We started by reaching out to a close colleague, Rick Wilson, a behavioral social scientist at Rice University, who studies decision making. Together we designed a series of online experiments, using risk maps, in which hundreds of Houstonians were randomly assigned to various levels of data resolution and risk type. We focused on the time spent searching a map: more time indicated citizens’ interest in storm risks and their willingness to take action to prepare. Although big hurricanes—say, category four—got the most notice, attention disappeared if the geographical data on storm effects were not local. People were not interested in maps divided into areas that were a kilometer wide or partitioned by zip code. But when the map showed data on almost every block, hundreds of users sought more information. We also learned that, particularly in inland areas, projected rain amounts got more interest than projected storm surge levels. Serious rain events affect people’s mobility, productivity and safety.
These behavioral experiments showed that individuals pay the most attention to risks they perceive as most relevant to their own situation. This is obvious to us in hindsight, but think of how it contrasts with the way most storm information is handed out today—official blanket statements for rare events covering areas of many hundreds of square kilometers, such as entire counties and zip codes.
With our new focus on local events, we began to build a system around rainfall runoff and accumulation. We call it the Hurricane and Rain Vectorized Exposure Yielder, conveniently abbreviated to HARVEY. Our computer model HARVEY has a much finer geographical terrain grid than our earlier map, using cells that are only several square meters, instead of square kilometers. A single street can have many of these new squares, and the total for the city is more than 100 million of them. This configuration provides much more precise estimates of overland water flows and their depths when it rains intensely.
We used a variety of sources to derive these estimates. We had the history of National Weather Service forecasts and data, of course, but our model also incorporates the locations of calls to Houston’s 311 city information service to report local flooding. We can also draw on emergency calls to fire and police departments asking for help. Repeated calls from a particular location indicate recurring trouble spots. Harris County has a network of rain gauges, and we pull data from them. (We are also testing a wireless network of street-level flood sensors.) Our prediction models also include radar data that show how much water is held in the clouds heading for the city and how fast the wind is moving them. Slower winds give the clouds time to dump a great deal of water. That scenario produces a lot of nonhurricane flooding and was behind the inundation created by slow-moving Hurricane Harvey last year.
All these data are superimposed on a high-resolution terrain map, derived from the Houston-Galveston Area Council’s laser-driven remote-sensing system, which captures minute differences in ground height. The entire thing is integrated by AI programs that use fancy-termed techniques such as ensemble regression models, deep-learning algorithms and high-dimensional vector spaces. But the basic point is they are much more capable of combining different types of data sets than were the engineering models and mathematics we used for our original storm calculator.
We have tested HARVEY by giving it several sets of initial conditions seen prior to storms since 2015 and have asked the program to produce flood estimates for multiple places across the city. The predictions HARVEY has churned out have matched actual field observations of these storms well. The program does best with heavy downpours, more than five centimeters—two inches—per hour that last several hours, and in spots with poor drainage because of bayou overflows and bay tides. For smaller events, we will be calibrating HARVEY one watershed at a time over multiyear periods, to capture local factors and longer-term effects of climate change.
What would this mean for our worried Houstonians Bob and Alice? Our new map would provide them different levels of risk, with more attention paid to the history of flooding near Alice’s house and the height of the land around Bob’s. The key difference is that even if Bob and Alice lived two blocks apart, rather than two kilometers, they would be given different risk levels. With the erratic rainfall patterns across the city for any single event, users like Alice and Bob may find very different estimates of street flooding around their homes, their workplaces and the routes in between. Our HARVEY system will show users like them the dangers that affect route choices, the possibility of getting trapped at their locations and the likely levels of flooding for homes during rainfall events. It will help the city government allocate emergency and planning resources in advance, allowing first responders, such as the fire department, to get to people in trouble faster. Storm-mitigation projects can be located in areas that need them most.
Our plan right now is to publically launch a beta version of HARVEY in 2019, designed specifically for residents of the hard-hit Brays Bayou watershed. This waterway crisscrosses a neighborhood called Meyerland, where homeowners have been surprised by flooding multiple times during the past five years. Their residences have been wrecked, rebuilt and wrecked again. On many occasions people have been stuck in these houses, watching the water rise. We hope to give them better and earlier warnings. Our next step will be to expand the system to reach the rest of the city. Our team is entering into a collaborative agreement with the city of Houston, the Kinder Institute for Urban Research, and the Severe Storm Prediction, Education, & Evacuation from Disasters (SSPEED) Center to test and deploy HARVEY in stages toward future city-wide coverage. And if the model works for Houston, it could be adapted to other cities across the world that face similar problems from severe weather.
A changing global climate is going to make rainfall worse in our region, according to conclusions reached by a 2018 Houston severe storm conference organized by the SSPEED center. Storms will stagnate more frequently, dumping more rain as a consequence. Tools such as HARVEY will provide flood estimates at a scale that public officials and private citizens seek as they try to plan for this intensifying chronic rainfall and runoff. Most important, these tools will give people who must live under these clouds the ability to answer, for their own safety and that of others, one urgent question: Should I stay, or should I go?