Urban Flooding Analytics Team – Leveraging Watson AI to create crowd-sourced flood maps for New York City.
Click here to watch the video of the Winning Team presenting their Final Presentation.
Case Title: Providing Real-Time Flood Data to Improve Climate Change Resiliency Efforts
Link to 1-minute Video: https://www.youtube.com/watch?v=E8YqO6JrvHo
Sector: City Services
Nathaniel Zinda, Junior, Economics, Baruch College
Augustus Kaptko, Senior, Media Technology, New York City College of Technology
Sett Hein, Sophomore, Computer and Information Sciences, Baruch College
The negative effects of climate change are rapidly intensifying. Chronic hazards, such as sea level rise and precipitation, present an increasing risk to our city’s neighborhoods. According to the New York City Panel on Climate Change (2015), sea levels are expected to rise anywhere between 11 inches to 21 inches by the year 2050, and up to 50 inches by 2100. Sea level rise for New York City is nearly twice the observed global rate. Precipitation events are likewise projected to increase. By 2050, the panel estimates that normal events of precipitation will increase from 11% to 13%, and intense events of precipitation will increase by 25% (NYCPCC, 2015). Needless to say, as the risks from these hazards continue to increase, so will the frequency and intensity of severe flooding events that threaten the economic and social well-being of New York City’s communities and people. The City predicts that a Sandy-like event in 2050 could result in $90 billion in economic losses – compared to the $19 billion from Sandy itself – because of sea level rise alone (OneNYC, 2015).
The Mayor’s Office of Recovery & Resiliency (ORR) is tasked with leading the City’s $20 billion OneNYC climate change resiliency plan. In pursuit of this, the ORR relies heavily on the Federal Emergency Management Agency’s (FEMA) 2007 Effective Flood Insurance Rate Map (FIRM) and the 2015 Preliminary Flood Insurance Rate Map (PFIRM) to assess the flood risk across the city and determine the OneNYC initiatives for City agencies to implement. Unfortunately, FEMA’s flood maps are notoriously inadequate.
The 2007 Effective FIRM grossly miscalculated New York City’s flood risk in 2012. When Hurricane Sandy stuck the east coast, only 47 percent of the flooded area in Brooklyn and 54% in Queens was predicted by the 2007 FIRM (ProPublica, 2013). When the 2015 PFIRM was released, the ORR immediately filed an appeal, citing numerous modelling errors. The appeal was won, and FEMA once again agreed to revise the City’s flood maps. Perhaps more disconcertingly, the Department of Homeland Security’s IG Office recently published a report accusing FEMA’s flood map program of being inundated by mismanagement and poor mapping standards. Furthermore, even the new maps do not account for rapid rain accumulation in the event of intense precipitation. According to a statement released by the Association of State Floodplain Managers, these maps will be obsolete immediately until they do (Bloomberg, 2017).
The ORR has recognized these shortfalls and has worked to augment the FIRM in order to improve the accuracy of the City’s flood map. In particular, analysts and engineers are focused on modelling rainfall accumulation in order to more accurately predict flooding from storm events. However, in order to accurately model rain accumulation and improve the accuracy of the FEMA maps more generally, additional data is needed on actual incidences of flooding. This data has proved to be remarkably difficult to collect, given that real-time, high-resolution datasets are rare if non-existence (Wang et. al., 2018). This data vacuum is not insignificant. The ORR’s resiliency initiatives are based on the data available to them, and these initiatives are the only line of defense against the increasingly hazardous impacts of climate change. The absence of real-time, comprehensive flood data deprives the ORR of information that could otherwise be used to more quickly deploy effective solutions in our City’s infrastructure and neighborhoods. As the negative impacts of climate change intensify, this puts all New Yorkers at risk.
This project includes a variety of stakeholders. The external stakeholders are comprised of New York City residents who are at risk (both now and in the future) of urban flooding, as well as the analysts and engineers within the ORR who are developing and executing initiatives to mitigate that risk. Beyond this, there are a number of city agents (such as architects, urban planners, emergency responders, etc.) who work alongside the ORR in executing resiliency and disaster response initiatives. These agents are also stakeholders as their operations would benefit greatly from real-time, comprehensive flood data. The last major external stakeholder is FEMA, who is responsible for financially reimbursing residents whose property is damaged from flooding events if their property was located outside the FIRM. Improving the accuracy of the FEMA flood maps would dramatically reduce the agency’s financial liability in the event of a flood related disaster.
