Big Earth data for disaster risk reduction

KEYWORDS:

• Big Earth data
• disaster resilience
• risk reduction
• remote sensing
• social sensing
• deep learning

1.

In an ever-changing world, where the frequency and intensity of natural and human-made disasters are on the rise, disaster risk reduction has emerged as a crucial focal point of interdisciplinary research, governance, and public discourse. Disaster risk reduction, which aims to safeguard humans and protect environments from hazards and threats, is of high societal relevance and closely related to several of the United Nations Sustainable Development Goals (SDGs). The findings from research into disaster risk reduction contribute significantly to making cities and other settlements more inclusive, safe, resilient, and sustainable.

Providing timely and reliable information to authorities and stakeholders in the form of accurate pre-disaster conditions, post-disaster damage, and recovery progression is critical for many stakeholders such as government agencies, insurance companies, and city planners to undertake response management, resource allocation, recovery activities, and preventive action to minimize the destructive effects of disasters. At the forefront of this quest is the rapidly evolving domain of Big Earth Data, which provides a revolutionary perspective on monitoring disaster evolution, impacts, and human responses. This wealth of information, sourced from advanced remote sensing, social media platforms, mobile applications, vehicles, sensor networks, and crowdsourcing efforts, facilitates a multi-dimensional lens to observe human activity, population dynamics, urban development and expansion, and ecological shifts before, during, and after disaster events. The insights gleaned from these data streams are pivotal for understanding disaster dynamics and their compound, cascading effects on human health, well-being, infrastructures, and the environment. Such understanding is critical in informing evidence-based decision-making and policymaking aimed at foreseeing and mitigating disaster risks and enhancing resilience.

This special issue comprises six articles that utilize Big Earth Data to measure, monitor, model, and manage present and future disaster risks, guiding the identification of optimal strategies for reducing disaster risks and promoting resilience and sustainability.

The first article in this special issue explores the application of machine learning techniques to refine open-access gridded climate data for the creation and management of weather index insurance in regions with limited data availability (Eltazarov et al., 2023). Weather index insurance is an innovative form of insurance that leverages predetermined indices, such as rainfall levels, to initiate payouts, circumventing the need to assess actual damages like crop yield reductions. However, the efficacy of such insurance is contingent upon the availability of climate data with long-term historical records, large geospatial coverage, and fine spatial resolutions, which are generally unavailable in open-access databases. To resolve this challenge, this study trained a random forest algorithm to enhance three vital climate parameters – temperature, precipitation, and soil moisture – at a 5 km spatial resolution, utilizing 30 km resolution data from ERA-5 and RSA databases. Focusing on county-level spring wheat yield data from 1982 to 2018 across 56 counties in Kazakhstan and Mongolia, the research demonstrates an increase in the hedging effectiveness of index insurances in 70% of the counties when employing the refined climate data. This finding underscores the capabilities of machine learning in processing large-scale, coarse-resolution earth observation data to generate reliable fine-grained climate data and mitigate the risks associated with weather index insurance.

The second article proposes a novel framework empowered by deep learning algorithms for landslide detection from multi-temporal Sentinel-2 satellite remote sensing images (Ghorbanzadeh et al., 2023). Landslides, triggered by factors such as earthquakes, human activities, and intense rainfall, are escalating in frequency and intensity due to climate change. This paper explores a new strategy for landslide detection by integrating a Fully Convolutional Network (FCN) model with rule-based Object-Based Image Analysis (OBIA). The study trains the ResU-Net model with diverse landslide inventories and tests it in untrained areas. It incorporates land cover and image difference indices from pre- and post-landslide images into the OBIA for enhanced segmentation. The enhanced OBIA framework refines and classifies landslide objects by applying rules grounded in expert knowledge. This integrated approach significantly improves the detection of landslide boundaries, optimizing the mean intersection-over-union metric by 22%. The research advances the use of remote sensing for rapid, accurate landslide mapping, crucial for effective disaster response and risk mitigation.

The third article introduces the street-level monitoring videos into extracting and visualizing the extent of urban flood (Wang & Ding, 2023). Urban flooding is a growing crisis that wreaks havoc on societies, but it is challenging to monitor and detect due to the complexity of urban drainage systems and lack of real-time, localized flooded area data. Traditional methods of flood data collection are hampered by high costs and technical difficulties. Emerging smart city technologies, especially traffic cameras and smartphones, are becoming invaluable for real-time flood data acquisition. However, processing challenges persist due to changing lighting conditions, complex urban scenes, and the need for heavy manual labeling in developing algorithms. This investigation introduces a novel semi-supervised flood water recognition scheme, requiring minimal labeling, coupled with monoplotting, a visual data projection method that facilitates high-quality, georeferenced data processing. By utilizing a combination of machine learning (i.e. Felzenszwalb algorithm) and photogrammetry techniques, the research secceeds to analyze urban flooding in unprecedented detail, overcoming the typical obstacles of variable weather, lighting, and background contrast. The innovative approach is promising in informing better urban drainage designs and flood mitigation strategies by providing accurate mapping of local flood extents and behaviors. Meanwhile, the study’s findings are poised to enhance decision-making for urban flood risk reduction.

