Post-disaster recovery assessment using remote sensing image analysis and agent-based modeling
Due to the COVID-19 crisis measures the PhD defense of Saman Ghaffarian will take place online in the presence of an invited audience.
The PhD defence can be followed by a live stream.
Saman Ghaffarian is a PhD student in the department of Earth Systems Analysis (ESA). His supervisors are prof.dr. N. Kerle from the Faculty of Geo-Information Science and Earth Observation (ITC) and prof.dr. T. Filatova from the Faculty of Behavioural, Management and Social Sciences (BMS).
One of the main components of the disaster risk management (DRM) cycle is recovery; however, it is the least understood one. Post-disaster recovery is the process of reconstructing communities in all their aspects (e.g., physical, economic, social, and environmental) in order to return life, livelihoods, and the built environment to their pre-impact or event better states. In this regard, timely and reliable information about the states of the damage and recovery process is vital for disaster planners and governments to make decisions. Remote sensing is an effective tool in providing information for post-disaster impact and recovery evaluations due to its agile data acquisition, synoptic perspective, growing range of data types, and instrument sophistication, as well as low cost. However, there is a need for diverse information to address the recovery process (including socio-economic aspects) comprehensively, and only a few of the parameters of interest can be extracted directly, while the majority have to be elicited indirectly.
In addition, although the variable recovery across different neighborhoods/sections of an affected area can be revealed by processing remote sensing data, there is a need for effective tools to explain those observations and identify suitable means to influence the recovery process (find bottlenecks, etc.). Spatial economic modeling, in particular, Agent-based modeling (ABM), permits to explore the dynamics of the recovery process from the bottom up. In an ABM, agents (decision-making entities) interact with each other and their environments to decide and act based on defined rules for their behavior in a specific situation such as a recovery process. Therefore, it constitutes an opportunity for policy-makers to test different policy scenarios in an artificial simulation environment and explore their consequences. Accordingly, policy and decision-makers can take advantage of the simulation outcomes to steer the recovery process. According to the above-described two main issues, the objectives of this research were first to develop remote sensing-based approaches to assess the post-disaster recovery and second, to explain further and explore the recovery process using ABM. In this regard, several investigations were conducted, and several approaches were developed as summarized below:
1) Review of remote sensing-based proxies for disaster risk management: While several spatial and non-spatial parameters required for detecting and quantifying of DRM-related elements can be extracted directly from RS imagery, many have to be elicited indirectly. Hence, a comprehensive review of the current remote sensing-based proxies developed for urban DRM and resilience were conducted. The proxies were sorted for two risk elements typically associated with pre-disaster situations (vulnerability and resilience), and two post-disaster elements (damage and recovery). The proxies were reviewed in the context of four primary environments and their corresponding sub-categories: built-up (buildings, transport, and others), economic (macro, regional and urban economics, and logistics), social (services and infrastructures, and socio-economic status), and natural. All environments and the corresponding proxies were discussed and analyzed in terms of their reliability and sufficiency in comprehensively addressing the selected DRM assessments. We highlight the strength and identify gaps and limitations in current proxies, including inconsistencies in terminology for indirect measurements. A systematic overview for each group of the reviewed proxies was presented that could simplify cross-fertilization across different DRM domains and may assist the further development of methods. While systemizing examples from the wider remote sensing domain and insights from social and economic sciences, a direction for developing new proxies was suggested, also potentially suitable for capturing functional recovery.
2) Post-disaster recovery assessment with remote sensing: A conceptual framework: The collected remote sensing-based information for pre- and post-disaster situations should have been used synergistically to address the post-disaster recovery assessment. Hence, a conceptual framework to monitor and evaluate the post-disaster recovery process and resilience was developed. In particular, available remote sensing image-based proxies were used to assess the recovery addressing not-only physical but also functional aspects. Also, this conceptual framework can be used to evaluate disaster resilience, assuming that the speed of the recovery is a proxy for resilience assessment. The proxies were mostly extracted using machine learning-derived land cover and land use maps. The proposed approach was used to assess the recovery of barangays (municipalities), including Tacloban city, in the Leyte region in the central Philippines.
