Crowdsourced DisasterMaps

There is no more critical service than that of responding to a disaster; peoples’ lives and property hang in balance. With DisasterMaps communities have direct input to emergency services by showing where they are needed. DisasterMaps also makes government actions more transparent by displaying official disaster information. With it, citizens feedback their current status so governments can be more responsive. Finally, it makes governments more accountable by recording the situation's evolution for later review. These are the “dimensions” of the “Community-Driven Innovation” challenge.

Responders and residents are frequently faced with making life and death decisions without any knowledge beyond what they themselves can see.

However, adding up what each of us sees and sharing the composite picture can save lives. DisasterMaps uses smartphone cameras to take pictures, analyze them and share them as icons on maps.

DisasterMaps can save tens of thousands of lives and help tens of millions yearly!

Currently the problem is that in most disasters residents don’t have the information they need to make informed decisions to protect themselves, their families and their property. Furthermore responders also typically lack the information to optimize their actions and resources. Both residents and responders need to know what's happening...now.

Shelter in place or evacuate? What evacuation route? What will responders face? Where are people needing rescue?

Our story:
    We went to sleep...the fire was over 30 miles away. No alarm!
    We awoke; our dog was barking at the window. No warning!
    We saw a red glow as we approached the window. Panic!
    We grabbed some clothes, important papers, and the dog.
    We took one car, afraid to be separated.
    But which way to go? Smoke, no information!
    The other way led deeper into the fire. We survived.
    When we returned our home was gone.The other car melted.

Other stories from Knoxfield News 12/2/16: “Gatlinburg fire evacuees: No warning, no text alert”

Has this happened to you? Or your relatives or friends? Read similar stories?

Wildfires, earthquakes, floods, hurricanes and other disasters don’t discriminate...they affect everyone but some people are forced to live in more vulnerable areas and some have more resources to deal with their aftermath. And almost all complain of the lack of real time knowledge of local disaster conditions that could have helped them.

DisasterMaps will begin by serving people in the US and then Canada, Europe and Australia. But as our focus disasters expand beyond wildfires, others around the world will become a greater focus. Even the poor in most countries have smartphones and connectivity and the Internet will come more and more from worldwide satellites. With the constant outpouring of new smartphones, old ones will become ubiquitous and useful for DisasterMaps. The need is worldwide and DisasterMaps will be freely available worldwide.

DisasterMaps is entirely software and runs in a browser on a smartphone. There are actually two web apps: Map and Alarm. Map displays a zoom and pan map with icons and official data overlays depicting the disaster data, the current situation.

Alarm is a sentinel watching for signs of a disaster. We’re recommending that people dig out their old Android smartphones or buy the cheapest Android they can find; keep it plugged in and online on their WiFi; and affix it to a window looking in the direction of greatest threat. In addition to taking pictures, analyzing them for a disaster, and uploading any disaster data, Alarm also displays the Map functionality.

For official data on wildfires in the US, we’ve had initial discussions with UCSD’s WiFire which works with FEMA and others to compile and provide such data. DisasterMaps would overlay WiFire Official data. The official data overlain as depicted below is satellite imagery.

For our prototype, we are using Javascript for coding; Firebase for the database; and mappa.js to interface to leaflet.js and OpenStreetMap for mapping. Currently we are working on the AI for recognizing fire to be used in Alarm for wildfires. Two technologies seem adequate for the task, neural nets from XNOR.ai and Google’s autoML Vision Edge. 

In addition large wildfires produce smoke that is hazardous to both short and long term health, especially children's health. General movement is tracked like weather but local variations can be significant and can be detected by visibility (from smartphone pictures) and temperature and humidity.

Similar AI efforts will be required for identifying floods, and other locally observable disasters. 

Earthquake warnings are already being provided in the western US and worldwide (shown below) based on crowdsourced smartphone accelerometers. This will become another Map and Alarm for our users and be useful for reporting conditions after the earthquake for responders. An app for providing earthquake alarms in Mexico has been downloaded over 3 million times.

Atmospheric disasters (hurricanes, tornadoes, heat waves and droughts) will rely on weather data augmented by local observations but have not been explored to date.

