Inspiration

When roads become inaccessible, some groups, such as residents without access to private vehicles, older adults, and people with disabilities, can face greater difficulties reaching evacuation resources than others.

There is a need to solve a problem that traditional risk maps do not consider in full: Knowing which communities are vulnerable is only half the problem. How do we figure out where to direct limited resources to assist them?

Crisees is born out of this question to develop an approach to turn geographic and demographic data into actionable evacuation planning decisions for southwest Houston.

What it does

Crisees is an equity-aware flood evacuation planning platform that enables disaster planners to identify access problems and optimize allocation of limited evacuation resources.

By processing geographic and demographic data, Crisees:

  • Analyzes accessibility of transportation to 151 tracts in southwest Houston.
  • Models possible disruptions in the road network related to flooding and their impact on evacuation accessibility.
  • Identifies vulnerable communities that may face difficulties reaching evacuation sites.
  • Uses mathematical optimization for selection of temporary evacuation pick-up hubs and allocation of limited transport means.
  • Compares baselines, disrupted scenarios, and potential interventions through optimizing resource allocation.
  • Visualizes findings in an interactive dashboard.

How we built it

We've designed Crisees as a full stack geospatial intelligence and optimization platform.

Data infrastructure: We've integrated publicly available American Community Survey demographic data, Social Vulnerability Index data by CDC, FEMA flood-hazard data, and derived roads and facilities based on OpenStreetMap. At the moment, our study region consists of 151 tracts and approximately 585,584 residents.

Backend: Our solution is built using Python and FastAPI for geographic data processing, network reconstruction, accessibility estimation, and exposing analytical capabilities using RESTful APIs.

Network analysis: By means of NetworkX and OSMnx-based data, we've performed shortest path calculation, estimated travel times, simulated reproducible disruptions in the road network.

Optimization: We've built a mixed integer programming optimization model by using SciPy to choose evacuation pick up locations and allocate limited resources taking into account vehicle capacity, available pick up hubs, shelters, time and other constraints. It uses transparently weighted priorities for equity maximization and independently verifies validity of the solutions with respect to mathematical constraints.

Frontend: Interactive visualization and scenario evaluation is implemented using Next.js with TypeScript, Tailwind CSS, and MapLibre GL JS and connected to backend API to perform all necessary computations.

Challenges we ran into

Firstly, we faced a challenge of working with publicly available geospatial data, which might be incomplete and have inconsistent format. Some of our main datasets could not be accessed during development due to restrictions. Therefore, we used alternative open sources of the data, preserving source attribution and marking their limitations.

Also, the problem we face is an issue with accurate modeling of evacuation access when some of the data is missing. FEMA flood-hazard data only covers about 52% of our current study area. Also, our road network doesn't contain any sidewalk information. Thus, we separate geographic information that is missing but assumed not to create hazards from the information that indicates danger.

Next one is that allocation of the resources is a quite difficult task to be done in our case because choosing pickup location does not mean that it is possible to evacuate everyone there. For this reason, it was necessary to take into account transportation, vehicle, shelter capacity and feasibility of allocating resources, so the model counts assistance only in case when it is possible to assist people.

The final challenge is related to the process of building geospatial application with single developer. Here we should balance computational efficiency, data processing, backend design, and frontend interface.

Accomplishments that we're proud of

We're most proud of our ability to turn Crisees into a working application instead of remaining with just a concept or data visualization.

Among other achievements, we are proud of:

  • Processing of 10 public datasets with geographic data covering 151 tracts.
  • Creation of the network consisting of approximately 88,404 nodes and 199,542 directed road edges.
  • Simulation of reproducible scenarios of disruption caused by flooding and estimation of tract-level accessibility.
  • Development of mixed integer programming optimizer and its independent constraint verification.
  • Implementation of the interactive geospatial dashboard with scenario simulation, resource allocation and impact analysis functionality.
  • Writing and passing 157 backend tests and 11 frontend tests, including testing in actual browser connected to the backend.

For example, in our severe flooding scenario, the model estimates 23,033 people needing to receive transportation assistance. In this situation, with allocation of 12 evacuation pick-up hubs and 15 vehicles, modeled assistance can be increased from approximately 5,214 people to 9,493 people.

That means an increase in assistance coverage by approximately 18.6 percentage points compared with disrupted scenario without any intervention.

We're especially proud that, in addition to recommendations themselves, Crisees provides explanations of limitations of resource capacities, unsatisfied demand and underlying trade-offs for each recommended allocation.

What we learned

We found out that identification of vulnerable communities and intervention to assist them are two very different problems.

Geographic data, network science, and mathematical programming can be used in combination to make decisions related to disaster planning transparent and verifiable. And one of the most important lessons we've learned was the difference between the actual observation, modeled assumptions and unknown conditions.

On the engineering side, we've got an opportunity to gain practical experience with integration of geospatial datasets, application of graph algorithms, usage of mixed integer programming optimization, asynchronous API and geospatial map visualization.

But most importantly, we have realized that creation of technology for societal benefit requires much more than just generation of impressive numbers.

What's next for Crisees

Our next step is the improvement of geographic coverage and quality of geospatial data by adding complete flood-hazard data by FEMA and detailed information about pedestrian network.

In addition to southwest Houston, we want to apply our technology to other geographic regions and to other scenarios of disaster and crisis situations.

Built With

Share this project:

Updates

Submission history