3) Improving the precision of the damage and recovery assessments through post-disaster debris identification: Most of the developed remote sensing-based damage detection methods have a common limitation, and it is using debris as a proxy for damage detection for both building/structural and regional damage detection. Thus, distinguishing the structural rubble from ephemeral debris can increase the accuracy of the damage and recovery assessments. The limitations of using debris and rubble piles as proxies for damage detection and subsequent post-disaster recovery assessment from remote sensing images were discussed, and two different approaches for post-disaster debris identification were investigated. Three feature extraction methods, i.e., Gabor filters, Local Binary Pattern (LBP), and Histogram of the Oriented Gradients (HOG), were investigated to identify the debris from UAV images. As the second strategy, an approach was proposed, which monitors the multi-temporal satellite images acquired days and weeks after the disaster to figure out the relation between debris type and their time of removal. The approaches were tested for Tacloban city using UAV and multi-temporal satellite images.
4) Automated deep learning-based post-disaster building database updating: The location of the damaged, reconstructed, and newly constructed buildings provide critical supporting information for both first responders and recovery planners after a disaster. The proposed method makes use of free OpenStreetMap building footprints available for a pre-disaster situation to automatically collect training areas from very-high-resolution satellite images for a convolutional neural network (i.e., U-net) which is supported with residual connections. The trained network is then transferred and retrained for the post-disaster situation at any time after a simple building-based change detection analysis over OSM data. The proposed approach was tested for different scenarios of damage and recovery assessments in very high-resolution satellite images selected from Tacloban, the Philippines, after Typhoon Haiyan. The results showed that the proposed approach significantly decreased the manual work of training area collection, while maintaining the accuracy of the detected damaged, reconstructed, and newly constructed buildings at a high level.
5) Post-disaster recovery monitoring with Google Earth Engine: Most of the developed RS-based approaches in post-disaster damage and recovery assessments focus on the use of costly very-high-resolution data that require extensive digital storage and computing capacity to make use of them. However, cloud-based platforms such as Google Earth Engine (GEE) provide free RS data and computing power with a coding environment to develop and implement user-defined methods and process the data. Hence, the aim of this study was to test the suitability of GEE for a large-scale post-disaster recovery assessment. To do so, the GEE was employed to assess the recovery process over a three-year period after Typhoon Haiyan, which struck Leyte island, in the Philippines, in 2013. The following steps were developed and followed (i) generate cloud and shadow-free image composites from Landsat 7 and 8 satellite imagery and produce land cover classification data using the Random Forest method, and (ii) generate damage and recovery maps based on post-classification change analysis. The method produced land cover maps with accuracies >88%. The model was used to produce damage and three time-step recovery maps for 62 municipalities on Leyte island. The study showed that GEE has good potential for monitoring the recovery process for extensive regions. However, the most important limitation is the lack of very-high-resolution RS data that are critical to assess the process in detail, in particular in complex urban environments.
6) Agent-based modeling of the post-disaster recovery with remote sensing data: Recovery planners and decision-makers need to monitor and collect information about the ongoing post-disaster recovery process and understand the effect of different response strategies in dynamics. Monitoring tools/data such as geospatial platforms/remote sensing has been recently employed to assess the post-disaster recovery. However, remote sensing can mostly provide information regarding the physical aspects of the recovery, which are relatively easy to monitor and evaluate compared to functional assessments that include social and economic processes. Therefore, there is a need for a tool to understand and explore the impacts of different dimensions of the recovery, including the behavior of individual actors and their interactions with socio-economic institutions. An agent-based model was developed to simulate and explore the recovery process in urban areas of Tacloban, the Philippines devastated by Typhoon Haiyan in 2013. The recovery patterns of the formal and informal (slum) sector households, which follow different decision-making strategies in the recovery process, were studied differently. The satisfaction of the formal building and slum households were tracked and mapped to understand and demonstrate each of which recovery patterns. In addition, the disaster resilience of the targeted groups was assessed, given that the speed of the recovery is a proxy for resilience. Also, the effect of the unemployment rate and presence of a relocation site far from urban areas and workplaces after a disaster were experimented using the developed model. The developed post-disaster recovery model can be used by decision-makers to understand the recovery process and carry out the most influential factors and components.