Our technical working group shares their feedback with One Concern’s modeling teams each year through letters. Below, you’ll find the two complete letters provided to our team from our Technical Working Group members, reflecting on our Flood and Seismic Modeling.
Flood Group Letter
We are pleased to summarize our thoughts and great impressions based on our review of One Concern’s model to estimate flood hazard throughout Japan. Our comments are based primarily on our meeting with you and your colleagues on June 15 and 22, 2021, where we heard presentations and had discussions about One Concern’s flood modeling pipeline.
One Concern’s flood modeling pipeline is a sophisticated and seamless serial combination of widely covered multiple state-of-the-art and well-accepted models. Coastal storm surge is simulated with SCHISM, which is well regarded throughout the hurricane and tsunami modeling communities. Rainfall-runoff and river routing is simulated with LISFLOOD, which is a fast-running workhorse model of the hydrology community. Direct urban ponding of rainfall is simulated with the NRCS curve number method, which is similarly a widely accepted method in the stormwater community. The LISFLOOD riverine model generates discharge boundary conditions for the upstream boundaries of the SCHISM model, while the urban NRCS method specifies direct rainfall ponding rates distributed across the SCHISM inundation model mesh. The flexibility of the SCHISM inundation model to apply high resolution to specific areas of interest, together with the rapid run speed of the LISFLOOD riverine and NRCS urban models, plus the seamless integration of the model data streams, has allowed One Concern to apply their flood model pipeline to the entire country of Japan. Such a broad geographic scope combined with building-level accuracy in areas of interest is an impressive achievement at the cutting edge of the integrated river and coastal flood hazard simulation.
The One Concern team made a strong effort to validate the full model chain, with very good results. The riverine model is driven with radar rainfall data and validated against stream gauge data throughout the country, with good agreement between simulated and measured peak flood flows. The coastal model is driven by tidal harmonics, plus hindcast pressure and wind fields for historical typhoon events. This model produces storm tides in good agreement with measurements by tide gauges along Japan’s coasts. The Urban model is driven by radar rainfall data falling directly on cities. The combined inundation model, based on the coastal model and using the riverine and urban models as inputs, is validated by comparing hindcast inundation depth maps with measured values during extreme historical events, again with impressive accuracy. In addition to the flood model pipeline itself, the team introduced a machine learning method to generate automatic diagnostic levee data along the many rivers where actual data was missing and validated this approach against data from cities with data that were held out of the training data set. This is a cutting-edge approach for dealing with data scarcity that has the potential to drastically improve model accuracy compared to the standard method of ignoring levees when data is not available.
Due to the flood model pipeline’s complexity and broad scope, further improvement of some specific aspects of each model can make the overall simulation even more accurate. Currently, each model in the pipeline has some assumptions that are likely to generate conservative (too large a flood extent in urban areas) results. The One Concern has explicitly stated these assumptions and is investigating ways to increase the accuracy and robustness of each of its models. However, the current assumptions are common in state-of-the-art models in the inundation modeling world, and a model pipeline such as One Concern’s that produces conservative results is better than one which produces non-conservative results. Nonetheless, the team is working on quantifying the effect of these assumptions, and this is the best that can be expected of any predictive model that exists.
Specific points that the One Concern team are currently working to address are:
- Underestimation of levee crest height due to the use of elevation to represent levees in the inundation model.
- Underestimation of floodplain conveyance and storage capacity due to the rectangular shape of river channels in the riverine model.
- Overestimation of peak flood flows due to the kinematic wave routing in the riverine model.
- Errors in baseflow due to neglect of snowmelt in the riverine model and dam operation.
- Overestimation of storm surge due to the use of increased wind speeds to account for breaking-wave-radiation-stress-induced water level setup in the coastal model.
- Overestimation of rainfall accumulation due to the neglect of the storage and conveyance capacities of stormwater drains and deep tunnels, in the urban model.
- Overestimation of ponding due to the neglect of pumping stations in the inundation model.
- Overestimation of flood extent due to the neglect of land-use-dependent surface roughness in the inundation model.
Overall, we are impressed with One Concern’s simulation capability, including the model pipeline’s accuracy, run speed, broad geographic scope, and high resolution. We do not know of other models or sets of models that can be applied so robustly and generally to the full range of coastal, pluvial, and fluvial floods possible. This makes One Concern’s developments unique and particularly useful.
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 to presenting information and responding to questions during our review.
Jeremy D. Bricker, PhD, PE
Associate Professor, University of Michigan and Delft University of Technology firstname.lastname@example.org
Research Professor, Center for Coastal Resource Management, Virginia Institute of Marine Science
Professor, Disaster Prevention Research Institute, Kyoto University
Seismic and Infrastructure Group Letter
We are pleased to summarize our thoughts and impressions based on our review of One Concern’s latest resilience-modeling research pertaining to earthquake-damage estimation and recovery times for critical infrastructure such as buildings, ports, airports, roads, and bridges. Our comments are based on materials presented in advance and discussed in detail with you and your colleagues in meetings on June 14, 17, 29, 30. Details of these discussions are nicely summarized in the reports you prepared following each meeting with suggestions ranked according to importance by consensus of the advisors. Consequently, we will focus here on only what we consider some of the high points of the review.
