Abstract: Insight into performance ability is crucial for successfully implementing AI solutions in real-world applications. Unanticipated input can lead to false positives (FP) and false negatives (FN), potentially resulting in false alarms in fire detection scenarios. Literature on fire detection models shows varying levels of complexity and explicability in evaluation practices; little supplementary information on performance ability outside of accuracy scores is provided. We advocate for a standardized evaluation dataset that prioritizes the end-user perspective in assessing performance capabilities. This leads us to ask what an evaluation dataset needs to constitute to enable a non-expert to determine the adequacy of a model's performance capabilities for their specific use case. We propose using data augmentation techniques that simulate interventions to remove the connection to the original target label, providing interpretable counterfactual explanations into a model's predictions.

Abstract: In the field of urban resilience, there is a great diversity of approaches to measuring the level of resilience in cities. This information is scattered among reports and academic articles. In this ongoing research paper, we explore the potential of Topic Modeling to analyze this information, in order to determine cluster indicators for a set of academic papers and resilience frameworks. These clusters are referred to as Key City Resilience Indicators (KCRI), which are used as reference to facilitate the measurement of urban resilience regardless of the context, including all the key dimensions required for cities to achieve resilience. Topic modeling outcomes can be used to generate indicators based on each topic or to automatically classify a new set of indicators in each of the established topics. These results can be applied to any resilience framework

Abstract: Flooding poses a significant and growing risk for many urban areas. Stormwater systems are typically used to control flooding, but are traditionally passive (i.e. have no controllable components). However, if stormwater systems are retrofitted with valves and pumps, policies for controlling them in real-time could be implemented to enhance system performance over a wider range of conditions than originally designed for. In this paper, we propose an autonomous, reinforcement learning (RL) based, stormwater control system that aims to minimize flooding during storms. With this approach, an optimal control policy can be learned by letting an RL agent interact with the system in response to received reward signals. In comparison with a set of static control rules, RL shows superior performance on a wide range of artificial storm events. This demonstrates RL's ability to learn control actions based on observation and interaction, a key benefit for dynamic and ever-changing urban areas.

Abstract: New crisis response and management approaches that incorporate the latest information technologies are essential in all phases of emergency preparedness and response, including the planning, response, recovery, and assessment phases. Accurate and timely information is as crucial as is rapid and coherent coordination among the responding organizations. We are working towards a multi-pronged emergency response tool that provide stakeholders timely access to comprehensive, relevant, and reliable information. The faster emergency personnel are able to analyze, disseminate and act on key information, the more effective and timelier their response will be and the greater the benefit to affected populations. Our tool consists of encoding multiple layers of open source geospatial data including flood risk location, road network strength, inundation maps that proxy inland flooding and computer vision semantic segmentation for estimating flooded areas and damaged infrastructure. These data layers are combined and used as input data for machine learning algorithms such as finding the best evacuation routes before, during and after an emergency or providing a list of available lodging for first responders in an impacted area for first. Even though our system could be used in a number of use cases where people are forced from one location to another, we demonstrate the feasibility of our system for the use case of Hurricane Florence in Lumberton, a town of 21,000 inhabitants that is 79 miles northwest of Wilmington, North Carolina.

Abstract: Nowadays, rescue actors still lack backing to exchange information effectively and ensure a common operational picture. Several studies report a low adoption of communication systems in rescue operations as well as a negative position of actors to such systems. The real needs of stakeholders, simply put, are not satisfied by the offered systems. Observing this circumstance through a user-centred design focal point, we notice that such issues ordinarily originate from inadequate design techniques. For this reason, we aim to implement Rescue MODES, a communication system oriented to support awareness amongst French actors in rescue operations based on their needs. In this paper, we propose an approach and introduce a platform that allows final users to design system interfaces in a customised way. This approach is based on an application ontology and an interaction model.

Abstract: Shock forest fires have short and long-terms devastating impact on the sustainable management and viability of natural, cultural and residential environments, the local and regional economies and societies. Thus, the utilisation of risk-based decision support systems which encapsulate the technological achievements in Geographical Information Systems (GIS) and fire growth simulation models have rapidly increased in the last decades. On the other hand, the rise of image and video capturing technology, the usage mobile and wearable devices, and the availability of large amounts of multimedia in social media or other online repositories has increased the interest in the image understanding domain. Recent computer vision techniques endeavour to solve several societal problems with security and safety domains to be one of the most serious amongst others. Out of the millions of images that exist online in social media or news articles a great deal of them might include the existence of a crisis or emergency event. In this work, we propose a Multi-Layer Fusion framework, for Real-Time Fire Severity Assessment, based on knowledge extracted from the analysis of Fire Multimedia Incidents. Our approach consists of two levels: (a) an Early Fusion level, in which state-of-the-art image understanding techniques are deployed so as to discover fire incidents and objects from images, and (b) the Decision Fusion level which combines multiple fire incident reports aiming to assess the severity of the ongoing fire event. We evaluate our image understanding techniques in a collection of public fire image databases, and generate simulated incidents and feed them to our Decision Fusion level so as to showcase our method's applicability.

