@prefix vivo: . @prefix edm: . @prefix ns0: . @prefix dcterms: . @prefix skos: . vivo:departmentOrSchool "Applied Science, Faculty of"@en, "Electrical and Computer Engineering, Department of"@en ; edm:dataProvider "DSpace"@en ; ns0:degreeCampus "UBCV"@en ; dcterms:creator "Mirmoeini, Farnoush"@en ; dcterms:issued "2009-12-23T18:36:45Z"@en, "2005"@en ; vivo:relatedDegree "Master of Applied Science - MASc"@en ; ns0:degreeGrantor "University of British Columbia"@en ; dcterms:description """Situation assessment is the task of summarizing low-level sensor data in a battlefield environment to produce hypotheses suitable to use in command and control. In this thesis a novel algorithm is devised for adaptive multi-stage situation assessment using a hierarchy of Bayesian networks that are reconfigured on two timescales. Network Centric Warfare concepts are used in designing the situation assessment system. The formulation and algorithms presented are suitable for dynamic battlespace situation changes. Furthermore, algorithms are provided to model the battlespace conditions as a stochastic feedback system that uses the hypotheses generated by the Bayesian networks to make decisions. Numerical examples are provided to demonstrate the effectiveness of these algorithms."""@en ; edm:aggregatedCHO "https://circle.library.ubc.ca/rest/handle/2429/17270?expand=metadata"@en ; skos:note "Adaptive Battlespace Situation Assessment Using a Hierarchy of Reconfigurable Bayesian Networks by Farnoush Mirmoeini B . A . S c , University of British Columbia, 2003 A THESIS S U B M I T T E D IN P A R T I A L F U L F I L L M E N T O F T H E R E Q U I R E M E N T S F O R T H E D E G R E E O F Master of Applied Science in T H E F A C U L T Y O F G R A D U A T E S T U D I E S (Electrical and Computer Engineering) The University of British Columbia September 2005 © Farnoush Mirmoeini, 2005 Abstract Situation assessment is the task of summarizing low-level sensor data in a bat-tlefield environment to produce hypotheses suitable to use in command and control. In this thesis a novel algorithm is devised for adaptive multi-stage sit-uation assessment using a hierarchy of Bayesian networks that are reconfigured on two timescales. Network Centric Warfare concepts are used in designing the situation assessment system. The formulation and algorithms presented are suitable for dynamic battlespace situation changes. Furthermore, algorithms are provided to model the battlespace conditions as a stochastic feedback sys-tem that uses the hypotheses generated by the Bayesian networks to make decisions. Numerical examples are provided to demonstrate the effectiveness of these algorithms. ii Contents Abstract ii Contents iii List of Tables vi List of Figures vii List of Acronyms viii Acknowledgements xi 1 Introduction and Preview 1 1.1 Introduction 1 1.2 Thesis Overview and Contributions 3 1.3 Outline 7 2 Background and Literature Survey 9 2.1 Situation Assessment 9 2.2 Bayesian Networks 12 iii 2.3 Network Centric Warfare 16 2.4 Related Work 16 3 A Hierarchical Multi-Timescale Situation Assessment System Using Bayesian Networks 18 3.1 A Hierarchical Multi-Timescale Architecture for Situation As-sessment System 19 3.1.1 Sensor System and Integrated Tracker 22 3.1.2 Hierarchical Bayesian Networks for Situation Assessment 28 3.2 Multi-Timescale Reconfiguration of the Bayesian Networks for Adaptive Situation Assessment 35 3.2.1 Reconfiguration of the Lower Level Bayesian Networks 35 3.2.2 Reconfiguration of the Higher Level Bayesian Network 40 3.3 Dynamic Situation Assessment with Smart Targets 41 3.3.1 Situation Assessment Based Decision-Making in Bat-tlespace 43 3.3.2 Modeling the Response of Smart Targets to Decisions . 47 3.3.3 A Stochastic Feedback Algorithm to Model the Bat-tlespace Dynamics 52 3.3.4 Markovian Analysis of Battlespace Dynamics 54 4 Numerical Examples 60 4.1 Numerical Examples for Reconfiguration of the Situation As-sessment System 60 iv 4.2 Numerical Studies for the Decision Making System and Bat-tlespace Evolution 64 5 Conclusion and future work 67 5.1 Conclusion 67 5.2 Future Work 68 Bibliography 70 Appendices 72 Appendix A The Junction Tree Algorithm 73 v List of Tables 3.1 Higher and Lower Level Action Sets 46 3.2 An example of the lookup table used in determining the vari-ables that change after an action taken by friendly forces . . . 48 3.3 An example of the lookup table used in determining the vari-ables that change after an action taken by enemy forces . . . . 