Crowdsourced smartphone pictures analyzed for signs of disaster and widely distributed as alerts with real-time situation maps is unique. But everything has an antecedent. The earthquake networks use crowdsourced smartphone accelerometers. Automated photo analysis is being used for fire detection. Using a smartphone as an observation platform is becoming popular, especially with medical and health apps. Disaster situation maps are often used by the media and real-time weather maps are commonplace. Finally, many apps today have alerts.

But DisasterMaps brings together many incremental steps to make a "disaster monitoring station", like a weather station, based on smartphones. The Weather Underground took the costly home meteorological station and built a volunteer community of over 250,000 stations worldwide. Using a consumer's smartphone, the Weather Underground distributes their analyzed data to millions in the form of apps. DisasterMaps builds on that consumer's smartphone model with local intelligence--machine learning models to analyze pictures for disasters and their effects--within the Alarm app.

So all the parts of DisasterMaps have had successes in reasonably similar applications. What makes DisasterMaps unique and exciting is the integration with a simple user interface and dramatic user experience: browse to the website and click on Alarm or Map and the app runs. Each has minimal barriers to use--no authentication (sign up, sign in) and only required permissions for the Alarm's use of location and camera. And minimal controls: normal zoom and pan for maps, and slide-out drawer for icon and overlay key. Everything is automated for the simple user.

DisasterMaps is software that relies on smartphones for observations (pictures); analysis (machine learning visual classifier); communications (uploading disaster conditions and downloading icons and maps); and display (of the maps.) The software could also run on tablets or computers with a camera.

The software itself is based on HTML and Javascript (runs in a browser) with Firebase (a database API) and mappa.js to interface to leaflet.js (a mapping program), and uses the OpenStreetMap map data.

For picture analysis, DisasterMaps Alarm apps will use machine learning models for each disaster running locally in specialized software from either XNOR.ai (a binary AI) or Google (autoML Vision Edge.)

The current problem as stated above is the lack of information during a disaster. DisasterMaps directly provides such information in the form of maps with icons and overlays to illustrate the current situation due to the cause of the disaster...where is the fire, smoke, flood waters, earthquake damage, etc. This assumes that people will install the Alarm and watch the Map for a disaster.

To start that behavior, DisasterMaps will directly solicit initial users through crowdfunding, local groups and local news broadcasters. Groups can start the viral spread of the apps and the social aspect of helping neighbors should appeal to them. Broadcasters have a vested interest...when a disaster strikes, they would like to have maps that they can use to explain the situation to their audiences. Thus we expect to get human interest stories before disasters and strong suggestions to install the Alarm as the disaster is in progress from broadcasters. Finally we expect to see many human interest stories of how DisasterMaps saved lives afterwards setting up further viral spreading of the app.

These were seen to play a role in the Weather Underground and earthquake networks, especially in Mexico where it has grown to three million.

Within a year we expect we'd be serving several thousands with a Wildfire Alarm. Wildfires in the US and Canada are almost all year and several major ones that persist over many days can be anticipated in Spring, Summer and Fall. These will be opportunistic targets and parallel our efforts in known "hotbeds" like California. All we need is several hundred Alarms in threatened homes in a wildfire for the maps to be interesting to those concerned which may number ten or more times as many people.

Adding earthquake disasters provide a reason to keep the Alarm apps going continuously and will add to viral spreading. With funds we may soon add Flood Alarms further increasing the population of users.

In five years, DisasterMaps should be serving millions of people worldwide. Most coastal lands and especially islands are subject to hurricanes, typhoons, cyclones and tsunamis which would be other modules for DisasterMaps. And lowlands are subject to flooding and mudslides, other modules. Of course, earthquakes are a major threat in many places around the world such as Japan where major adoption would be expected. The records of the Weather Underground app and the earthquakes networks suggest 3-5 million is possible but today's social networking may make viral spreading really supercharged a lead to substantially higher usage.

Within the next year our goals are to have a thousand users who actually use DisasterMaps to navigate their way out of a threatening wildfire, have a hundred provide testimonials and have a hundred media articles noting the successes. Supporting these goals our objectives for the next year include:

• incorporating as a not-for-profit with credible Board;
• raising $250,000 from crowdsourcing and foundations;
• completing development with minimal staff, using consultants and contractors for flexibility;
• development of an open source machine learning model classifier that is 99% accurate for wildfires through contests;
• integrating wildfire and earthquake disasters;
• crafting terms of use that provide simplicity for users and legal protection; and
• obtaining the endorsement of a major broadcast network.