The fourth article attempts to mitigate the risk of landslides from a forward-thinking perspective by mapping landslide susceptibility with weighted-based machine learning methods (Trinh et al., 2023). Ha Giang province, Vietnam, one of the most landslide-prone areas, is selected as the study area. Although traditional approaches like Analytic Hierarchy Process (AHP) and Frequency Ratio (FR) methods are widely used to assess such risks, their effectiveness is limited by the quality and availability of data. This study combines the strengths of expert-driven weighting methods (AHP and FR) with the analytical power of machine learning, namely random forest, support vector machine, and logistic regression to produce landslide susceptibility for Ha Giang. The chosen approaches are validated through statistical metrics and field-verified landslide data. Among the tested methods, Random Forest with AHP or FR weighting emerges as the most effective, suggesting its suitability for landslide risk reduction efforts. The research demonstrates the value of integrating expert knowledge with Big Earth Data and machine learning to improve landslide prediction and mapping, contributing to more informed decision-making for land management and civil protection in susceptible areas.

The fifth article epitomizes the definition of disaster resilience, popular measurement methods, and the emerging uses of Big Earth Data in measuring and enhancing disaster resilience (Qiang et al., 2023). Recent years have seen a surge in efforts to create quantitative resilience measurements aimed at identifying crucial resilience factors to inform policy decisions. However, challenges persist, primarily due to varied definitions of resilience, a lack of empirical validation of theoretical models, and insufficient understanding of the underlying mechanisms. This study provides a comprehensive, comparative summary of how previous efforts have defined community disaster resilience and developed indexes to measure it. Additionally, this work illustrates the novel uses of different types of Big Earth Data in measuring disaster resilience with case studies, including data from physical sensors (e.g. remote sensing and street view images), data from social sensors (e.g. social media, volunteer-generated information, and crowdsourced data), and human mobility data (e.g. human movement data captured by GPS signals, connections to Wi-Fi networks, credit card transactions, etc.). Finally, this study outlines the challenges and future directions, including the uncertainty of data-driven resilience analysis due to concerns about the veracity of Big Earth Data, the effects of scale in spatial analysis and modeling of resilience, embracing advanced data fusion and AI methods, and the difficulty in quantifying causal linkages between resilience indicators and outcomes to inform resilience-related decision-making.

The sixth article focuses on developing a web-based interactive 3D emergency response and visualization system for effective decision-making in disaster response and risk reduction (Yang et al., 2023). Due to the three-dimensional nature of disasters and their impacts, offering a 3D perspective is crucial for risk reduction. This study leverages Cesium, a platform for 3D geospatial data, and Vue.js to realistically and dynamically represent disaster scenes in a web interface. Landslides are selected as the disaster focus for developing the system. The 3D system is equipped with diverse functions, including interactive route planning, search and rescue, and the ability to seamlessly incorporate text, image, and video messages into both the 3D web interface and mobile GIS applications. Additionally, this system provides progressive 3D construction and Augmented Reality (AR) visualization capabilities using LiDAR and camera technologies, operable over both local emergency networks and the Internet. Furthermore, the system is equipped with advanced professional landslide management functions, such as landslide susceptibility mapping, continuous landslide monitoring, spatial-temporal contingency plan management, landslide-related information display, overseeing personnel and equipment, and emergency communication. The developed 3D system removes the technical barrier for rescue workers and decision makers in accessing time-sensitive data sources, which significantly improves the efficiency and effectiveness of emergency response operations and aids in better predicting and preparing for potential landslides.

