In this paper, the necessity of analysing and classifying the risks of multi-hazard threats by their origin and extent has been investigated. The proposed method of multi-hazard risk evaluation is based on the disposition of the set of valuable objects at critical risk, the set of active threats, and the set of manpower and resources for response operations. The result of applying the method is a categorization of the situations which allows decision-makers and local representatives to timely make adequate decisions in real-time situations.
Since the full-scale Russian military troops invaded Ukraine, numerous buildings, bridges,
railroads, dams and other critical infrastructure objects have been exposed to numerous risks,
impairment, and demolishment, not mentioning the dire aftermath, such as casualties, devastation,
uncertainty and refugees. From the very beginning of the invasion, they started spreading havoc
and destruction, encroaching not only on military objects but also on residential areas, malls, gas
stations and critical infrastructure [
Moreover, there are not only physical threats to dissemination. The majority of critical objects
and infrastructure are under attack from hackers and blackmailers wanting to disrupt the integrity
of the system and outage entire regions and cities. Digital or cyber-threats comprise a type where
a criminal combines two or more ways to commit system breakthroughs [
All the above-mentioned threats are considered multi-hazardous and comprise a major percentile of the global scale. Generally, such threats are concentrated on undermining the defense and integrity of a country. Therefore, the European Union is devising stated risk assessment methods to properly inform aimed forces and representatives about emerging risks of threats. The aim is to evaluate the extent of the multi-hazard risk of threats and forward them to early warning systems and risk assessment mechanisms.
Those threats can be simultaneous and of different kinds, e.g., bombing, cyber-threats and
shelling. They are considered destructive forces as they demolish both critical and civil
infrastructure [
Nevertheless, when there is a risk of a disaster occurring, it may lead to numerous interwoven disasters causing each other, called domino, or cascading effects, and are explicable through various scientists.
The next section is devoted to analyzing the related research pieces of work. The subsequent section depicts the problem and the solution methods. Accordingly, the fourth section presents the achieved results and a description thereof. Ultimately, the last section comprises the used literature.
The multi-risk assessment and analysis are conducted by numerous scientists, and our local
researchers have also performed certain investigations. The paper [
The paper [
The paper [
The anthropogenic, or human-driven threats are sometimes referred to as terrorism and
disobedience. These may be cyber-attacks, explosions, critical infrastructure malfunction or
invasion; transportation accidents, failures, and hazardous material accidents (Fig. 1).
Authors of the paper [
Also, in the paper [
The paper [
There are a lot of techniques representing interactions and interrelations among hazards. In this paper, we are working on the proper classifier for the potential hybrid risk assessment. Classification is a part of data analysis and pattern recognition that requires a class label for the described instances by a set of attributes and can be implemented in various ways ranging from decision trees, graphs, lists, neural networks, random forests, and k-nearest classifiers.
One of the most efficient classifiers, in the sense that its predictive performance is competitive
with state-of-the-art classifiers, is the so-called Naive Bayesian classifier, which learns from the
training data, the conditional probability of each attribute Ai with the class label C. Classification
is then performed by applying Bayes' rule to calculate the probability of a given C particular
instance A1, ..., An and then predicting the class with the highest posterior probability. This
computation is made feasible by a strong independence assumption: all attributes of Ai are
conditionally independent by class value C [
Another Bayesian classifier is a Bayesian network (BN) or directed graphical model (DGM) which represents itself as the joint probability distribution of a set of random variables with possible causal relationships. The network consists of nodes representing random variables, edges between pairs of nodes representing the causal relationship and a conditional probability distribution (CPD) at each node. The main purpose of the method is to model the posterior conditional probability distribution of the variable after observing new evidence. Bayesian networks can be built either manually with knowledge of the underlying domain, or automatically from a large dataset using e.g., Python libraries.