Rapid population growth with increases in urbanization and infrastructure contribute to ever increasing losses from earthquakes and other natural disasters, such as flooding and forest fires, that are further accelerated by climate change. As a result, the need to develop models to facilitate recovery and resilience of potentially affected communities is becoming increasingly important.
As we all know, development of earthquake damage and recovery-time models is extremely challenging, especially for buildings and other infrastructure, due to limited data for calibration and validation of the models. Therefore, such models are traditionally developed based on a combination of theory and judgment, informed by available data from past earthquakes and detailed analyses of various structures’ performance. In contrast to other software for regional damage and loss analyses, such as HAZUS, One Concern’s cloud based platform offers a promising new approach that leverages statistical machine learning and artificial intelligence (ML/AI) techniques to integrate observed and/or simulated data to characterize earthquake ground shaking, detailed inventories, structure performance, and recovery times. T
This approach permits the latest advances in hazard and damage estimation to be augmented with rapidly developing improvements in ML/AI techniques as developed across a broad range of fields. While the ML/AI techniques have great potential, it is important to carefully train and validate the models so as to understand and communicate to users their capabilities and limitations.
Validation and calibration efforts of the One Concern model for earthquake damage as reviewed and documented in the TWG Review letter of September 17, 2020 were found to be in reasonably good agreement with observed data from the Northridge and Kumamoto earthquakes and those calculated for a potential future earthquake on the Hayward fault. Since that time One Concern has incorporated many of the recommendations of the 2020 Advisory Group. It has continued to expand, validate, calibrate, and train the ML model using observed data from the 1995 Kobe earthquake and simulations based on the updated
Global Earthquake Model fragility functions, together with recent ground motion maps based on Japan’s national state-of-the-art Kik-Net and K-Net seismic-observation networks. In addition, the model has been expanded to incorporate exposure data developed using state-of-the-art imputation procedures for situations with limited data. These commendable efforts have led to more robust models with potential applications to other areas of Japan and the US.
Some of our key observations and recommendations regarding One Concern’s latest developments for the ML/AI damage estimation model in Japan are:
1) In situations with limited data on building inventories, we encourage One Concern to leverage information from local municipalities and expert opinion as much as feasible to augment and validate data developed using data imputation procedures. We also suggest exploring the use of additional data sets, developed through emerging technologies (e.g., Microsoft AI-assisted tools for building footprints, LIDAR scans, etc.) to determine or confirm building parameters as story height and occupancy.
2) We encourage One Concern to review the redundancies in how Vs30 is considered in accounting for variations in soil conditions when estimating ground-motions and building damage. To the extent that Vs30 is considered in ground motion prediction equations, one should be careful to avoid any double counting of its effect in the damage estimates.
3) We support and encourage One Concern’s efforts to review and, where necessary, adjust GEM fragility functions for their applicability to observed Kobe damage, especially those for one- and two-story wooden structures. This is important both to improve the accuracy of the damage estimates and to demonstrate the added value of One Concern’s ML/AI model for validation with earthquake damage data.
4) We further encourage One Concern to continue efforts to train and validate the ML/AI model using observed building damage data from the Tohoku and Hokkaido earthquakes, which can provide new insights based on the building inventories in the affected areas and the earthquake characteristics. In particular, the long duration shaking from the large magnitude Tohoku earthquake offers opportunities to incorporate ground motion duration in models for structural damage and ground movement.
With regards to One Concern’s efforts to develop and validate the model for the building inventory and earthquake hazard in the United States, we appreciate the development of ML/AI techniques to develop a high-resolution inventory by combining and imputing data from a variety of sources (e.g., Corelogic, Estated and others). We also see important value in continuing efforts to train and evaluate One Concern’s inventory, damage, and loss components of the ML/AI model using inventory data from the Boston area and damage/loss data from the Los Angeles area. The following are a few observations and suggestions regarding the United States model development:
1) The validation results of One Concern’s building inventory model using the Corelogic data for the Boston area are encouraging, but we would encourage validation against other available ground truth data (i.e., other independently developed data sets or semi-automated spot checking of some results). We would also encourage expanding the validation to other regions of the United States where earthquake risks are a concern.
2) Given the sparsity of well-documented earthquake damage, loss and recovery-time data in the United States, we are supportive of One Concern’s approach of conducting earthquake scenario studies, results of which can be compared to other published studies (e.g., earthquake scenario studies by USGS or other agencies), and sensitivity studies, which can be evaluated by expert opinion. We encourage One Concern to continue to conduct and document these studies to build further confidence in the ML/AI model.
