, accessibility, and shape indexes. The hazard risk includes the index of use and prevention and target indexes.

Given that rationale, Fig. 4.1 summarizes the qualitative and quantitative data structures of indexes and parameters involved, highlighting the major references.

As far as the significance of parameters, the main elements, properties, and details of values and ranges of the K parameters shown in Table 4.1 can be discussed as follows, in correlation with the related indexes shown in Fig. 4.1).

Hazard Indexes and K-Parameters

Vulnerability Indexes and K-Parameters

Exposure Indexes and K-Parameters

All the presented K-parameters are valued following the rules of the participatory Delphi technique [18], in order to ensure the acceptability of relations among K-parameters and i_n indexes, as well as the formulation and ranging appropriateness of K-parameters.²

Finally, in order to solve the weighting of each index in the calculation of single determinants (Eqs. 4.2, 4.3, 4.4), an analytic hierarchy process (AHP) application has been processed³ highlighting the higher relevance of three main indexes: Target, accessibility, and crowding indexes in each risk determinant (Fig. 4.2).

A table depicts the weights of indexes. It lists the target index, index of use, prevention index, shape index, accessibility index, obstacle index, index of attack type, crowding index, and index of reaction. The accessibility and crowding indexes have the highest weight of 0.65.

Even if the final aim of the formulation for the terrorist risk assessment presented in [16] is structured to support operative and comparative evaluation of real case study, the same can be declined to determine possible attack points. Considering the structure of the formulation that provides a qualification of determinants for SoRs and street/square, and the geometric rules identified for the identification of SoRs in the OAs, the formulation can be focused on the single case study, providing a plan distribution of SoRs and risk properties. In the details, the structured formulation has determined a set of reduced bi-dimensional matrices which allow a brief discussion of the risk of OAs and their parts, where to variable condition of hazard proneness, levels of damages are determined for the setup of level of risks. Table 4.2 shows the details of such matrices, determined in [16], which became the way to qualify the OAs in all their parts (open space and SoRs), as specific target types.

As for other kinds of emergencies affecting the urban built environment [19‐21], the scenario creation in case of terrorist acts in OAs should consider the organization of data not only about OAs physical vulnerability/morphology and terrorist hazard (see Sect. 4.2), but also the user exposure and vulnerability. Recent works within the BE S²ECURe project⁴ developed a joint approach for scenario creation based on the assessment of spatiotemporal variations of user-related factors depending on the OAs characterization [6], so as to mainly derive inputs for emergency and evacuation simulation [22]. To pursue replicability and quick application, the proposed methodology essentially relies on:

Remote analysis via: (a) web mapping platforms such as Google Maps/Street view or Open Street Maps, to derive dimensions, typologies, intended uses, scheduling of areas in the OAs and the facing buildings, and to detect the presence of specific elements composing the OAs layout (including obstacles, street furniture); (b) national census databases, to determine the typologies of users depending on their age.

Standard occupant loads, such as those of fire safety codes, to determine a quick index of users’ exposure by density [persons/m²] depending on the intended use of the OAs and the facing buildings.

Nevertheless, the integration of specific GIS-based datasets and census data from local authorities can increase the accuracy of quick results. The whole methodology is shown in Fig. 4.3 and described above.

A workflow chart depicts three phases. It starts with phase 1, O A morphology and use, followed by phase 2, individual vulnerability, and phase 3, users' exposure over space and time. In phase 3, a line graph with fluctuating trends of density or number of users is observed.

The first phase concerns the identification of intended uses placed outdoors and indoors and that can generate overcrowding in the OA, by also detecting the related surface, and of the user-related factors such as the main use behaviours, the quick occupant loads and the related temporalities.