The internal stakeholders include the Watson FMS Project Team, as well as the project’s sponsors. Sponsors would include the ORR as well as the IBM Watson Division. Given that the project utilizes Watson AI and Cloud technology, IBM would have both a reputational and financial interest in its long-term success. This list is not exhaustive.
The Watson FMS Project Team envisions a world in which coastal urban centers like New York City are resilient enough to withstand the devasting effects of climate change. Resiliency does not just refer to buildings or infrastructure, but to people as well. Watson FMS is about utilizing technology to create new forms of collaboration between citizens and government in order to solve the difficult problems associated with urban flooding.
The IBM Watson Flood Mapping Software (Watson FMS) represents a novel approach to the collection, processing, and visualization of flood data. The application leverages Watson’s Natural Language Understanding (NLU) and Visual Recognition to extract data from unstructured Twitter posts during a disaster- or intense storm- situation and uses that data to construct a crowdsourced map of where residents experience flooding throughout the city.
Watson FMS collects data from the microblogging service, Twitter. Twitter’s freely available Streaming API allows for the continuous retrieval (and temporary storage) of tweets that meet certain pre-established filter predicates (Fohringer et. al., 2015). Multiple case studies have demonstrated the viability of this approach in constructing informative flood maps (Fohringer et. al., 2015; Smith et. al., 2015; Eilander et. al., 2016; and Wang et. al., 2018).
While the Streaming API can filter out tweets with certain keywords, NLU capabilities are necessary to extract references to location and intensity within the unstructured text.1 Watson FMS will do this by classifying certain text strings as ‘location’ entities or ‘quantifier’ entities. Once this is complete, the Bing Maps Location API is used to convert the ‘location’ entities into an address with a latitude/longitude coordinate (Wang et. al., 2018). These coordinates are then displayed on the web application’s front end for analysts and engineers to query. In addition, Watson’s Visual Recognition capabilities can be used to analyze the attached photos to determine flood depth.
Two important design considerations are privacy and participation. The use of the Twitter Streaming API obviates many of the concerns around privacy. Twitter data is largely publicly accessible, and the Streaming API only streams tweets that are. Because of this, there is no way of discretely accessing private social media data. As mentioned above, restricting the application to publicly-accessible data does not compromise the viability of the approach.
The second concern is over citizen participation. Watson FMS is only viable insofar as residents in the New York City area tweet about flood events and include either photos or location and/or intensity information within the tweet. This is not as implausible as it may appear. 20 million tweets were posted in the Northeast in the five days covering the approach and aftermath of Hurricane Sandy, 25% of which involved people sharing photos and videos (Pew Research, 2013). However, there are ways to improve the posting of relevant information. A behavioral outreach campaign (similar to ‘See Something, Say Something’) can be used to encourage Twitter participation in the days leading up to a major storm event. Another option is to incorporate a feedback loop with Watson Assistant in which the Assistant requests additional information from profiles who have already tweeted about the event.
Watson FMS is a novel tool to collect real-time, comprehensive flood data in order to better inform the strategic planning efforts of the ORR. When used to conjunction with FEMA’s modernized maps, it has the potential to provide significant financial savings by limiting the economic damage from flooding events. The Cost-Benefit Analysis includes the first three years of implementation, as well as a fourth year ‘n’. This year is any year in the future in which New York City experiences a 100-Year-Flood (a storm on the scale of Hurricane Sandy for example).
The initial implementation costs for the FMS system are expected to total $320,000. These costs include the one-time cost of developing the web application, along with the first-year costs associated with the Watson API Subscription and marketing campaigns. In subsequent years, the web application cost is deducted but an on-going maintenance cost is added that is equal to 10% of the web application cost. Annual maintenance costs in subsequent years are estimated to be $185,000. In the event of a 100-Year-Flood, marketing costs are increased substantially for an annual estimated cost of $5,135,000.
The project’s savings are calculated using a methodology outlined in Appendix 3. It is expected that the project will take a loss in the first two years. This is because the system needs time to record a certain number of flooding incidences before the data can be measurably useful. This loss is estimated at $505,000 at the end of the second year. However, the project becomes exponentially more profitable as the years go on, particularly in the event of a 100-Year-Flood. At the end the third year, savings are estimated to be
$6.46 million, and in the event of a 100-Year-Flood, savings are estimated to be over $2 billion. These are rough estimates. Savings are in the form of reduced loss of property and business, as well as more efficient resource deployment (Board of Earth Sciences and Resources, 2009).