Abstract: Emergency Traffic Management (ETM) is one of the main problems in smart urban cities. This paper focuses on selecting an appropriate object detection model for identifying and counting vehicles from closed-circuit television (CCTV) images and then estimating traffic flow as the first step in a broader project. Therefore, a case is selected at one of the busiest roads in Christchurch, New Zealand. Two experiments were conducted in this research; 1) to evaluate the accuracy and speed of three famous object detection models namely faster R-CNN, mask R-CNN and YOLOv3 for the data set, 2) to estimate the traffic flow by counting the number of vehicles in each of the four classes such as car, bus, truck and motorcycle. A simple Region of Interest (ROI) heuristic algorithm is used to classify vehicle movement direction such as \quotes{left-lane} and \quotes{right-lane}. This paper presents the early results and discusses the next steps.

Abstract: This paper seeks to establish a machine learning driven method by which a military veteran with Post-Traumatic Stress Disorder (PTSD) is classified as being in a crisis situation or not, based upon a given set of criteria. Optimizing alerting decision rules is critical to ensure that veterans at highest risk for mental health crisis rapidly receive additional attention. Subject matter experts in our team (a psychologist, a medical anthropologist, and an expert veteran), defined acute crisis, early warning signs and long-term crisis from this dataset. First, we used a decision tree to find an early time point when the peer mentors (who are also veterans) need to observe the behavior of veterans to make a decision about conducting an intervention. Three different machine learning algorithms were used to predict long term crisis using acute crisis and early warning signs within the determined time point.

Abstract: A simulation-based model used to measure resilience indicators of the railway transportation system is presented. This model is tested through a perturbation scenario: the inoperability of a track which links two stations in the system. The performance of the system is modelled through two indicators: (a) the number of passengers that reach their destination and (b) the total delay of passengers after a serious perturbation. The number of passengers within a given station at a given time is considered as early warning in the model. Furthermore, a crisis management plan has been simulated for this perturbation scenario in order to help the system to recover quickly from this perturbation. This crisis management plan emphasizes the role and the importance of the proposed indicators when managing crises.

Abstract: The initial impact of a disaster can lead to a variety of associated hazards. By taking a multi-hazard viewpoint with respect to disaster response and recovery, there is an opportunity to allocate limited resources more effectively, particularly in the context of long-term planning for community sustainability. This working paper introduces an approach for extending quantitative resource allocation models to consider multiple interrelated hazards. The discussion is motivated by a literature review of existing models and then focuses on changes necessary to take the multiplicity of hazards into consideration in the context of decision support systems for disaster operations management.

Abstract: The explosion of an improvised nuclear device (IND), in any American city, would cause devastating physical and social impacts. These impacts would exceed the response capabilities of any city, state or region. The potential loss and suffering caused by an IND detonation can be dramatically reduced through informed planning and preparedness. By incorporating estimates of the impacts associated with the detonation of an IND into the planning process, jurisdictions can estimate the scale and scope of their response requirements. A prototype, computer-based tool was developed to quantify the human impacts associated with an IND detonation. Using various types of information such as the approximation of the prompt radiation footprint, blast footprint, and thermal footprint of the detonation, along with an estimation of the level of protection provided by building structures the system calculates the number and type of injuries that can be expected in a monocentric urban area.

Abstract: Assessing a system's vulnerability is a widely used method to estimate the effects of risks. In the past years, increasingly dynamic vulnerability assessments were developed to display changes in vulnerability over time (e.g. in climate change, coastal vulnerability, and flood management). This implies that the dynamic influences of management strategies on vulnerability need to be considered in the selection and implementation of strategies. For this purpose, we present a strategy evaluation framework which is based on dynamic vulnerability assessments. The key contribution reported in this paper is an evaluation framework that considers how well strategies achieve a predefined target level of protection over time. Protection Target Levels are predefined objectives. The framework proposed is inspired by Goal Programming methods and allows distinguishing the relevance of time-dependent achievements by weights. This enables decision-makers to evaluate the overall performance of strategies, to test strategies, and to compare the outcome of strategies.