Concerning the intended uses, the approach excludes residential areas since they essentially represent a sort of background level in users’ exposure and vulnerability and have a limited impact on the terrorist act attraction due to negligible symbolic and strategical values [7, 23]. The approach hence considers outdoor and indoor areas characterized by users’ gatherings (e.g., cinemas, sights, parks), public buildings, special buildings with symbolic value (e.g., worship places, museums) and hard targets. Indeed, areas not accessible to users such as fenced areas are excluded. Then, the available gross surface GS_i [m²] is calculated for each selected indoor and outdoor area. Freeware web mapping tools can be used to measure the plan (or covered) gross surface of each area, e.g., by Calcmaps.⁵ Then, for buildings, this surface is multiplied by the number of floors hosting the given intended use, e.g., by Google Street Maps.

Concerning use behaviours, different typologies of users are associated to the way they spend time in the intended uses, by mainly distinguishing behaviours between [8]: (a) only outdoor users (OO), who generally walk and move in the outdoors with a limited permanence times in the OA due to physical, social, and leisure activities, including sightseeing; (b) prevalent outdoor users (PO), who spend a long time walking in the outdoors or staying/sitting during social and leisure activities; (c) non-residents users (NR), who essentially populate buildings facing the OA and having a direct access to it, and could also contribute to the pedestrian volumes (moving towards or from the buildings) and gather in front of them while waiting to enter them. Access doors, gates and passages can be remotely identified by Google Street View.

Concerning quick occupant loads OL_i [persons/m²] data from Italian fire safety codes [24] are herein adopted and combined with previous works assumptions to extend their applicability to both indoor and outdoor area [25, 26]. OL_i can be then arranged depending on specific data based on surveys. For example, OL_i can be substituted by using the number of seats instead for restaurants, cinemas and theatres, by the number of students and teachers for educational buildings, and by the number of workers for office buildings close to the public. Furthermore, OL_i can vary during the daytime depending on the scheduled activities of the intended uses. Then, temporalities for the considered areas can be derived according to timetables accessed via web search (e.g., opening times via Google Maps or websites of the specific activity open to the public), derived from national⁶/local regulations, or reasonable estimated by local habits [6, 25]. The occupant loads are applied to the timetable range while 0.00 persons/m² are considered out of the addressed timetable. The methodology also distinguishes between working days (as the most common and recurring conditions over the year) and holidays (Sundays and other national Holidays), because they can imply specific occupation variations both depending on timetable and use conditions. In addition, seasonal variations can be taken into account by revising OL_i, e.g., by increasing values for tourist destinations.

In view of the above, Table 4.3 resumes the selected intended uses categories associating typology of users and quick occupant loads, which are also determined in terms of related temporalities.

The second phase concerns the collection of data on individual vulnerabilities, over space and time. Only outdoor users (OO), prevalent outdoor users, non-residents (NR) are users’ typologies related to their position in the OAs, while additional users’ typologies are related to their individual vulnerability due to age and gender. In fact, these factors can imply significant variations in the response to the emergency conditions in terms of pre-movement and evacuation behaviours, motion speed and susceptibility to direct/indirect damages from the attackers and the crowd phenomena [22, 27‐29]. Age and gender data can be easily collected from local, regional and/or national census and statistics databases[30].⁷ Municipalities-related distributions of population by age and gender can be considered valid for the urban areas, and thus for the OAs too, although refined on-site surveys can be then carried out at the microscale. According to a quick assessment approach, five age classes are assumed according to Chap. 3, Sect. 3.​3, to represent motion issues and assistance needs in evacuation [27]. These classes are: toddlers T (0–4 years), parents-assisted children PA (5–14 years), young autonomous YA users (15–19 years), adult users AU (20–69 years), and Elderly users EU (70 + years). For each age class, the users’ percentage by age class UP_age [%] is directly derived from population statistics databases. According to a quick assessment approach, UP_age is homogeneously considered for all the intended uses, except for educational buildings, where the age classes are referred to the typology of hosted students.

