When you say the terms ‘earth hazards’ or ‘geological hazards’, you may assume that the best way to gather data and assess these would be through studies on the ground. This has, of course, been the norm for much of the time scientists have been studying the world beneath our feet, but in the past few decades there has been a significant shift in how geoscientists view and assess natural hazards.
The use of ground-based observations alone come with a number of limitations, including accessibility, the speed at which large areas can be covered, and inconsistencies between different surveys. For large-scale features such as faults, or remote locations such as some volcanoes, this means it becomes difficult to fully and accurately assess the hazard posed. Satellite imagery can cover features such as faults, volcanoes and landslides on a global scale, allowing for a more complete picture than ground-based observations, and providing access to even the most remote or inhospitable locations. The imagery is independent of ground-based monitoring methods such as tiltmeters, and by comparing multiple images through time using methods such as InSAR (Interferometric Satellite Aperture Radar) ground movement and deformation over a range of timescales can be observed. Comparison against ground-based observations can help to improve accuracy and account for instrumentation errors, and the combination of both ground and satellite monitoring can provide a more complete picture of features such as fault systems and magma complexes.
One of the most exciting aspects of using satellite imagery for geohazard analysis is that there have been continuous improvements recently in the frequency, type, and availability. The launch of satellites such as Sentinel-1 mean it is becoming increasingly feasible to routinely study volcanic and seismic hazards in remote and inaccessible regions, however this comes with its own set of challenges. The sheer amount of data produced by Sentinel-1 is too large to be manually analysed on a global scale, so work has been done to use machine learning algorithms and convolutional neutral networks (CNN) to automatically detect volcanic (e.g. Anantrasirichai et al. 2018, 2019) and co-seismic (e.g. Brengman & Barnhart 2021) ground deformation and differentiate it from atmospheric noise.
Detection of landslides is another hazard being examined using a combination of satellite imagery and machine learning, trying to reduce resources such as expert knowledge, supervision and fieldwork needed for traditional mapping methods. Work by Ghorbanzadeh et al. (2019) uses optical data from the ‘Rapid Eye’ satellite to analyse the potential of a number of machine learning methods such as convolutional neural networks (CNNs) for landslide detection, and although the paper concludes that this method is still in its infancy, it shows a promising start for the use of deep learning in this field. More recently, the 2022 Landslide4Sense (L4S) competition organized by the Institute of Advanced Research in Artificial Intelligence (IARAI) examines how landslides can be automatically detected using large-scale multiple source satellite imagery collected on a global scale, as well as the best performing deep learning algorithms. The resulting article by Ghorbanzadeh et al. (2022) aims to foster interdisciplinary research in the field, inviting researchers to submit more results and evaluate the accuracy of methods used in the hope of improving the landslide detection results reported.
The use of satellites for monitoring geohazards has had an effect on the insurance industry as well as academia, with an ever-increasing catalogue of data allowing for probabilistic analysis. An example of this can be seen in the paper by Biggs et al. (2014), which examines the links between volcanic deformation and eruption using a catalogue of 540 volcanoes monitored over a period of 3 years. This work went on to win a Lloyd’s of London prize, and was part of a larger project by the University of Bristol’s Volcanology Group and the British Geological Survey. These organisations co-developed the Global Volcano Model (GVM) network in 2011, and found that at least 80% of the world’s volcanoes lacked ground-based monitoring. The University of Bristol’s work on regional-global surveys, atmospheric corrections and machine learning identified active deformation at more than 25 volcanoes previously considered inactive, and this work contributed to the UN Office for Disaster Risk Reduction publishing their 2015 Global Assessment of Risk report (GAR15) with volcanic risk considered for the first time.
Work by the University of Bristol and GVM was featured in our 2016 Annual Review. This work on the relation of global databases on volcanic hazards and their impacts, together with volcanic risk profiles of all countries with active volcanoes and the risk metrics developed for GAR15, has been described as “key data requirements to build models for financial decision making for the insurance industry”. Bristol’s work has significantly advanced the potential for volcanic eruption risk modelling and supported the accurate pricing of risk-based insurance policies.
Further improvements to the range of satellite data available, and the methods for detection and processing of images, means that documents such as GAR15 could be revised to provide more accurate probabilistic hazard analysis for the parametric insurance and reinsurance industries. Further work in this area is a key goal of the WTW Research Network for 2023, following on from previous partnerships with the University of Bristol and GVM.
In addition to volcanoes, earthquakes and landslides, satellite imagery, InSAR and machine learning have been used to improve understanding of earth hazards and their effect on the built environment.