49 4.1 Mean Duration of the Battle and its simulated variance for var-ious Starting States 66 vi List of Figures 1.1 Structure of Situation Assessment System together with Feedback 4 3.1 Hierarchical architecture of the adaptive situation assessment system with sensors, integrated tracking system and two levels of Bayesian networks 19 3.2 Network Centric Warfare information structure with three levels of information sharing 21 3.3 Three lower level Bayesian networks 30 3.4 Structure of Higher Level Bayesian Network in Different Modes. (P denotes Protagonist and A denotes Antagonist) 33 3.5 Structure of adaptive situation assessment system together with stochastic feedback and decision making systems 42 3.6 Interaction of an Abstract Situation Assessment and Decision Systems 47 4.1 Likelihood of Lower level Bayesian Network in parametrical re-configuration 61 vii Likelihood the three different structures of the higher level Bayesian network in structural reconfiguration viii List of Acrony SA Situation Assessment BN Bayesian Network M L E Maximum Likelihood Estimation NCW Network Centric Warfare Infostructure Information Structure E M Expectation Maximization C2 Command and Control CPD Conditional Probability Distribution CPT Conditional Probability Table DAG Directed Acyclic Graph V E Variable Elimination JPN Joint Planning Network JDN Joint Data Network J C T N Joint Composite Tracking Network RWR Radar Warning Receiver ELINT Electronic Intelligence ix IR Infrared ESM Electronic Support Measure JDL Joint Directors of Laboratories IFF Identification of Friend or Foe E A Electronic Attack MDL Minimal Description Length x Acknowledgements I am grateful to Prof. Krishnamurthy for introducing me to such an interesting branch of applied mathematics and engineering, which, most probably, I would continue to work on for the rest of my career. I am also grateful to him for helping me learn the process of scientific writing through patiently going over numerous revisions of the papers that I wrote. The support for this work was partially provided by Defence Research and Development, Canada. I would like to thank the Countermeasure Tech-niques group, especially Ms. Christina O'Regan for their valuable comments and giving me the opportunity to work on their projects. My parents have let me make my own decisions since I was very young, and I am grateful for that. I thank my mother for always encouraging me to use my full potential and my father for encouraging me to be a well-rounded person. I thank my beloved Tissa for everything; things that I never feel I have enough space to write about. FARNOUSH MIRMOEINI The University of British Columbia September 2005 xi Chapter 1 Introduction and Preview This chapter gives an introduction to our work in solving the problem of situation assessment in airborne conflicts. 1.1 Introduction Situation assessment (SA) is the highest level of abstraction in understanding the conditions of a battlespace environment. It is the ongoing process of infer-ring relevant information about the forces of concern in a military situation. It is, in other words, the task of inferring \"what is going on\" or \"what is happen-ing\" from data collected by sensors and processed by lower level data fusion and target tracking systems in a battlefield environment. We define a situation as the general state of the battlespace in the combination of circumstances and conditions at a given time. A situation consists of entities, attributes and rela-tionships between these entities. Situation assessment or situation awareness is the task of identifying all entities within a situation and their relationships 1 and gaining knowledge and understanding about the battlespace. It is also the task of assigning significance and priorities to these entities and assessing how well each side is doing in a battlespace scenario. Situation assessment involves the task of making probabilistic inferences on the data collected by sensors. Advances in electronics and information sys-tems together with advances in sensor networks and Network Centric Warfare have facilitated collection of large amounts of sensor data in a short amount of time in a battlefield. Moreover, the collected data in prone to contain errors and missing values due to sensor malfunction, enemy forces' deceptive tactics and resource sharing (i.e. a sensor is not always dedicated to observe a certain quantity). The information from hundreds of sensors cannot be reviewed and processed by human operators at the same rate as they are collected, there-fore information systems are needed to summarize and fuse this incomplete and uncertain data into a number of comprehensible hypotheses that military decision makers can act upon. Moreover, as the battlespace situation changes with time, the assessment of the battlespace situation needs to be dynamically adapted to take these changes into account. Hence, an adaptive situation as-sessment system is needed that can be easily reconfigured as the battlespace situation changes. The goal