Every year natural hazards such as wildfires cause massive destruction of physical infrastructure and loss of lives. A wide range of activities is carried out at different stages of a wildfire under wildfire management to minimize the associated risks. Representing the dynamics of wildfires with complex mathematical and empirical in the form of wild- fire models is one of the effective ways to understand the behavior and form strategies against threatening wildfires. The current practice of wildfire management uses operational fire models such as Spark, Phoenix, FARSITE, and Prometheus, which are ideally expected to retrieve predictive information on the outspread of fires in as little time as possible. Risk metrics for a geographical location can be estimated by running multiple simulations, referred to as an ensemble prediction, with possible input factor conditions and conducting statistical analyses on simulation outputs. Such an approach of ensemble predictions can be analyzed in different ways to enable the estimation, analysis, and identification of the risks associated with wildfires. However, even a single simulation in an ensemble is a complex calculation based on interrelationships between different parameters and must also deal with geographical information data sets. Consequently, running ensembles on a single computer or a small cluster can result in bottlenecks due to data access and processing constraints and take longer than the window available for preparation for any imminent disaster. Research carried out in recent years has put forward Cloud Computing frameworks as a possible solution to increase the efficiency of the ensemble results from the prediction tools and make these services available to many users in a scalable way. Cloud infrastructure itself does not decrease the computation time for individual simulation in an ensemble. But it provides a means to reduce the overall time of the ensemble as it allows elastic on-demand access to almost unlimited storage, network, and computational processing. However, this access to the Cloud resources must be coupled with an effective control mechanism and innovative solutions in the system design to manage the resources and support the ensemble predictions in optimal manners to rapidly estimate, identify, and analyze the associated risks. As such, intending to enable ensemble predictions for rapid risk estimation, analysis, and identification, this thesis first presents the existing challenges through a comprehensive review of the adaptation of Cloud Computing in disaster modeling and management systems. To enable rapid ensemble wildfire predictions over Clouds for rapid risk estimation against the associated computational challenges, the thesis proposes a Cloud-based framework that offers ensembles of wildfire simulations as a service in a cost and time-efficient manner. As an improvement, the thesis extends the framework to facilitate running the fire simulations with sampled values of input parameters, referred to as sensitivity analysis, to perform a rapid risk analysis and determine the conditions with significant threats, which can be prohibitively time-consuming in local machines. Finally, against the naive comprehensive sweep methods in conventional ensemble predictions where simulations are run at all start locations to identify high-risk areas, the thesis proposes a novel quadtree-based search mechanism that can rapidly identify potential high fire-risk areas and produces an increasingly detailed risk map within a given time frame without running simulations at all start locations. The wildfire model analyses carried out with real use cases in the Tasmanian region verify the efficacy and usability of the proposed solutions in a real operational environment. The solutions proposed in this thesis are model-agnostic and can be easily transferred to other natural hazard models. This thesis adds to the body of the knowledge by making the following contributions: 1. A comprehensive survey that reflects the current research trends in utilizing ICT infrastructures including Cloud Computing to support different aspects of natural disaster management. 2. A validated foundation system design (framework) to deploy the ensemble of wildfire simulations as end services over the Clouds considering the user requirements with minimal cost for rapid risk estimation. 3. A brief report on parametric uncertainty quantification in Australian fire spread models used in Australian Fire Danger Rating System (AFDRS). 4. A comprehensive sensitivity analysis of input parameters in the widely used fire spread models with an insight into the implications of results on the understanding and interpretation of the fire models. 5. A cloud-based framework that can efficiently handle the high computational need of sensitivity analysis of operational disaster models for rapid risk analysis. 6. A novel and innovate quadtree-based mechanism for rapidly identifying areas of wildfire risk in operational management.
Copyright 2021 the author Chapter 2 appears to be the equivalent of a pre-print version of an article published as: KC, U., Garg, S., Hilton, J., Aryal, J., Forbes-Smith, N., 2019. Cloud Computing in natural hazard modeling systems: current research trends and future directions. International journal of disaster risk reduction, 38, 101188. Chapter 3 appears to be the equivalent of a pre-print version of an article published as: KC, U., Garg, S., Hilton, J., 2020. An efficient framework for ensemble of natural disaster simulations as a service, Geoscience frontiers, 11(5), 1859-1873. Copyright 2020 China University of Geosciences (Beijing) and Peking University. Production and hosting by Elsevier B.V. The published article is an open access article under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0) license (http://creativecommons.org/licenses/by-nc-nd/4.0/ Chapter 4 appears to be, in part, the equivalent of a pre-print version of an article published as: KC, U., Aryal, J., Garg, S., Hilton, J., 2021. Global sensitivity analysis for uncertainty quantification in fire spread models. Environmental modelling & software, 143, 105110. Chapter 4 also appears to include a pre-print equivalent of an article published as: KC, U., Garg, S., Hilton, J., Aryal, J., 2020. A cloud-based framework for sensitivity analysis of natural hazard models, Environmental modelling & software, 134, 104800.