Work has been done to measure and reconstruct ground displacement from coal mining in China (Chen et al. 2020), by combining the deformation rates derived from both TerraSAR-X and Sentinel-1 satellites and using a small baseline subset (SBAS-InSAR) algorithm. Chen et al.’s paper highlights limitations in traditional geodetic surveying methods such as precision levelling, total station measurement and GNSS, as well as conventional multitemporal (MTInSAR) techniques. In the UK, research has been to look at subsidence and uplift associated with coal mining, ground water withdrawal, landslides, and tunnelling engineering works (Anantrasirichai et al. 2020).
This proved to be a more challenging application of deep learning networks than for hazards and locations previously discussed, due to the sparsity of measurement points, presence of noise, slower deformation signals and insufficient ground truth data for constructing a balanced training data set. But the work presented in Anantrasirichai et al.’s paper proposes improvements and can detect both subsidence and uplift from anthropogenic activities, demonstrating the potential applicability of their proposed framework for development of automated ground motion analysis.
Earth Observation isn’t only being used to advance hazard analysis, but has enabled better understanding of urban development and exposure. Work done by the Global Earthquake Model foundation (GEM) has created a framework for forecasting the spatial distribution of population and development of residential buildings in Costa Rica using historical satellite imagery, in order to forecast exposure to seismic risk (Calderon & Silva 2021). Similar work analysing urban growth modelling and vulnerability assessment for the Kathmandu Valley in Nepal has been done in conjunction with academics at UCL (Mesta et al. 2022), and this article states that substantial advances in remote sensing technologies and spatial modelling have enabled the creation of worldwide spatial datasets on human settlements that can be used in natural-hazard risk modelling.
This broad range of research undertaken within the last decade highlights that the quality and availability of Earth Observation data is continually improving, and together with increased computational power and the development of machine learning algorithms to aid in processing and detection, is aiding in more complete and accurate assessments of hazard, risk and exposure invaluable to the insurance industry. By continuing to support work in this field through academic partnerships, it is hoped that we can utilise data from new satellites and develop reliable workflows for detecting, quantifying and even forecasting geohazards, as well as the exposure and risk to developing urban areas.
Anantrasirichai, N., Biggs, J., Albino, F., Hill, P., & Bull, D. (2018). Application of machine learning to classification of volcanic deformation in routinely generated InSAR data. Journal of Geophysical Research: Solid Earth, 123(8), 6592-6606.
Anantrasirichai, N., Biggs, J., Albino, F., & Bull, D. (2019). A deep learning approach to detecting volcano deformation from satellite imagery using synthetic datasets. Remote Sensing of Environment, 230, 111179.
Anantrasirichai, N., Biggs, J., Kelevitz, K., Sadeghi, Z., Wright, T., Thompson, J., ... & Bull, D. (2020). Detecting ground deformation in the built environment using sparse satellite InSAR data with a convolutional neural network. IEEE Transactions on Geoscience and Remote Sensing, 59(4), 2940-2950.
Biggs, J., Ebmeier, S. K., Aspinall, W. P., Lu, Z., Pritchard, M. E., Sparks, R. S. J., & Mather, T. A. (2014). Global link between deformation and volcanic eruption quantified by satellite imagery. Nature communications, 5(1), 1-7.
Brengman, C. M., & Barnhart, W. D. (2021). Identification of surface deformation in InSAR using machine learning. Geochemistry, Geophysics, Geosystems, 22(3), e2020GC009204.
Calderón, A., & Silva, V. (2021). Exposure forecasting for seismic risk estimation: Application to Costa Rica. Earthquake Spectra, 37(3), 1806-1826.
Chen, Y., Tong, Y., & Tan, K. (2020). Coal mining deformation monitoring using SBAS-InSAR and offset tracking: A case study of Yu County, China. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 13, 6077-6087.
Ghorbanzadeh, O., Blaschke, T., Gholamnia, K., Meena, S. R., Tiede, D., & Aryal, J. (2019). Evaluation of different machine learning methods and deep-learning convolutional neural networks for landslide detection. Remote Sensing, 11(2), 196.
Ghorbanzadeh, O., Xu, Y., Zhao, H., Wang, J., Zhong, Y., Zhao, D., ... & Ghamisi, P. (2022). The Outcome of the 2022 Landslide4Sense Competition: Advanced Landslide Detection from Multi-Source Satellite Imagery. arXiv preprint arXiv:2209.02556.
Mesta, C., Cremen, G., & Galasso, C. (2022). Urban growth modelling and social vulnerability assessment for a hazardous Kathmandu Valley. Scientific reports, 12(1), 1-16.