While this proposal is primarily focused on the benefit that Watson FMS will have on the strategic planning efforts of the ORR, the software has many other potential applications. The Department of Emergency Management, for example, could use Watson FMS improve their emergency response in the event of a Hurricane or related natural disaster. Real-time flood data would provide the agency with broad-based situational awareness and help them prioritize their rescue efforts in the most efficient way possible. Furthermore, in the immediate aftermath of a disaster, a crowdsourced flood map would provide invaluable data to federal and state relief agencies as they coordinate their response efforts.
Watson FMS is an application designed to provide a dataset that was never before possible. It represents a first of its kind ‘big data’ approach to the problem of urban flooding. By leveraging the technology afforded by IBM Watson, it is possible to use data as a remarkably effective weapon in our City’s fight against climate change.
Cost-Benefit Analysis Table + Methodology
|Action||Year 1||Year 2||Year 3||Year N
|Costs||Web Application Implementation1||$150,000||–||–||–|
|Watson API’s Subscription Costs2||$120,000
|Behavioral Outreach Campaign3||$50,000||$50,000||$50,000||$5,000,000|
|Benefits||Est. Flood Damage5||$174m||$174m||$174m||$25.4b|
|W-FMS Accuracy Contribution7||–||–||5% y2y||10% y2y|
- Cost estimate for the development of a Web Application (large enterprise) using a ‘moderate class’ custom development http://www.comentum.com/web-development-cost-rate- comparison.html
- Watson Service Subscription (Cloud) based on consistent usage over the https://console.bluemix.net/pricing.
- Estimates will fluctuate based on the number of severe storms in a given In the event of a 100-Year-Flood (or a storm on the scale of Hurricane Sandy), campaign costs
- Estimated to be 10% of Implementation
- Estimate taken from a collection of 549 simulated storms ranging from small storms to severe The average damage per year from a simulated storm was $174m, while the damage from a Hurricane simulation was $25.6b. https://www.scientificamerican.com/article/massive- seawall-may-be-needed-to-keep-new-york-city-dry/
- Research data suggests that the B/C ratio for accurate flood maps is an average of 8. The B/C Ratio is measured as the incremental benefits of improved map accuracy divided by the incremental costs of flood damage absent the improved map accuracy. Chapter 6: Benefits and Costs of Accurate Flood Mapping in Mapping the Zone: Improving Flood Map Accuracy. https://www.nap.edu/read/12573/chapter/6
- Watson FMS is estimated to account for 5% y2y of the improved accuracy of FEMA’s Flood Maps in the early years of implementation, and 10% y2y in the later years. It is estimated that it will provide a negliable contribution in the first two years as the system needs time to record a certain number of flooding incidences before the data can be measurably These are rough ball-park estimates.
- ($174,000,000) (0.80) (.05) where $174m is the est. annual flood damage, 0.80 is the proportion of every dollar of flood damage that is saved by improving map accuracy, and 0.05 is the proportion of improved map accuracy that can be attributed to Watson FMS.
Board on Earth Sciences and Resources / Mapping Science Committee. 2009. ‘Chapter 6: Benefits and Costs of Accurate Flood Mapping’ in Mapping the Zone: Improving Flood Map Accuracy. Online Edition.
Eilander, D., Trambauer, P., Wagemaker, J., van Loenen, A., 2016. Harvesting social media for generation of near real-time flood maps. Procedia Eng. 154, 176-183.
Fohringer, J., Dransch, D., Kreibich, H., Schroter, K., 2015. Social media as an information source for rapid flood inundation mapping. Natural Hazards Earth System Sciences. 15 (12), 2725 2738.
Keller, M., Rojanasakul, M., Ingold, D., Flavelle, C., Harris, B., 2017. Outdated and Unreliable: FEMA’s Faulty Flood Maps Put Homeowners at Risk. Bloomberg.com
OneNYC. 2015. One New York: The Plan for a Strong and Just City. nyc.gov.
Pew Research Center. 2013. Twitter served as a lifeline of information during Hurricane Sandy.
Shaw, A., Thompson, C., Meyer, T., 2013. Federal Flood Maps Left New York Unprepared for Sandy and FEMA Knew It. ProPublica.com.
Smith, L., Liang, Q., James, P., Lin, W,. 2015. Assessing the utility of social media as a data source for flood risk management using a real-time modelling framework. Journal of Flood Risk Management.
Wang, R.Q., Mao H., Wang, Y., Rae C., Shaw W., 2018. Hyper-resolution monitoring of urban flooding with social media and crowdsourcing data. Computers and Geosciences. 111, 139-147.