of a situation assessment system is, therefore, to summarize and fuse the lower level data and generate up-to-date hypotheses about what situation the battlespace is in. A computational technique that is suitable in situation assessment should be capable of processing different types of uncertain, incomplete and possibly 2 conflicting types of data. It has to be able to reflect the dependence relation-ships amongst such data and be able to make inferences on it. Probabilistic reasoning using Bayesian networks is an artificial intelligence technique that satisfies all these requirements. Bayesian networks are an artificial intelligence technique for learning and reasoning with random variables. Bayesian net-works (BN) are particularly useful in situation assessment. They are a graph-ical representation of probability distributions that can combine information and make probabilistic inferences, and are able to handle causality and uncer-tainty in data using Bayesian probability theory. Bayesian networks provide a formal method of reasoning over uncertain and partially missing information. A situation assessment system should be able to make inferences on uncertain and incomplete information rapidly and economically, and efficient algorithms for making inferences in Bayesian networks exist. Furthermore, a situation assessment system needs to be adapted to account for changes that occur in the battlespace, and algorithms for efficiently learning Bayesian networks us-ing maximum likelihood estimation (MLE) exist. Bayesian networks therefore are the tool of choice to be used in an adaptive situation assessment system. 1.2 Thesis Overview and Contributions The main contributions of this work are summarized below: • A novel hierarchical Bayesian network based situation assessment system (Figure 1.1) is developed that is able to be reconfigured and adapted to the changes that occur in the battlespace over the duration of the con-3 Hypotheses (h) Higher-level Bayesian Networks Higher-level Decision Maker Evidence and Training Data Averaging Device Evidence and Training Data Lower-level Bayesian Netowrks Evidence Higher-level Action an Lower-level Decision Maker Integrated Tracking System Sensor Readings Sensor System Lower-level Action O i | Signals Targets and Battlespace Environment Figure 1.1: Structure of Situation Assessment System together with Feedback flict. The Bayesian networks used in situation assessment are hierarchi-cal and are dynamically reconfigured in two timescales to cope with the changes that occur as the battle proceeds on two different timescales. It uses an Expectation-Maximization (EM) type algorithm for the re-configuration of their parameters and structure. This reconfiguration occurs in two different timescales that correspond to two different lev-els of abstraction. We also show there is a need for the reconfigurable Bayesian networks to be used in situations assessment. The lower level Bayesian networks evolve on a fast timescale (typically a few seconds) and make low level hypotheses about the battlefield from data provided by an integrated target tracking system. These lower level Bayesian Net-works have fixed connections and are reconfigured parametrically on a faster timescale than the higher level network. The higher level Bayesian network represents higher level command and control decision variables (such as enemy resources) that do not have known dependence relation-ships and the dependence amongst them changes with time. The higher level Bayesian network is reconfigured both parametrically and struc-turally, but on a slower timescale (typically minutes). This results in a hierarchical reconfigurable Bayesian network that is able to process the data in different levels of abstraction and is able to produce high-level hypotheses about what is happening in the battlespace and how well each side is doing. These hypotheses can be used in decision making as well as simulations. The situation assessment system presented here 5 uses a hierarchy of Bayesian networks for estimating hidden variables and making inferences. This situation assessment system uses a Network Centric Warfare (NCW) system, in which robust networks for information sharing exist. NCW uses robustly networked forces in order to achieve improved information sharing capabilities and thus a higher level of situational awareness [1]. These networks have a variable quality of service that affects the level of information sharing in the networks. The situation assessment system provided here also fits in the hierarchy of information networks that form the information structure (infostructure) of Network Centric warfighting enterprise provided in [1]. We extend the situation assessment system by designing a stochastic