Then, temporalities of presence for users by age classes are associated to a presence coefficient at the given time t cp_t [-], which should be multiplied to UP_age and which varies from 0 (no user of the given age class for the considered intended use is present, thus UP_age = 0%) to 1 (the number of users for the given age class is maximum, thus equal to UP_age). Such temporalities and thus cp_t can be assessed hourly. For OO, cp_t = 0 during nighttime, according to Table 4.3 insights, while for PO and NR, cp_t = 1 only during the opening times. Such data can be also refined by additionally considering the percentages [%] of male MU and female FU users, still according to the same databases. Moreover, additional user vulnerability factors can also relate to individual features regarding motion disabilities, since they can affect the motion and sensory abilities of the users [31]. Such kind of elements could be added in the proposed approach, by adding a specific presence coefficient and users’ percentage by motion and sensory ability to Eq. 4.7. Nevertheless, related input data could be not available from a consolidated statistical perspective (i.e., using the same databases defined above), their collection could be time consuming, and data should be carefully collected and managed according to data protection authorities regulation since they could related to sensitive data, being related to individual health information.⁸ Therefore, they are not considered in the proposed method since they are not easily managed according to quick and remote survey approaches.

The last phase concerns the time-dependent organization of data about users’ exposure and individual vulnerability. From a wider perspective and considering a certain time t of the day (hour), the total number of users in the OA NU_t [persons] is calculated (Eq. 4.6) by summing the overall number of users by age-classes for each intended use i in the OA NU_age,i,t [persons] (Eq. 4.7). NU_t dynamically varies from minimum (no user is present since intended uses are close to the public) to maximum (full opening of intended uses) conditions and describes the users’ exposure without specifying individual vulnerability issues. In this sense, the total number of users by age in the OA NU_age,t [persons] (Eq. 4.8) can describe the dynamics in individual vulnerability since it aggregates intended-based occupancy data by age classes.

Finally, the aggregation of NU_age,i,t by intended uses having the same users’ behaviours (OO, PO and NR) can be performed too, thus depicting vulnerability-related factors depending on the users’ position and habits at the starting of the attack. In particular, in case the attack is performed in outdoors, as in considered in this work, the effective number of exposed users NU_t,exp [persons] is equal to the number of users performing OO and PO behaviours, plus those preforming NR who are waiting to enter buildings.

In view of the above, Table 4.4 summarizes the KPIs on time-dependent users’ exposure and vulnerability. All the KPIs can be used to compare different risk scenarios within the same OA, and between several OAs, since they concern values which are normalized or expressed in reference to the OA surface data. Some KPIs can be also used to provide input data for emergency simulations.

Furthermore, the KPIs proposed in Table 4.4 are organized over the time t, still using the hourly sampling mentioned above, and they can be also assessed by separately referring to working days, holidays, and exceptional use conditions (e.g., fairs, exhibitions, concerts, other one-off events and mass gatherings). Furthermore, KPIs can vary depending on the season or the day of the week, depending on the specific OA use conditions. KPIs statistics on maximum values and average (for normal data distribution) or median (in case of non-normal data distributions) values can be calculated regardless of time, to respectively provide a quick description of critical and recurring OA conditions.

Emergency and evacuation simulations can be performed through the model defined in Chap. 3, Sect. 3.​4, by using input data about the OA morphology and layout, the position, quantity and quality of exposed users (Sect. 4.3), and the quantity and quality of the points of attack (Sect. 4.2). Due to the stochastic effects related to the users’ behaviours within the simulation models (i.e., initial user distributions, individual speed calculation, path selection and motion loops), a significant number of runs repetition (≥10) has to be performed for each scenario, and the general convergence indicators shown in Table 4.5 should be evaluated [22, 32, 33]. These indicators can be analysed to evaluate if the number of consecutive runs is enough to provide statistically-reliable simulation outputs. Thresholds for each indicator varies depending on the given acceptance criteria, but general works remarks that evacuation time and related standard deviation should be at least ≤ 5 ÷ 10% to ensure confident preliminary analysis.

