As part of our continuous journey to refine our resilience modeling capabilities, One Concern regularly meets with our Technical Working Group to review our latest research and models. Comprised of independent leaders in the fields of disaster science, data science, engineering, and hydrology, the Technical Working Group helps to guide One Concern as we develop our hazard and climate resilience modeling capabilities by providing insights and feedback informed by the most up-to-date knowledge from peer-reviewed academic research.
2022 TWG Review of One Concern Flood Hazard Models
We are pleased to summarize our thoughts and great impressions based on our review of One Concern’s model to estimate flood hazards throughout Japan. Our comments are based primarily on our three-time meetings with One Concern’s flood team on November 7, 16, and 29, 2022, where we heard presentations and had discussions about One Concern’s flood modeling pipeline.
Building on the flood modeling pipeline that we reviewed in 2021, One Concern has improved its simulation methodology by addressing the comments brought up by the Technical Working Group last year. Most of the comments brought up in 2021 focused on the tendency of the One Concern models to be implemented in a conservative fashion, causing an overestimation of flood hazards. However, it is acknowledged that overestimation of flood hazards in an early warning system can lead to false positives (false alarms), which themselves can lead to reduced public confidence in the early warning system. Therefore, One Concern applied a mix of machine learning and calibration to available data to make the models more accurate.
Addressing topics brought up last year, they have innovatively applied machine learning to increase accuracy and reduce the degree of conservatism in model results. This is being done to address the issues of unknown river channel cross-sectional shape and depth, as well as the tendency of the kinematic wave hydrological routing method to overestimate flood peaks. The application of ML in this way is truly novel in the field of hydraulic engineering. Despite the typical tendency of kinematic wave to overestimate flood peaks, calibration to measurements shows that the river model used tends to underestimate observed, historical flood peaks. This will be addressed via further calibration of river channel cross-sectional geometry and roughness to this data.
The second topic was addressed by recalibrating the flood model train to an expanded dataset of measurements. This was done to determine snowmelt-induced baseflow, and infiltration rates to represent stormwater system storage capacity and land-use-dependent equivalent roughness values. One data type of pump station operating rules were not available, so the model remains conservative in this aspect. However, local clients could provide this data in order to make the product more useful to them.
For the river flood model, however, some conservative assumptions remain, such as the need to represent dikes via topography, due to the inability of the inundation model to implement thin barriers. The flood team is working on this issue by considering higher resolution simulations (which are possible due to the implicit discretization scheme that allows large time steps), as well as other possibilities such as GPU processing and sub-grid scale topography. ML was furthermore used to generate synthetic river dikes where dike height and location data were unavailable, though it could not yet be determined whether this would always lead to a conservative result. Like river dikes, the inundation model is also unable to resolve small (<30m wide) channels that could be responsible for local flooding, so the resolution-enhancing methods described here could also be used to address this issue in the future. Since the 30m is close to the DEM resolution, an alternative approach like ML may also be explored. In general, though the use of ML to fill in the gaps for missing data is innovative and novel, it is still difficult to certify whether conditions could occur in which these methods would generate a non-conservative result. This could be the subject of future sensitivity analyses.
Additionally, a new model of future changes in heat/cold waves due to climate change can be evaluated as an additional step toward climate services by One Concern. It is important to clarify with the customer of the projection data and to provide unique projections that are not available from IPCC, etc. We recommend considering using not only GCMs but also downscaled projections in the US, such as CORDEX.
Overall, we are still impressed with One Concern’s simulation, modeling, and analysis capability, as well as their novel introduction of ML methods to make up for missing data and to improve model results. We appreciate this opportunity to review and provide input on One Concern’s flood model pipeline. We also acknowledge the cooperation and responsiveness of your team in presenting information and responding to questions during our review.