dy-namic model of a battlespace. This model consists of a situation assess-ment based decision making system and Markovian model for friendly and enemy forces. The targets are assumed to be \"smart\" [14], that is to say they respond to the actions of the decision maker. The deci-sion making system mimics a Command and Control (C2) system. The effects of these actions on the enemy are measured by sensors and the situation is assessed again and new hypotheses are generated. This pro-cess allows us to analyze the battlespace dynamics off-line and predict the probability of winning an engagement and also the expected time to complete an engagement. This decision system has also been used to show the effectiveness of our situation assessment system, and can be re-6 placed with a human decision-maker or other more advanced automatic decision making systems. The architecture provided in this paper can be used in simulation of airborne conflicts, and also to construct a library of suitable modules of action under different battlespace circumstances. 1.3 Outline This thesis is organized in five chapters. In the next Chapter we present a literature survey on situation assessment and Bayesian networks. We first talk about the concept of situation assessment and what it entails according to various researchers and then state our own definition of situation assessment. In the second section of the chapter we introduce Bayesian networks briefly and in the third section we discuss Network Centric Warfare. At the end of this chapter we review the previous work done in the area of situation assessment using Bayesian networks. In Chapter 3 we explain the architecture of the hierarchical multi-timescale situation assessment system and provide information about each abstraction level and also discuss how Bayesian Networks are used in the sit-uation assessment system. We then discuss the reconfiguration algorithms to adapt different levels of the situation assessment system to cope with changes that occur in the battlespace over time. Furthermore, we describe the decision-making system used in this research, and also mathematical analysis of the situation assessment and decision-making system. We present some numerical results and examples of our work in Chapter 7 4. These numerical results show that there is a need for reconfiguration in a situation assessment system as well as providing numerical examples of our algorithms. Finally, Chapter 5 contains some important conclusions of this research and also guidelines for future work. 8 Chapter 2 Background and Literature Survey In this section we present background information about several important concepts discussed in this thesis. We start by introducing the concept of sit-uation assessment and stating its elements and properties. We then proceed to defining Bayesian networks and discuss the procedure of making inference on them and also learning them from data. Section 2.3 introduces Network Centric Warfare and its relationship to our work and Section 2.4 given a sum-mary of related work done in the area of situation assessment using Bayesian networks. 2.1 Situation Assessment In the previous chapter we introduced the concept of situation assessment and stated a working definition of it. However, it should be noted that there is 9 not a universal agreement on the nature of situation assessment and what it entails. Walz and Llinas [24] observed that: \"the SA process is relatively ill defined and many elements enter into a situation. In general, however, we are attempting to perform a contextual analysis of (data fusion) level 1 products (position and identity of individual entities). This next level is thus concerned with force deployments and as-sociated events and activities, knowledge derived from some type of pattern analysis applied to the Level 1 data. Moreover, this knowledge is coupled to a contextual background or setting that may involve both the physical and so-ciopolitical environments... The SA process therefore is concerned with what is happening and what events or activities are going to happen... it is focused on the behavioral aspects within the area of interest ... In SA force deploy-ments are considered in the context of environmental effects (e.g. terrain and weather)...(SA) can also be thought of as a means to estimate the enemy battle plan; that is, what the enemy is doing (activities) and attempting to achieve (intent, goals).\" Blackman and Popoli [3] observe that some authors assign different identities to situation and threat assessment. Walz and Llinas [24] also state that: \"...whereas situation assessment established a view of activities, events, maneuvers, locations and organizational aspects of force elements and from that estimates what is happening or going to happen, whereas threat assess-ment estimates the degree of severity with which engagements are to occur.