Bayesian networks [
A traditional Bayesian network [
=1 ( 1, … , ) = ( | ( ))
The concept of Bayesian networks [
The novelty of this work in comparison with the related ones is that the proposed risk analysis covers all three main phases of the crisis management cycle such as pre-crisis, response, and postcrisis. Within each phase, the risk is evaluated differently. As a result, it is divided into potential (for the pre-crisis phase), active (for the response phase), and posterior risk (for the post-crisis phase), which allows decision-makers to make more conformed decisions at every phase of the crisis management cycle.
Hereby, we suggest a crisis management cycle that is divided into three phases: pre-crisis, response, and post-crisis. The main characteristics of risk in the research context imply that it is dynamic and spatially distributed. As for the first one, we presume that for any spatial location, risk will change depending on affecting of multi-hazard threats. For the latter, let us suppose the risk can be evaluated for each area of the territory or for each vulnerable object, which provides its spatial reference. The subject of risk analysis in the proposed model is the evaluation of the chance of losses as a result of the involvement of the target object (here – critical infrastructure objects).
The risk originates from the interaction of multi-hazards and targeted objects, such as infrastructure (including critical), communities and governments affected by that threat. In this regard, risk dimensions include: 1) the probability of threat occurrence which depends on whether the actor has a certain goal and threat potential which depends on the availability of tools used by the actors (if applicable); 2) the characteristics of the targeted object such as object vulnerability, potential damage, and the speed of object restoration; 3) the availability of the object for the actors created the hazard.
Subsequently, multi-hazard risk assessment can be represented as a combination of the following components: 1) evaluation of the probability of a threat occurrence (PT); 2) availability of tools at the disposal of the actor (threat potential evaluation) (ET); 3) availability of targeted object for actor (AO); 4) vulnerability of targeted object (VO); 5) speed of object recovery (SO).
Thus, a qualitative multi-hazard risk evaluation for the targeted object at any time moment t will be a point or an area in the n-dimensional space of qualitative values of the multi-hazard risk components:
( ) = ( , , , , , )
This risk evaluation is dynamic and can be assigned to each vulnerable object and area of the territory. Next thing, we create a directed graphical model (Fig. 3).
The above model represents bombardment (B), cyber-threats (C), and potential or imposed damage (D) to the infrastructure (I), including critical objects. The probability algorithm for bombardment (B) is constructed below:
Let’s modelling this with a Bayesian network. Assume that we have two risk levels: low (0-0.5) and high (0.5-1). Our goal is to classify the risks given according to their uprising intensity utilizing the classification method. Let’s create the case probability table (see Table 1).
In this case, we can apply the Bayes theorem to model the probability of infrastructure damage (D) given the occurrence of air bombardment (B) and cyber-threats (C). We can express this probability as: ( | , ) = ( | ) ( | ) ( ) ( , ) where P(D | B, C) is the posterior probability of D given B and C, P(D) is the prior probability of D, P(B | D) is the conditional probability of B given D, P(C | D) is the conditional probability of C given D, and P(B, C) is the joint probability of B and C.
In the Bayesian network, we can represent these probabilities using conditional probability tables (CPTs) and prior distributions for the variables B, C, and D. The CPTs specify the conditional probabilities of each variable given its parents in the network, while the latter represents the initial probabilities of each variable before any evidence is observed.
Specifically, the Bayesian network for multi-risk critical impairment includes the following components: ▪ ▪ ▪ the B variable represents the occurrence of bombardment, which has a prior distribution P(B) = [0.6, 0.4] for low and high-risk levels; the C variable represents the occurrence of cyber-threats, which has a prior distribution P(C) = [0.7, 0.3] for low and high-risk levels; the D variable represents the potential damage to infrastructure, which depends on both B and C.
