Can Twitter Accurately Document Flooding Incidents?

Bright Planet has explored whether social media could "help in a flood", especially in the context of reviewing tweet data for insurance claims.  Here's what they did in relation to the July 8, 2013 Toronto storm and flood:

"To assist with the vetting and triage of insurance claims, BrightPlanet harvested all tweets within the Greater Toronto Area. BrightPlanet then filtered and curated the tweets down to only tweets discussing specific effects of the flooding. Any tweets containing a latitude and longitude were then mapped in a heat map format in Toronto to show where the most chatter was happening about the flooding."

See the full description on their page here: BrightPlanet.  The harvesed image is here:

There are a lot of tweets downtown, mostly south of Bloor Street. Matthew Dance analyzed tweets about Toronto - not about flooding, and produced this cluster map:

Of course the extents are different, but I'd suggest that we see the same pattern in BrightPlanet "flood tweets" as we do for Matthew Dance's "generic Toronto tweets".  Dance explains tweet density in a few ways:

  • "The most densely Tweeted area is bounded by Bloor Street to the North and Lake Ontario to the South, connected by Young Street.  There is a greater density along the Lake, away from Young to the West."
  • Hot spots include the Eaton Center, Rogers Center and tourist and suburban destinations - the areas around Yonge and Bloor and Front Street, including the sports stadiums - "destinations for those interested in shopping or taking in the sights in Toronto."
  • Areas that are strictly neighbourhoods (Hillsdale Avenue running east from Yonge) have low density.
  • Poorer neighbourhoods like Regent Park have few tweets (between Dundas and Gerrard west of the Don Valley)

Where does the City of Toronto report flooding for July 8 2013? It does not match the BrightPlanet flood tweet locations very well as shown below:

Tweet-rich downtown had only 50 mm of rain and limited reported flooding, compared to 130 mm of rain in tweet-poor Etobicoke where most of the flooding was reported.  Perhaps the BrightPlanet flood tweets should first be "normalized" by the density of generic non-flood tweets, to give a better representation of relative spatial flood activity?

It would also be worth checking if what people tweet more about downtown is spectacular surface flooding (e.g., a sewer geyser blowing a manhole 10 feet in the air, a Ferrari floating in an underpass lake, an amphibious GO Train) as opposed to basement flooding, which is the focus of City of Toronto flood complaints.  Tourist and passersby have time to tweet about flooding, while a homeowner with sewage gurgling up their basement floor drain may have more pressing thing to do (save the family photos honey!).