\" 10 Blackmail and Popoli [3] further state that: \"The possible scope of situation assessment is enormous... situation awareness serves as a primary input into sensor management ... SA provides this contextual information (varying tactical requirements) for the sensor man-agements system ... SA serves to recognize patterns of behavior in tracked objects ... SA designs must strive to relieve workload of detailed control while preserving the ultimate control and authority of the operator ... a more overall role of SA is to act as a decision aid for the operator.\" They go on to state that feedback from a situation awareness capability to the underlying sensor and data fusion system would afford \"far reaching\" benefits: \"Situation Assessment (SA) can assist in target acquisition performance by recognizing potential multi-element enemy tactics ... SA can help cue the sensor management system to look for as yet undetected likely participant ... SA can help recognize the relative unimportance of maintaining individ-ual tracks ... SA could then aid the tracking data association process by authorizing a group track and interpreting conflicting spatially unresolvable ID signatures as a tactical identifier for the group as a whole. SA can rec-ognize the importance of establishing the strength of a potentially attacking formation and cue the sensor management for a raid count assessment.\" There is, however, a lack of total system approach to the problem of sit-uation assessment. In our research, we have tried to come up with a definition of situation assessment that entails as much of the most abstract quantities 11 of battlespace conditions as possible. This definition treats the battlespace situation as a quantity (though unobservable) that can be inferred from other observable quantities in a battlespace environment. This definition of situation assessment has led us to a situation assessment system that deals with these quantities in a hierarchical manner. This hierarchical architecture enables us to reconfigure different levels of abstraction in different timescales. We state our definition of situation assessment here once more: We define a situation as the general state of the battlespace in the com-bination of circumstances and conditions at a given time. A situation consists of entities, attributes and relationships between these entities. Situation as-sessment or situation awareness is the task of identifying all entities within a situation and their relationships and gaining knowledge and understanding about the battlespace. It is also the task of assigning significance and priori-ties to these entities and assessing how well each side is doing in a battlespace scenario. 2.2 Bayesian Networks In the previous section we explained why Bayesian networks are useful in sit-uation assessment. We now explain more about Bayesian networks and their properties and elements. Bayesian Networks are a powerful tool in making in-ferences from inconsistent and incomplete sources of data. They are a graphical representation of probability distributions that are capable to be learned from data. The capability of being learned from incomplete and possibly conflicting 12 sources of data, and also their capability to be reconfigured as the environment changes makes them an ideal candidate in situation assessment. A Bayesian Network (BN) is a model. It reflects the states of some part of a world that is being modeled and it describes how those states are related by probabilities. A l l the possible states of the model represent all the possible worlds that can exist, that is, all the possible ways that the parts or states can be configured. Typically, some states will tend to occur more frequently when other states are present. For example, if an unidentified aircraft is using fire control radar, it is more likely a fighter aircraft than a commercial carrier. Bayesian networks are networks of relationships. They encode the prob-abilistic relationship among several random variables. They consist of nodes and directed links. The nodes of a Bayesian network are random variables, which can be discrete or continuous (though we deal solely with discrete nodes in this work). The direction of the link arrows roughly corresponds to \"causal-ity\" . That is the nodes higher up in the diagram tend to influence those below rather than, or, at least, more so than the other way around. More precisely, the direction of an arrow represents the conditional probability of inferring the state of a particular node given the states of its parent nodes. In addition to the graph structure, it is necessary to specify the