2022 TWG Review of One Concern Seismic and Infrastructure Models
As members of the Technical Working Group of One Concern, we are pleased to summarize our thoughts and recommendations based upon our review of One Concern’s resilience-modeling work pertaining to seismic damage and loss estimation models, as well as recovery modeling for critical infrastructure such as buildings, seaports, power networks, roads, and bridges. Our comments are based on materials presented by members of the One Concern Resilience Modeling Team (1C Team) in advance and discussed in detail with the 1C Team in meetings on November 7, 8, 10, and December 1, 2022. Details of our comments, meeting discussions, and the written response of the 1C Team to our questions were summarized well in documents, slides, and subsequent discussions during the meeting on December 1. Consequently, we will focus here only on what we consider to be a summary of some of the key points of the review.
The pursuits of One Concern are timely and essential. Natural hazards and the effects of climate change, population growth, and urbanization, along with infrastructure aging all contribute to the increasing risks of damage to the built environment and cascading consequences in communities. The ability to predict damage and losses, along with downtime and recovery of buildings and infrastructure systems, is crucial to support a range of decisions related to resilience enhancement.
We commend One Concern on the advances in both their Ready and Domino/DNA products and the outstanding 1C Team of researchers and modelers that One Concern has assembled. Importantly, the 1C Team reflects not only state-of-the-art knowledge in artificial intelligence and machine learning (AI/ML), but also expertise in domain sciences related to hazard analysis, engineering performance assessment of buildings and infrastructure, and catastrophe modeling that is critical for resilience modeling. The 1C Team is technically strong, engaged with the research community, and aware of the latest developments in the field.
Traditional approaches to risk analysis, such as those employed in HAZUS and similar regional risk assessment software, rely on models developed based on a combination of theory and judgment, and these models generally remain unchanged or change slowly as new data emerges from recent events. In contrast, One Concern’s cloud-based platform offers a promising new approach that leverages AI/ML techniques to integrate observed and/or simulated data to characterize earthquake ground shaking, detailed asset inventories, structure performance, and recovery times. This approach permits the latest advances in hazard and damage estimation to be augmented through rapidly developing improvements in AI/ML techniques. While we are all aware that while AI/ML techniques have great potential, it is important to carefully train and validate the models so as to understand and communicate to users the capabilities and limitations of the models. One Concern seems to appreciate these points, as evidenced by its continuing efforts to employ AI/ML carefully in the context of model calibration and improvement, along with inventory development, taking steps toward explainable AI/ML.
Since our last review, documented in our letter of September 15, 2021, One Concern has continued to enhance its scientific and engineering team with several additional Ph.D. and career-experienced members. This enhanced team has allowed model developments to proceed rapidly as they pertain to various hazards posed by earthquakes, windstorms, and floods as well as their effects on man-made structures and associated infrastructure. These developments include recommendations documented in our 2021 review letter, as well as additional notable advancements, including:
1) Development and review of resilience statistics for One Concern’s Seismic, Windstorm, and Flood models in the US and Japan for buildings, electrical-power networks, highways, bridges, seaports, and integrated lifelines,
2) Inclusion of climate scenarios with baseline updates for the years 2035 and 2050,
3) Addition of new probabilistic resilience metrics (damage, losses, downtime, and recovery curves),
4) Significant efforts to further validate and train AI/ML models based on several additional historic events (e.g., four additional seismic events used in the Japan Ready product), including validation at multiple geographic scales,
5) Significant updates in fragility functions based on recent extensive expansions in associated data sets,
6) Further development of platforms to facilitate evacuation and resource allocation in case of damaging events,
7) Extensive augmentation and validation of inventory databases, loss and downtime analyses, and review of state-of-the-art platforms for model calculations, such as for catastrophic risk simulation,
8) Calibration and validation of the models using recent data from Hurricane Ian (9/23/22- 10/2/22), which severely impacted Florida and the east coast of the US.
We appreciate One Concern’s efforts to document its modeling methodology and its validation, allowing for internal review and publication in scientific and engineering research conference proceedings and journals (e.g., One Concern, 2021; Pant et al. 2022, and Chhabra et al. 2022).