The articles in this special issue substantially contribute to disaster risk reduction. They enhance risk estimation and reduction through precise climate data (Eltazarov et al., 2023), improve disaster mapping to support disaster response using various Big Earth Data (Ghorbanzadeh et al., 2023; Wang & Ding, 2023), promote disaster mitigation and preparedness based on susceptibility mapping with machine learning models (Trinh et al., 2023), outline the potential of measuring and facilitating community resilience via emerging geospatial big data (Qiang et al., 2023), and support effective disaster response, visualization, and mitigation through advanced 3D webGIS platforms (Yang et al., 2023). However, these six articles do not cover the full spectrum of topics under the theme of “Big Earth Data for Disaster Risk Reduction.” The recent advances in geospatial digital twins, geospatial data fusion, geospatial AI (GeoAI), and artificial general intelligence (AGI) bring new opportunities to enhance the potential of Big Earth Data in understanding disaster risks, identifying causal factors, forecasting disaster occurrences, affected areas, and intensity, and designing more effective risk reduction strategies. We anticipate more research and practical deployment combining Big Earth Data with advanced spatial technologies to build a more resilient and sustainable future.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Lei Zou

Lei Zou is an Assistant Professor in the Department of Geography at Texas A&M University, directing the Geospatial Exploration and Resolution (GEAR) Lab (https://www.geoearlab.com/). He received his Ph.D. in Environmental Sciences from Louisiana State University in 2017. His research focus is fostering a healthy, resilient, and sustainable future for human communities amidst changing climates and environments through spatial and responsible thinking, novel big data, AI-supported methods, interdisciplinary collaboration, and community outreach. He spearheads over 20 interdisciplinary projects and has published over 60 articles in academic journals and conference proceedings and as book chapters. Meanwhile, he serves as the vice chair for the International Cartographic Association Commission on Geospatial Analysis and Modeling (ICA-GAM), the vice chair for the Geographic Information Sciences & Systems Group at the American Association of Geographers (AAG-GISS), a board director of the U.S. Cartography and Geographic Information Society (CaGIS), and an editorial board member of the International Journal of Digital Earth.

Fang Chen

Fang Chen is the Deputy Director General and a Professor of International Research Center of Big Data for Sustainable Development Goals (CBAS). He is also serving as the Secretary-General of Integrated Research on Disaster Risk (IRDR) Chinese National Committee, the Executive Deputy Director of the CAS-TWAS Centre of Excellence on Space Technology for Disaster Mitigation (SDIM), and member of World Federation of Engineering Organizations Committee on Disaster Risk Management (WFEO-CDRM). Prof. Chen is the chief scientist of Digital Belt and Road Platform of Big Earth Data Science Engineering Program (CASEarth) of Chinese Academy of Sciences (CAS), which provides the cloud service in support of data sharing and policy-making for sustainable development goals (SDGs). He has ample experience in Big Earth Data for sustainable development, and his work spans the public, private and non-profit sectors. His current work focuses on adapting Big Earth Data technologies to meet the SDGs assessment needs (mainly for SDG 11 and SDG 13) of developing countries including Nepal, Thailand and Small Island Developing States (SIDS). Prof. Chen also conducts interdisciplinary work combining, remote sensing, ecology, and other fields of study to assess spatial patterns of disaster risk. He has published over 100 academic papers and book chapters and was elected to the CAS “Hundred Talent Program” in 2011 and the TWAS Young Affiliate Fellow in 2014.

Xiao Huang

Xiao Huang is an Assistant Professor in the Department of Environmental Sciences at Emory University. His research primarily focuses on human-environment interaction, computational social sciences, vulnerability and resilience, urban informatics, disaster mapping and mitigation, GeoAI and deep learning, and disaster remote sensing. Different from traditional geospatial analysis, his research takes advantage of rapidly growing data availability through the utilization of emerging, innovative data sources and the development of advanced geospatial analytical techniques to address existing/future challenges in disaster studies (e.g., real-time monitoring, assessment, and mitigation) and human society (e.g., human behaviors, vulnerability, and inequity) from Big Data and spatiotemporal perspectives. He built an academic career integrating his research and wide‐reaching public service while engaging with communities, particularly socially disadvantaged communities.

Bandana Kar

Bandana Kar is an AAAS Science, Technology and Policy Fellow for building decarbonization in the Building Technologies Office. Her primary research is in energy and urban/community resilience from the perspective of science policy and security. Using geospatial and computational sciences, she develops and deploys risk-informed and process-oriented algorithms and tools that account for human dynamics and extreme events for decision and policy making. Her research has been funded by NSF, DHS, DOE and NASA. She was a recipient of the 2019 Emerging Scholar Award from the American Association of Geographers and was a fellow of the 2009 NSF’s Enabling the Next Generation of Hazards and Disasters Researchers Fellowship Program. She has published more than 50 articles, book chapters and proceeding papers in geospatial, remote sensing, computational science journals. She coedited the book “Risk Communication and Community Resilience,” and she is the vice-president of the American Society for Photogrammetry and Remote Sensing, a co-organizer and co-founder of the ACM SIGSPATIAL Workshop - Advances in Resilient and Intelligent Cities (ARIC), and a member of ASCE’s Risk and Resilience Measurement Committee.