parameters of the model. For a directed model, we must specify the Conditional Probability Distribu-tion (CPD) at each node. If the variables are discrete, this can be represented as a table (CPT), which lists the probability that the child node takes on each of its different values for each combination of values of its parents. We can 13 now present a mathematically rigorous definition of Bayesian networks: A Bayesian network is a pair B =< G,Q >. G is a directed acyclic graph (DAG) that encodes the dependence relationships, in the sense that the nodes represent elements of U and the directed connections represent the conditional probability of inferring the state of a particular node given the states of its parent nodes UXi • G also encodes the conditional independence assumption that each node Xt is conditionally independent of the rest of the network given its direct parents i lx ; . © represents the set of parameters that quantify the network. For each node Xi it contains the conditional probability table (CPT) of each node, which in turn contains the conditional probability of a node given its parents. Parameters are of the form 9Xi\\nx. = P(xi\\UXi) for each possible value of Xt and Tixi • Therefore, a Bayesian network B defines a joint probability distribution over U that can be factored as follows: n PB(Xu...,Xn) = Y[pB(Xi\\UXi) (2.1) The most common task we wish to solve using Bayesian Networks is probabilistic inference. Inference involves calculating the conditional prob-ability of a set of nodes given the known state of another set of nodes in a Bayesian network. The underlying principle of the process of inference is Bayes Rule, which gives a way to calculate an unknown conditional probabil-ity value based on other known conditional probabilities. Let B\\, B 2 , B n be a partition of a sample space S so that U\"=1 Bi = S and Bi D Bj = 0 for all 0 < i,j < n. Suppose that event A occurs, we would like to know what is the probability of event Bk, that is, we would like to know P(Bk\\A). According 14 to Bayes Rule we have [18]: P(Bk\\A) - - ^ j - - E U m B j ) p { B j ) (2-2) Bayes Rule is highly useful in making inferences in Bayesian networks and this is the reason they are named after Bayes. The simplest algorithm for making inference in Bayesian networks is the Variable Elimination (VE) algorithm, which uses Bayes rule directly to calculate conditional probabilities. Other more efficient algorithms exist that are used in making inference in Bayesian networks. One of the most efficient ones is the Junction Tree algorithm that is explained in the Appendix. Bayesian networks, as mentioned earlier in this section, have the ca-pability to be learned and updated to model the current state of the world. Learning is one of the most important problems in Bayesian Network theory and involves the estimation of the probability tables and existence of connec-tions from data. This is called learning. Learning Bayesian Networks has two aspects: parametrical learning and structural learning. Parametrical learn-ing deals with determining parameter 0 or the conditional probability tables (CPTs) from data. Structural learning deals with learning the causal relation-ships that exist among the random variables that comprise the nodes of the Bayesian Network, and also determining which random variables are useful to be a node in a Bayesian Network. We will explain Bayesian network learning algorithms in the next chapter, where we use them to reconfigure the Bayesian networks in the situation assessment system. 15 2.3 Network Centric Warfare Network Centric Warfare (NCW) is a new scheme introduced by the US Armed Forces Joint vision 2010. Recent developments in information technology have caused a new conceptual scheme to be proposed in order to achieve informa-tion superiority. Information superiority is a state that is achieved when a competitive advantage is attained from the ability to extract a superior infor-mation position [1]. Network Centric Warfare uses robustly networked forces in order to achieve improved information sharing capabilities and thus shared situational awareness. We use the concept of Network Centric Warfare in our research by assuming that a network of sensors and also a network for higher level information sharing exist. 