Inspiration

With one of our teammates being a native New Yorker, we understood that hurricanes frequently devastate NYC and the rest of the east coast. When Hurricane Sandy hit New York in 2012, the city paid $19 billion dollars to repair damage. Long Island experienced had massive floods, broken trees, homes and businesses destroyed by fire, loss of electricity for 1 week, among many other tragedies. Thousands of civilians were left without electricity, food, or homes. There were no phone networks and people couldn't call 911. Disaster Management would visually inspect the damaged area to evaluate and report the situation before providing aid. They didn't use the most critical resource in an efficient way!

But we knew that natural disasters weren't limited to the east coast. Our resident Californian knows about living in constant fear of the next massive earthquake. The 1906 San Francisco earthquake and the Loma Prieta earthquake demolished buildings, caused infrastructure to crumble, and isolated cities and people from necessary aid. New massive earthquakes seem to rock the news every day: Haiti, Japan, Chile, Nepal, and many more. Natural disasters that isolate people from aid aren't limited to hurricanes and earthquakes. Wildfires and tornadoes are notorious culprits as well.

Current methods for creating plans of action to distribute aid require manual labor to go into damaged communities for visual inspection. We knew there had to be a better way.

What it does

DisasterFaster is a real-time web application that predicts and recognizes regions struck by a natural disaster that have the greatest need for immediate relief. The model is trained using geographical and environmental parameters incorporating the ground elevation, the likeliness of property destruction, and the severity of the disaster. It also looks into real-time traffic data to identify areas with a high population density which would require immediate attention. Emergency Management Teams can use this service to monitor the heat maps and target the damage-prone areas in a drastically faster duration and send resources more efficiently. The lightweight web app combined with beautiful data visualization allow these teams to quickly put together a plan of action.

How we built it

We used HTML, CSS, and JavaScript along with Google Maps APIs (elevation, geocoding, reverse-geocoding, heatmaps, and markers APIs) to create a visualization that highlights areas prone to extreme damage. The heat map presentation of critically damaged areas enables responders to quickly and efficiently identify where to focus their efforts.

Challenges we ran into

Learning prediction models need data. We scoured many federal, state, and independent datasets to find workable data, and while we got lucky to find a treasure trove of open source data released by the state of New York, we still had to heavily process the data we extracted. The lack of available data on some parameters we wanted to include greatly limited how complex of a model we could generate. Moreover, our heavy reliance on map APIs and the quotas on requests to them also limited how much data we could present.

Accomplishments that we're proud of

Despite a lack of data and relative inexperience working with map analysis, we were able to develop a functional model that predicted high risk areas. The thrill of seeing our model work was an indescribable mix of emotions, including excitement, relief, and a hunger to improve it further.

What we learned

We all learned new technical skills including working with asynchronous JavaScript requests, map analysis with GIS, and data visualization principles to present and highlight the most important information upfront so that the user quickly obtains the information for which they came.

What's next for DisasterFaster

The biggest struggle and biggest room for improvement was the quality and quantity of data that we fed into our model. We can improve the accuracy of our prediction map with a more advanced machine learning based model such as an artificial neural network. We could feed in our parameters (elevation, average age of buildings, status of communication lines, amongst others) and compare the prediction with damage maps from previous natural disasters in the area.

Share this project:

Updates