Besides, convergence analysis, the statistical-based analysis of simulation results has to be performed also to derive KPIs for risk-assessment purposes [22, 34]. The first level of aggregation of data concerns the definition of the evacuation curve for the given scenario, since it traces the overall effects of evacuation interactions between users, the OA and its components, the attackers and their effects on the users and the OA. If the normality of simulation results could not be confirmed, the median evacuation curve, expressing the median number of users reaching a safe area (for the OA, one of the access streets) over time, should be considered. In fact, median values refer to the 50th percentile of distributions and they seem to be robust enough to trace results being not easily affected by extreme values in distributions [35]. In additional quartile-based curves (e.g., 5th, 25th, 75th and 95th) can be calculated. In particular, the curves referring the 5th and 95th percentile of users arrived at a safe are over time can trace the reasonable limits for the effectiveness of the simulated scenario, excluding behavioural outliers [33].

While these curves trace a time-based overview of the evacuation process, the KPIs listed in Table 4.6 summarize the overall risk conditions of a given terrorist act scenario in the OA. These indicators have been developed within the BE S²ECURe project to trace main behavioural issues in terrorist act evacuation [22], and to be consistent with previous works also concerning other kinds of emergencies, such as general purposes, fire and earthquake [36‐40]. To ensure the KPIs robustness [35], they take advantage of median values from simulation results on the set of simulation runs are considered as for the evacuation curve.

Moreover, the KPIs are normalized to make them ranging from 0 (minimum risk) to 1 (maximum risk). Therefore, they can compare different input scenarios on the same effects scale. In that sense, they can both compare several conditions related to the current scenario of the analysed OA, e.g., as in pre-retrofit conditions, by varying the simulation input factors related to the attack typology, the points of attack, the users’ exposure and vulnerability depending on the time of the day. Similarly, they can be used to compare pre-retrofit scenarios with post-retrofit scenarios implementing specific RMRSs, given that these RMRSs can modify the OA layout, the effects of the attack, the spatiotemporal distribution of the users, and also the user behaviours in emergency and evacuation. To this end, simulation models should be adapted to represent possible specific behaviours apart from those defined in Chap. 3, Sect. 3.​4.

The comparison between KPIs can be then performed in absolute terms, as the difference between the KPIs, since all of them range from 0 to 1, so as to derive how specific conditions scen can impact the KPI levels in respect of a given reference scenario ref. Nevertheless, percentage variation of a given KPI PV [%] can be calculated according to Eq. 4.9, which is based on previous works on behavioural-based design [41]. PV-based assessment can better stress the final KPI levels in respect of the original one. Indeed, both absolute differences of the KPIs < 0 and PV < 0% imply an increase in the users’ safety considering the KPIs in Table 4.6.

The mitigation of risk and the prevention of violent acts are related to the qualification of risk itself, the identification of intrinsic vulnerabilities and the full knowledge of the phenomenon. In that sense, the application of expeditious formulation for risk assessment combined with the setting up of possible critical scenarios in the real OAs involving users’ behaviour and the use of fast performance indicators allow the analysis of the identified scenarios in an as-built conditions. The reduction of risk and the improvement of resilience of urban place and users can be achieved through a comprehensive and effective system of strategies, which comprehend physical and technical solutions. If the effectiveness of such strategies can be solved by the standards and regulations, also coherently with the attack types, the compatibility of strategies requires to be declined in terms of compatibility of solutions with the real OAs. As already highlighted in discussed in the theories of the “design of security” in British counterterrorism activities [7, 43, 44], the transformation of the physical places may affect the integrity of a real place and the security perception of its users. Considering the relevance of cultural and symbolic places in the proneness to a terroristic attack [45], the resolution of compatibility became pivotal when applied in cultural or historic places. In that sense, all the RMRS strategies identified for their classification (discussed in Sect. 2.​2), a system of datasheets is setup, linking to each potential physical element involved in the strategies, technical solutions, and their possible levels of physical or aesthetical compatibility. The solutions are derived from the analysis of the current regulation about countermeasures of terrorist threat at the international level and properly highlighted in the following Table 4.7.