Within the next five years our goals are to have a million users worldwide who actually use DisasterMaps to navigate their way out of a threatening disaster and have the emergency responders and local media worldwide routinely rely on DisasterMaps. Supporting these goals our objectives for the next year include:

• becoming a worldwide brand;
• building a sustaining staff of 5-10 talented, committed individuals;
• developing a business model that generates sustaining and evolving funds ($1-2M/yr);
• integrating all disasters that fit the crowdsourced smartphone model;
• extending the fixed smartphone model to compile mobile observations; and
• scaling the codebase and supporting services for millions of users.

Within the next year the anticipated barriers to accomplishing our goals are:

• Financial: raising $250,000 is difficult especially as donations and grants, without equity.
• Technical: no barriers but there are some challenges: ML model accuracy, communications at scale with minimal spending.
• Legal: using simplistic “terms of use” and depending on good Samaritan laws has some risk.
• Cultural: some are averse to technology and some will worry about privacy  with taking pictures and sending data.
• Market: this app could be subsumed by one of the big guys, such as Google’s SOS.

Within the next five years the anticipated barriers to accomplishing our goals are:

• Financial: developing a revenue model will be difficult when users don’t want to pay or to be delayed.
• Technical: we’ll need to keep up with future changes in technology.
• Legal: as we gain value, we’ll need better legal protection from errors.
• Cultural: going worldwide will require understanding other cultures.
• Market: as we have success, commercial competitors may arise, ex SkyAlert provides commercial earthquake warnings.

Within the next year we’ll overcome the anticipated barriers by:

• Building a respected Board and advisors to help with financial, cultural and market barriers. 
• Using consultants and contractors to overcome the technical barriers.
• Consulting with lawyers on the legal barriers.

Within the next five years we’ll overcome the anticipated barriers by:

• Reviewing free, not-for-profit app revenue models and consulting with social media experts. Partnering with groups supplying services to stressed peoples.
• Building a small nimble in-house technical team to direct use of emerging technologies.
• Employing legal services to minimize legal risk.
• Using multi-cultural and local experts to adapt the text, UI, and UX for different cultures and their needs.
• Working cooperatively with potential competitors including commercially-focused ones.

N/A

Only one person has been behind the development of DisasterMaps. He has been supplemented by colleagues, mentors and consultants as needed. His plan has been to evolve the project in private through development of a proof-of-concept prototype and then open it up to the public to bring in others. He has successfully used this same formula in founding projects as diverse as planning the industrialization of space, developing an emergency response system for chemical industry, and designing and manufacturing the first tablets.

Now with the prototype nearing completion, the public and money-raising phase is starting with this application for recognition.

I’m a PhD physicist who enjoys solving technical problems and assembling groups to package the solutions as products. Most of my career can be characterized as the application of computers to solve problems. Most of the problems have been in the environmental sciences especially atmospheric transport (of nuclear material, chemicals and air pollution) and emergency response (for chemical plants, shipping chemicals, and highway problems.) Some of my colleagues in atmospheric transport focused on fires since the techniques were overlapping. Many of my colleagues in emergency response focused on fires, floods, hurricanes, mudslides, etc. I’ve been in the management of two international emergency response systems companies, worked for US EPA and NOAA, and worked with literally hundreds of companies and government agencies concerned about emergency response. This project is a perfect capstone to my career.

Some colleagues who are supporting DisasterMaps have experience consulting and managing projects for Federal, State and Local emergency management agencies; working for the VA State Department of Emergency Services, Department of Transportation Emergency, and Deputy Assistant to the Governor for Preparedness; as a senior executive in telecom and computer software industries; and as a consulting meteorologist and working on the National Emergency Response computer models.

