Gaining accurate situational awareness after a disaster remains a big challenge in any crisis. While we have more data than ever before, it is almost inevitable that disaster responders will get overwhelmed by the amount of data and struggle to interpret it.
How do we know, which pieces of information are relevant and new? Which ones are outdated or repetitions? What is the big picture? Where should we focus our efforts?
Many emergency response organizations try to answer these questions by creating dashboards that combine different streams of information to help disaster managers make decisions.
The UN’s innovation initiative, UN Global Pulse, is currently working on a prototype of their version of such a dashboard called MIND (Managing Information for Natural Disasters).
MIND combines a number of data streams, such as automated information from GDACS, OpenStreetMap, a logistics database with road information, HDX, Twitter, Google Trends, Wikipedia and a news-API on a dashboard.
Users can easily find information such as a frequency analysis of Twitter terms, calculate travel times or see what casualty figures the media are reporting.
UN Global Pulse published a thorough description of the project on Medium, which I recommend you read, if you want all the details. However, I still had some questions after reading the article and Faizal Thamrin, Humanitarian Data Advisor at the Pulse Lab Jakarta, was kind enough to talk to me.
Here are my thoughts following the conversation:
First and foremost, I believe that this tool can be very valuable in emergencies. There is a need for this kind of fully automated data analysis. I also recognize that it is currently only a prototype and issues around non-English languages and alphabets have not been resolved.
As for the different integrations:
GDACS: MIND automatically starts collecting information for all disasters marked red or orange on GDACS. This means that nobody has to manually decide which disasters warrant data collection.
Twitter: I have mixed feelings about using geotagged tweets for frequency analysis, since the location where a tweet is sent from might be very different from the location a tweet is describing. This happens, for example, when relatives outside the affected zone convey information from within the affected area, for example because people inside the affected area don’t have internet access anymore, but can still text their relatives elsewhere. This is quite common and might lead to information being missed. There is also the general problem of relying on geotagged tweets, since in many countries only a very small number of tweets has geographic metadata (globally, only 1% of tweets are geotagged, but in some countries, like Indonesia, the percentage is much higher).
On the other hand, I rather like their approach of not trying to predict which words and phrases affected people are going to use to describe their situation. As MIND is not searching for terms, but looking at frequencies of anything within the geographic boundary of the disaster, I think it’s plausible that you might find more unexpected information in the data – and that is a good thing. Users can see individual tweets when drilling down inside the frequency analysis.
Google Trends: This is easily my favourite feature: MIND can show you the most frequently googled terms in a disaster-affected area. I love this, because it is probably a more accurate representation of needs and issues than geotagged tweets. Primarily, because more people are using Google than Twitter, but also because the percentage of geotagged Google queries is much higher than the percentage of geotagged tweets. While Google queries cannot actively convey information through content in the same way a tweet can – i.e. you won’t find an image of a collapsed bridge in someone’s Google search query – Google Trends will probably show you more accurately, which topics are really on people’s minds.
Casualty Count: MIND analyses new articles to see which casualty numbers the media are reporting. I have to admit that I’m a bit puzzled by this. In my experience the media get casualty numbers from the disaster management authorities, not the other way round. But I guess that depends on the country and it can alert you to something being wrong if the media report dramatically different numbers from your own. I was also curious about how the platform deals with articles that compare casualty numbers for the current disaster with previous ones (e.g. “1,000 people died, compared to the 3,000 people who perished in the same area in 1982”), but apparently the team has taken that into account.
MIND has more features than the ones I discussed above; take a look at the Medium post if you want more details.
The MIND team is still looking for organizations to help them test the platform. The current plan is to make the platform publicly available in the fourth quarter of 2019.
What are your thoughts? Please leave a comment below!