In particular, the CPT for D given B and C specifies the following probabilities: This CPT specifies that the probability of infrastructure damage depends on both the levels of bombardment and cyber threats. For example, if both B and C are low, then the probability of D being low is 0.9, and the probability of D being high is 0.1. Using the Bayes theorem, we can compute the posterior probability of D given specific values of B and C, based on the prior probabilities and the conditional probabilities specified in the CPT. For example, if we observe that both B and C are high-risk, then we can compute the posterior probability of D being low-risk as follows: ( = | = ℎ, = ℎ) = ( = ) ( = ℎ | = ) ( = ℎ | = ) ( = ℎ, = ℎ)
Based on the given DAG of the Bayesian network, we can define the joint probability distribution:
( , , ) = ( ) ∗ ( ) ∗ ( | , )
In the expression above, P(B) and P(C) are the marginal probabilities of the nodes B and C, and P(D | B, C) is the conditional probability of D given B and C. We can now represent this using a probability table (probability matrix). Let's assume that each node can take on only two values: 0 or 1, representing the absence or the presence of each event.
( ( ( (
( = 0) = 0.6 ( = 0) = 0.8 = 0 | = 0, = 0) = 0.9, ( = 1 | = 0, = 0) = 0.1 = 0 | = 0, = 1) = 0.3, ( = 1 | = 0, = 1) = 0.7 = 0 | = 1, = 0) = 0.2, ( = 1 | = 1, = 0) = 0.8 = 0 | = 1, = 1) = 0.01, ( = 1 | = 1, = 1) = 0.99
Global warming leads to snowless winters, which in turn cause a decrease in floodwaters, shallowing the rivers and lakes and thus they desiccate. Moreover, anthropogenic factors also create problems. For instance, such an anthropogenic factor as illegal sand mining leads to the Dnipro River shallowing: the soil moves towards the formed voids, the pits are tightened, but the river channel changes (see Fig. 5).
The next set of issues is tightly associated with industrial, agricultural and domestic wastewater. Phosphates, which enter the Dnipro in unlimited quantities, become the main cause of water bloom. The decomposition products of algae absorb oxygen and evolve into an ideal breeding ground for bacteria and lack of oxygen result in fish extinction.
The normal existence of fish is hindered by obstacles that stand in the way of their migration, which is caused by a change in the chemical composition of the water caused by an increase in its temperature due to a slowdown in its flow due to the corresponding influence of hydroelectric power plants built on the river. 5.1.
Here is the example of artificial coniferous forests planted around the Lower Dnipro Sands (Oleshky Sands) in the Kherson region, in Southern Ukraine. Those sands sometimes are qualified as a semi-desert. Sands are surrounded by dense coniferous forests planted to prevent dunes from moving. Despite the relatively small areas of the steppes, they are composed mainly of sand, so they often experience sandstorms.
The first reason leading to deforestation is global warming. Another reason is the increasing frequency of forest fires, the scale of which can already be regarded as a global disaster. In addition, forests are prone to insect infestations, which are destroying them at an increasing rate. A completely different reason is that the underground level of water is falling more and more, causing the forests to dry out, which leads to the desolation of forests covering large areas. Such territories gradually turn into sand deserts and causes the movement of sands (Fig. 6).
Unfortunately, in recent years, these processes have acquired a systemic character, significantly changing the natural landscape in many parts of the territory, which causes serious impacts on the ecosystem and affects slowly proceeding climate changes, exacerbating them. 5.2.
Shallowing the rivers and lakes in Polissya (the northern part of Ukraine, both the part of the territory of Belarus and Poland) gives rise to fires in Chernobyl. Thereby, the ecological danger in the exclusion zone is caused by the presence of nuclear and radiation hazardous objects. Unfortunately, radioactive contamination can spread far beyond this zone, especially as a result of fires. Clearly, the ecosystem of the Kherson region suffers by itself, and, although it causes significant risks to human life, but does not pose an immediate threat to life and health of people. Unlike it, the ecosystem of the Chernobyl exclusive zone is under significant risk of the transfer of radioactive dust that settle down at the forests in this region. The influence of various factors such as strong winds and precipitation during large scale forest fires pose immediate and permanent risk to people health and life (Fig. 7).
The proposed model has been implemented using Visual C++ based on the Free-BN Library and approbated on the simulated area.