We’ve had initial discussions with Jessica Block, Associate Director of UCSD’s WiFire which compiles and provides official (US, state and local agencies such as FEMA, National Fire Program, CalFire) and power company data for wildfires to the public. Our preliminary agreement is that DisasterMaps would  have access to and overlay WiFire data, and WiFire would add DisasterMaps point observations to their compilations. This reciprocal data sharing would be our typical partnership that we’d expect to have with groups focused on making disaster data public such as MIT’s Riskmap (floods and other hazards), UH’s Pacific Disaster Center (hurricanes), and UNIBG’s Earthquake Network and UCB’s ShakeAlert.  

Though working with for-profit, private companies could be more problematic, we’ve already had acceptance for the data sharing model with EnviroVision Solutions who have networks of mountaintop automated cameras to detect wildfires in the West and worldwide.

Key customers and beneficiaries: Residents who are threatened by a disaster. News broadcasters who want to disseminate information about an ongoing disaster. Emergency responders who want detailed information on disaster conditions.

Service provided and how provided: Real-time disaster conditions at points where an Alarm is installed, ex. observed fire or no fire, primarily delivered over the Internet as icons on Maps and secondarily as commentary on news broadcasts.

We plan to rely on donations and grants to complete the product development and for initial marketing. We’ll start with crowdfunding sites such as Kickstarter, Indiegogo and GoFundMe. They provide a funding platform and offer an audience for generating interest in our success. We will also explore Foundations and government grants. On a continuing basis we expect our users, especially ones that have used DisasterMaps to evacuate, to fund our continuing efforts. There is also a possibility of corporate sponsorships or advertising services that evacuating people would need. It is our intention to make DisasterMaps always be free to the public and to have a very lean and efficient operation.

Since DisasterMaps is planned around a lean operation, so the prize money will be significant revenue this year. But of much greater value is the credibility that will accrue to being an MIT Solver. It will help us raise funds from public donations and foundations. And of course the introductions and expert advice will be invaluable.

Introductions to government agencies and potential corporate sponsors.

To start, DisasterMaps would like to partner with MIT’s Riskmap group headed by Professor Miho Mazereeuw to share disaster data.

In the government we would like to partner with FEMA, DHS, National Fire Programs, DOI, Forestry Service both to share disaster data and for agency response personnel to be aware of (and possibly use) DisasterMaps. Of course we would be consumers of data from NOAA, NWS, and NASA (satellite data.) 

A group that responds to emergencies that we’d love to partner with is the Red Cross. We’d share target audiences and being a partner with the Red Cross would lend a great deal of credibility to DisasterMaps.

A corporation that could help us is Google. We expect to be users of Cloud Storage, Vision ML, and ML Kit for on-smartphone disaster detection. So their technical assistance and possibly reduced charges would be helpful.

Apple could be helpful. Currently old iPhones without an active SIM card may not work on WiFi, as planned for most DisasterMaps Alarms. Perhaps there is some workaround.

DisasterMaps is based on smartphones that can detect features of disasters in their photos. The needed vision algorithm is developed via machine learning using images of disasters (ex. wildfires, floods) and similar images without any sign of disasters (ex. Sunsets, highway tail lights, rivers, lakes, ocean.) Beyond the disasters themselves (ex. wildfires, floods, mudslides, tornados), DisasterMaps should recognize destruction (ex. burnt structures, earthquake rubble, water or mud inundation) in the wake of a disaster. Initially we plan to use Google’s auto ML to train an algorithm and then deploy it using ML Kit into smartphones. Also we plan to hold contests to maximize the accuracy of an algorithm and to optimize its execution on a smartphone. This prize will help fund those activities.

N/A

N/A

N/A

DisasterMaps is based on smartphones that can detect features of disasters in their photos. The needed vision algorithm is developed via machine learning using images of disasters (ex. wildfires) and similar images without any sign of disasters (ex. Sunsets, highway tail lights.) Beyond the disasters themselves (ex. wildfires, floods, mudslides, tornados), DisasterMaps should recognize destruction (ex. burnt structures, earthquake rubble, water or mud inundation.) Initially we plan to use Google’s auto ML to train an algorithm and then deploy it using ML Kit into smartphones. We’re focused on doing the disaster detection within the smartphone so pictures are never sent to the cloud, only compressed data (ex. fire or no-fire, or flood or no-flood) is broadcast.

Also we plan to hold contests to maximize the accuracy of an algorithm and to optimize its execution on a smartphone. This prize will help fund those activities.

N/A

N/A