(I've been geocoding my photos since 2004.)
Graffiti complaints in San Francisco over 90 days.
There's a trade-off between seeing the precise location of each data point, and seeing the aggregate regional blobs that make trends easier to see. Here I did both and composited one on top of the other using ImageMagick.heatmap.py -b black -p graffiti.coords -r 30 -W 1200 -o g1.png -P equirectangular ;
Data from datasf.org. I don't know what system the coordinates are in, so I just pretended they're square and mapped them as-is.
heatmap.py -g orkney.gpx -o orkney.png --height 800 --osm
heatmap.py --decay 0.8 -g bman.gpx -o bman.png --height 800 --margin 10 -b black --gradient gradient-1.2.png
(I carried a GPS around all week.)
There are a few kinds of heat maps. This program is for when you have data points, each with a lat-lon coordinate, and you want to plot them on a map such that they blob together and get bright where they pile up, showing off the data density.
If you've divided your map into regions and want each region colored a certain way (a choropleth map), or you have a formula that gives you a value for every pixel on the map, this is probably not the right tool.
It's just one file: heatmap.py. I've put it on github in case anyone wants to contribute. If you want the --osm option to work, you'll also need osmviz and should read the bit about OSM licensing issues.
Now get yourself some data (or grab my list of photo coordinates) and just try it out. See the examples on this page, read the command line help for a very brief rundown on the options, and maybe eventually read the rest of this page. In general, the program will tell you if needs more information.
I wanted to make some data maps and I couldn't find any free tools that did a good job. I found proprietary tools (ick) that produced gorgeous output, and free tools that I felt took the wrong approach. Had I discovered gheat sooner, I probably would have started with that, but this is a bit different, better for some things, worse for others. There are some obvious next steps, but it's time to set it free.
It's a Python script, which makes it most suitable for server-side image generation or preparing images in the comfort of your own home. That's great for my needs, but doing the work in a browser would be a better choice for many applications these days.
How long it takes is roughly proportional to
num_points * radius2,
so if you want to try things out quickly, start with a small radius,
and/or less data.
The memory footprint depends on the settings. With --decay set
between 0 and 1, the memory use will be dominated by
num_points * radius2. With decay
set to 0 or 1, the rendering only needs to hold data proportional to
the size of the output image, which is usually negligible. At that
point, the major factor will be whether it needs to read in all the
data at once.
Currently it needs to know the geographic extent of the data at the
outset, so if you don't specify one using --extent, all the data will
be read first, and then held in memory for the rest of the processing.
So memory use will be proportional to the size of your data set.
If you use --decay of 0 or 1 and use --extent and give either the height
or width (or --osm and --zoom), then the memory footprint won't depend
on the input significantly and you should be able to push any amount of
data through. (I tried 10 million points, worked great.)
It's to make it prettier and to make it convey your information better. If you leave it at 1, it works like most people's heatmaps do, just adding points on top of each other linearly. Depending on your data distribution, this may mean areas with relatively low density are hard to see. The highest concentrations stand out, but the dimmer areas are nearly invisible.
If you set it to 0, every data point will be visible, turning the plot into "how far is it to the nearest point" and conveying nothing at all about distibution density. It will look more like a scatter plot (or a line drawing, if you have GPS tracks).
All the other heatmap software I've looked at works at one of these extremes (usually the former), but I find it looks better and conveys more information somewhere in between. I was unable to find a magic value that works well for everything, so you may want to fiddle with this. There's no right answer. You just have to decide what makes sense for your application and your data.
The catch is that when the decay is 1 or 0, the heatmap can be computed using much less memory. if it's anything in between (as it is by default), the memory footprint is considerably greater.
You can specify the color gradient on the command line by providing hue, saturation, value, and alpha for the two extremes. The program will use linear interpolation in HSV space to render the in-between values. Note that hue is specified with three digits in case you want to go around the other way on the color wheel. For example, 000ffff00 and 100ffff00 specify the same color, but pass through different hues on the way to 02affffff.
The other option is to supply an image. The pixels down the left edge are used to specify the gradient. (For example.)
If you use the --osm option to overlay your data onto OpenStreetMap tiles, you are bound by their terms of service.
Here's what you get when you use --help.
Usage: heatmap.py [options] Options: --version show program's version number and exit -h, --help show this help message and exit -g FILE, --gpx=FILE -p FILE, --points=FILE File containing one space-separated coordinate pair per line, with optional point value as third term. --csv=FILE File containing one comma-separated coordinate pair per line, the rest of the line is ignored. --ignore_csv_header Ignore first line of CSV input file. --shp_file=FILE ESRI Shapefile containing the points. -s FLOAT, --scale=FLOAT meters per pixel, approximate -W INT, --width=INT width of output image -H INT, --height=INT height of output image -P NAME, --projection=NAME choices: mercator, equirectangular; default: mercator -e RANGE, --extent=RANGE Clip results to RANGE, which is specified as lat1,lon1,lat2,lon2; (for square mercator: -85.0511,-180,85.0511,180) -R INT, --margin=INT Try to keep data at least this many pixels away from image border. -r INT, --radius=INT pixel radius of point blobs; default: 5 -d FLOAT, --decay=FLOAT float in [0,1]; Larger values give more weight to data magnitude. Smaller values are more democratic. default:0.95 -S FILE, --save=FILE save processed data to FILE -L FILE, --load=FILE load processed data from FILE -o FILE, --output=FILE name of output file (image or video) -a, --animate Make an animation instead of a static image --frequency=FREQUENCY input points per animation frame; default: 1 --straggler_threshold=STRAGGLER_THRESHOLD add one more animation frame if >= this many inputs remain -F STR, --ffmpegopts=STR extra options to pass to ffmpeg when making an animation -K, --keepframes keep intermediate images after creating an animation -b COLOR, --background=COLOR composite onto this background (color name or #rrggbb) -I FILE, --background_image=FILE composite onto this image -B NUM, --background_brightness=NUM Multiply each pixel in background image by this. -m HEX, --hsva_min=HEX hhhssvvaa hex for minimum data values; default: 000ffff00 -M HEX, --hsva_max=HEX hhhssvvaa hex for maximum data values; default: 02affffff -G FILE, --gradient=FILE Take color gradient from this the first column of pixels in this image. Overrides -m and -M. -k KERNEL, --kernel=KERNEL Kernel to use for the falling-off function; choices: linear, gaussian; default: linear --osm Composite onto OpenStreetMap tiles --osm_base=URL Base URL for map tiles; default http://tile.openstreetmap.org -z ZOOM, --zoom=ZOOM Zoom level for OSM; 0 (the default) means autozoom -v, --verbose --debug