Such publications provide a basis for in-depth reviews of the methodology and validation procedures by appropriate peer reviewers. They help establish credibility and adoption of One Concern’s models by clients and the professional scientific and engineering research communities. Moreover, as One Concern’s software tools and supporting databases grow in breadth and sophistication, comprehensive documentation of the models and model validation is critical to both facilitate development and ensure the reliability of the software products. In addition to documenting the models, it is important to establish sustainable processes for rigorous internal review, which depending on the topic, may benefit from the selective involvement of external experts. In these regards, we strongly encourage One Concern to continue to (1) document and publish model developments and their validation in a timely manner as appropriate, (2) establish regular and well-documented internal review processes that include external experts as deemed appropriate, and (3)continue to participate, interact, and contribute papers to the scientific and professional community concerned with damage, loss, and recovery estimation from natural hazards.
Since our 2021 review, One Concern has made considerable progress on models to evaluate recovery time, which is essential to losses from interruption of infrastructure services, displacement of populations, and resilience planning. Based on the information presented during our review, the recovery models, which relate physical damage to recovery/restoration times, are largely based on empirical data from past events, e.g., associating the calculated damage to observed recovery times for electric power or post-event occupancy of damaged buildings. This approach follows accepted practice in recovery modeling, owing to the lack of detailed knowledge and techniques to directly simulate the underlying phenomena that affect recovery. Nevertheless, to the extent that recovery modeling is central to One Concern’s Domino product, we would encourage One Concern to push the technology beyond empirical models to ones that integrate state-of-the-art knowledge into simulation-based models of recovery, e.g., using agent-based or discrete choice models of operations and resources needed for recovery. Related to this, it may be necessary to acquire additional expertise in areas related to economics and financing of recovery, logistics and supply chains, and social science (organizational and human behavior). Additionally, the recovery models currently available for different infrastructure vary widely in sophistication and maturity. One Concern has an opportunity to expand its recovery models to more fully address the diverse assets and interdependent operations of the range of infrastructure systems covered in its products.
We further encourage continued efforts to articulate the unique capabilities of One Concern’s technology for resilience planning that is not available in other damage and loss assessment methods. One Concern’s e-book on “Resilience and Business Downtime” is a good start towards explaining the benefits and offering examples of the benefits offered through high-resolution (parcel-level) modeling of assets. The examples in this book could be extended to go beyond simple measures of downtime to evaluate the specific impacts on businesses and ways to mitigate these effects. Further, cloud-based computing and information access technologies offer computing speed and near-real-time data access and model updating (through AI/ML) which are hallmarks of One Concern’s approach. These advancements can provide a unique and interactive decision support system for investments in disaster risk mitigation and resilience planning on portfolio, community, and national scales.
To summarize, we commend One Concern for their notable advancements in the development of resilience models pertaining to seismic damage, loss estimation, and recovery for critical infrastructure such as buildings, seaports, power networks, roads, and bridges. We are encouraged by One Concern’s progress in the documentation and publication of the methodologies and validation procedures for various models, along with engagements with the risk analysis research community in natural hazards engineering. Finally, we continue to stress the importance of transparency in communicating the capabilities and limitations of models to clients and other stakeholders.
We appreciate this opportunity to review and provide recommendations regarding the latest developments in One Concern’s resilience-modeling efforts. We would like to acknowledge the cooperation and efforts of the One Concern Resilience Research team who have provided well-organized information in advance of our review meetings and provided detailed responses in subsequent meetings to questions we posed during our review.
Gregory G. Deierlein, Ph.D., NAE
Director, John A. Blume Earthquake Engineering Center, Stanford University
Roger D. Borcherdt, Ph.D.
Research Scientist Emeritus, Earthquake Science Center, United States Geological Survey
Jamie E. Padgett, PhD
Professor, Department of Civil and Environmental Engineering, Rice University