The simulation area is the Lower Dnieper Sands (Oleshky Sands) in the Kherson region, in Southern Ukraine. The sands are surrounded by very dense artificial coniferous forests that prevent the sands from moving during strong winds. Global warming leads to the loss of forests in this area. As a result of global warming, we can observe chains of cascading effects. Due to warming, the groundwater levels are decreased, which further increases fire danger, rapid destruction of forests in large areas, desertification of the territory, and the revival of sand movement. Due to warming, forests are also being affected by invasions of insects, and also become more prone to forest fires (Fig. 8).
The results of the conducted simulation show that the proposed model provides enough performance to real-time modeling of a wide range of natural processes from climate change to forest fires and adequate knowledge representation about cascading events taking into account the uncertainty of the observations. The developed software contains a set of functions and procedures that allow defining events, their hierarchies, build DAGs, and assign probabilities transforming the event model into Bayesian networks. The infrastructure hierarchies can also be defined based on the spatially referenced objects’ definitions. The developed software is intended for use in the decision support system, which aimed to assess dangers, threats and risk with respect to the pre-defined objects.
The Bayesian Network multi-hazard risk model has been approved and tested within the simulated spatial model of Kherson Area, Ukraine. The simulations of multi-hazard disasters have been carried out to assess dangers, threats and multi-risk posed to various objects within the simulated area by multi-hazard disasters. We evaluate the performance of the decision-support queries directed to assessing multi-risk for the separate objects (buildings), their multitudes (quarters), and the entire areas. Thus, the query response time has been evaluated and averaged. The obtained result of the simulation is represented in Fig. 9. The adequacy of the multi-hazard risk model is confirmed by the above-mentioned experiment. The obtained results allow us to confirm that the developed multi-risk assessment model based on the Bayesian Network provides sufficient performance. Thus, it allows to simulate multi-hazards that contain cascading and triggering effects such as shallowing of Dnipro river and lakes, as well as variety of deforestation issues, dehydration of water sources, etc. Of course, forest fires, tornadoes, sandstorms, and other disasters can be also considered.
Due to the sufficient efficiency of the proposed models, the decision-support system helps decision-maker timely assess threats and risks from emerging events for various infrastructures and objects spatially distributed in an analyzed area of interest.
In this paper, the necessity of analysing and classifying the risks of multi-hazard threats by their origin and extent has been investigated. The proposed method of multi-hazard risk evaluation is based on the disposition of the set of valuable objects at critical risk, the set of active threats, and the set of manpower and resources for response operations. The result of applying the method is a categorization of the situations which allows decision-makers and local representatives to timely make adequate decisions in real-time situations.
The crisis management cycle is propped to divide into three phases: pre-crisis, response and post-crisis is presented. The risks according to their uprising intensity utilizing the classification method were sorted. The comprehensive approach of combining relevant tools to prevent, counteract, and recover from the impact of multi-hazards in a coordinated manner to support riskinformed decisions at all phases of the crisis management cycle is depicted. Moreover, the novel method of multi-hazard threat classifying is proffered. As a result, it is divided into potential (for the pre-crisis phase), active (for the response phase), and posterior risk (for the post-crisis phase), which allows decision-makers to make more conformed decisions at every phase of the crisis management cycle. [18] A.V.Kolesnyk, S.V.Smelyakov, P.H.Berdnyk, O.I.Kolodyazhnyy, A Bayesian networkbased method for assessing the risk of engine failure on an aircraft in flight, collection of scientific papers of the Kharkiv National University of the Air Force, 2(64), 2020, pp. 53-60 [19] N.Ruozzi, Bayesian Networks, Erik Jonsson School of Engineering & Computer Science at the University of Texas at Dallas. URL: https://personal.utdallas.edu/~nrr150130/gmbook/bayes.html C [20] J.van Westen, S.Greiving, N.R.Dalezios (Ed.), Chapter 2, Multi-hazard risk assessment and decision making, in: Environmental Hazards Methodologies for Risk Assessment and Management, IWA Publishing, 2017, p. 64