2.4 Related Work A number of papers discuss using Bayesian networks in Situation assessment. Kott et al. present a general overview of the situation assessment problem and discuss various research methods that can be used to approach the problem [13]. Bladon et al. advocate the usability of graphical models in situation assessment and show that they satisfy various requirements such as robustness [4]. Several papers present the use of Bayesian network in situation and threat assessment. Das et al. discuss the merits of Bayesian network technology as compared to other reasoning methods and introduce several scenarios and constricts Bayesian networks for them [7]. Das and Lawless [8] also propose 16 a truth maintenance system to check for information consistency in Bayesian networks when they are used in Situation Assessment. Nguyen discusses the application of Bayesian networks in threat assessment and breaks down threat assessment into capability and intent assessment [20]. He then introduces a sample network for intent assessment [20]. Okello and Thorns formulate the threat assessment problem and give a simplified but mathematically rigorous approach to it; using both discrete and continuous variables in the Bayesian network they propose [21]. Laskey and Laskey propose a Bayesian network approach to the problem of combat identification [16]. McMichael discusses a statistical approach to situation assessment using Bayesian networks and dynamic programming techniques [19]. Suzic uses Bayesian networks in enemy policy recognition [22]. A l l these researchers, however, treat situation assessment as a static problem and do not use any kind of reconfiguration. This results in situation assessment systems that are not a true representative of what goes on in a battlespace since the parameters and structure of these Bayesian networks are always constant. We also present a hierarchy of Bayesian networks for situation assessment which is able to be reconfigured on two timescales. The situation assessment system developed in our research also uses the broadest meaning on situation assessment, instead of focusing only on threat assessment or plan recognition. We also use the concepts of Network Centric Warfare in building our situation assessment system. 17 Chapter 3 A Hierarchical Multi-Timescale Situation Assessment System Using Bayesian Networks In this chapter we present the hierarchical multi-timescale situation assessment system we developed. We first explain the structure of such situation assess-ment system and then describe how the Bayesian networks in this situation assessment system can be reconfigured. In the third section of this chapter we present a stochastic feedback system that makes decisions in the battlespace. 18 Joint Planning Network (JPN) Hypothess (r4) Joint Data Network (JDN) Higher-level Bayesian Networks Evidence and Training Data Evidence and Training Data { Y 3 ) Averaging Device Lower-level Bayesian Netowrks Evidence andTraining Data fa) Joint Composite Tracking Network (JCTN) Integrated Tracking System Sensor Readigs ( T I ) Sensor System |.Signals Targets and Battlespace Environment Figure 3.1: Hierarchical architecture of the adaptive situation assessment sys-tem with sensors, integrated tracking system and two levels of Bayesian net-works 3.1 A Hierarchical Multi-Timescale Architec-ture for Situation Assessment System Consider a battlefield situation that comprises of friendly and enemy forces and their platforms, assets and interactions. At each time instance, sensors record measurements from the battlefield. The data collected by sensors is plentiful but usually low in information content; furthermore this data cannot be used directly in a situation assessment system because it is noisy and also 1 9 does not contain all quantities that represent the battlespace situation (since some quantities are not directly measurable and some are not observable all the time), therefore it needs further processing before it can be used in decision making. The adaptive situation assessment system consists of five functional-ity blocks, as outlined in Figure 3.1. The function of this hierarchical adaptive situation assessment system is to assess the situation by evaluating the likeli-hood of different hypotheses. The lowermost block is the sensor system, which collects data using both passive (e.g. Infrared) and active sensors (e.g. radar). The second block is the integrated tracking system (integrated tracker), which processes the sensor readings. The integrated tracker is mainly responsible for processing the incoming noisy sensor data and feeding it to the Bayesian networks for inference. We will explain the integrated tracker fully in the next section. There are two blocks of Bayesian networks. Al l of these Bayesian networks perform adaptive situation assessment. The reason we have two dif-ferent levels of Bayesian networks is that these two levels operate on different timescales and therefore they cannot be integrated in one large Bayesian net-work. For this situation assessment system, we introduce a multi-timescale model where each block processes the data on a faster timescale than the ones above it. We call these timescales T i , r 2 , r 3 and r 4 (rx