We got a link from a friend about a "Homeless Heat Map" he helped create to track homeless populations in downtown LA. Here is an image of the map, and his notes on it:
From George Mokray:
Back on December 5th Cartifact launched the Downtown Homeless Map and
I got to write about the process here. The project takes LAPD Central
Division's bi-weekly homeless counts and turns the data into a map,
visually telling the story of changes in Downtown's street population.
Today we've put online a new version of the maps, using a radically
different methodology for showing the data. Instead of the dots of
the old maps, this version takes the data and turns it into a "heat
map" that shows the density of the population in different areas.
More about the new style after the jump...
Interesting to note, though, is the way in which temperature affects
the number of people on the street. It's cold outside, and has been
for several days now. The count for January 15th (Monday) was down
271 people from January 2nd. It got cold and the people who could
find somewhere to go did so.
I think the main thing this new style brings is a more instant
understanding of what's going on. The dots made an interesting
picture, and one that did work to tell the story, but in the end they
generated a lot of questions. Real world data collection inevitably
means compromises in your methodology, and in this case it led to
confusing results like dots showing up on top of other dots.
Aside from just looking cool, the heat map was a tecnically
interesting thing to create. The process involved taking irregular
point data and generating an approximated surface from it. That
surface data was then brought back into the GIS and the statistical
models were tweaked this way and that until they generated something
that felt true to the situation on the streets.
Update (9pm): A little on the technical side...
To generate the approximated surface I'm using surfit, an open-source
gridding application. Initially I was struggling because the data I
get only has positive points -- there are no zero points to bring the
elevation back to the plane in areas where no homeless were counted.
I eventually figured out how to normalize the computation against a
flat area I set up that covers Downtown.
Once I have the surface grid from surfit I use VTBuilder from the
Virtual Terrain Project to georeference that data and clean out
really low data (elevations less than 0.5 or so). VTBuilder outputs
an Arc ASCII GRD file and I pull that back into ArcGIS.
The color ramp is applied against a set of baseline statistics that
don't change count to count.
View the map for yourself here.