3) Given the increasing recognition of recovery-time as a critical measure of indirect losses and community resilience, we appreciate One Concern’s efforts to incorporate recovery time (in addition to damage and direct economic loss, i.e., repair costs) into their earthquake model. To the extent that recovery modeling has not received as much attention by the research community, One Concern’s modeling platform can provide an important resource to test and evaluate alternative building recovery models, including ones that scale with the earthquake magnitude and include resource availability and lifeline systems (e.g., water and electric power) in the recovery assessment. We applaud and strongly encourage One Concern’s continued efforts in these directions.
We encourage One Concern to continue to collect, evaluate, and incorporate legacy data on damage and losses from past earthquakes and to continue to engage with other groups involved with collecting data from future earthquakes (e.g., the Earthquake Engineering Research Institute, the Structural Extreme Event Reconnaissance network, and others). We encourage continued development of comparative methods to document the contributions of the ML/AI technologies. We also encourage One Concern to continue to make use of synthetic data from detailed performance-based earthquake engineering analyses (e.g., FEMA P-58 analyses), new national and global data sets, and advancements in ML/AI technologies.
Advancements in community resilience to major disasters starts with developing a better understanding of the recovery time for damaged communities to return to an acceptable state of operation. Recovery of an acceptable level of operation as defined by a majority of people includes recovery of infrastructure, such as power distribution and transportation facilities, including seaports, airports, roads, and bridges. Recovery also depends on interdependencies between systems, the impacts of which can differ for different organizations (e.g., impact on a specific business enterprise versus the community in general). In this regard, One Concern’s efforts in developing recovery-time models for power distribution and each of the transportation facilities are commendable.
Our key observations regarding the infrastructure recovery-time models and their application for resilience planning include:
1) With regards to the electric power distribution model, as validated using Northridge data, we encourage One Concern to continue considering sensitivity analyses for recovery of the power network to a range of earthquake magnitudes that will affect areas of significantly different sizes. Simulated scenarios could be helpful, such as the USGS Haywired study and those utilized at the NSF NHERI SimCenter or NIST Center for Risk-Based Community Resilience Planning.
2) We encourage One Concern to proceed with their plans to incorporate the influence of geotechnical conditions on damage and recovery of infrastructure systems, especially roads, airports and ports.
3) We commend One Concern’s resilience-planning efforts to address recovery of a broad range of infrastructure components and their interdependencies. In this regard, we suggest that One Concern may want to consider expanding the recovery models for container ports to include liquid-storage and liquid-transfer ports, which are critical for recovery of disrupted oil and gas supply chains for transportation and electric power plants. We encourage One Concern to continue to develop state of-the-art models for resilience planning that define, adopt, and communicate clear metrics for characterization of “performance functionality” for various infrastructure systems.
4) We support One Concern’s approach to incorporate planning horizons into the calculated earthquake-loss and infrastructure-recovery metrics. However, the current implementation based on earthquake hazard maps is limited insofar as it does not reflect spatial correlation of ground motions across an affected region. Therefore, to develop event-specific damage and recovery-time estimates, we encourage One Concern to continue to develop model predictions based on scenario earthquake ground motions. We also encourage studies to compare loss and recovery metrics derived from the different approaches to simulating the various components of the earthquake risk.
In summary, we commend One Concern for their significant progress on validation and calibration of the ML/AI model for earthquake damage estimation. In addition, their efforts to develop recovery-time models for a wide diversity of infrastructure components is a significant accomplishment needed for resilience planning. The advances in the ML/AI and recovery-time models continue to differentiate One Concern’s products and further establish their leadership in this domain.
We are pleased to see One Concern’s effort to engage more with the risk analysis research community in natural hazards engineering, including presenting some aspects of its work at national conferences and publishing in scholarly journals, which we believe would lead to improvements in One Concern’s models and credibility among experts in the field.
Finally, beyond efforts that we reviewed to develop and validate One Concern’s models, we think it is important to demonstrate the unique capabilities of One Concern’s technology for resilience planning that is not available in other damage and loss assessment methods. Certainly, the cloud-based computing and information technologies offer computing speed, immediate access, and high-resolution output that are exemplary. Aside from convenience, these capabilities and the new recovery-time models offer an interactive decision support system for investments in disaster risk mitigation and resilience planning on portfolio, community, and national scales.
We appreciate this opportunity to review and provide input on One Concern’s latest resilience modeling developments, particularly as they pertain to earthquake-damage estimation and associated recovery times for critical infrastructure. We also acknowledge the cooperation and responsiveness of the One Concern Resilience Research team to presenting well organized information in advance of the meetings and responding to questions during our review.
Gregory G. Deierlein, Ph.D., NAE
Director, John A. Blume Earthquake Engineering Center, Stanford University
Roger D. Borcherdt, Ph.D.
Scientist Emeritus, Earthquake Science Center, United States Geological Survey
Jamie E. Padgett, PhD
Professor, Department of Civil and Environmental Engineering, Rice University