Long Beach Airport, about 1928. Photo: Winstead / Long Beach Public Library
A map for seeing how Southern California came to be, with 1920s aerial photography laid over the region as it is today.
My background is in theater design, where you learn to take a production apart by working backward from the effect. What pieces, cues, and changes over time made it work? I wanted to look at Los Angeles the same way and understand how the city I know came to be.
I built it for myself first. I also had policy researchers in mind: people studying how transit shaped development, or why neighborhoods ended up with such different outcomes.
UCSB's FrameFinder holds the historic aerial archive, along with the flight and camera details needed to correct the images. But each photo is a point on a generic map. You click it, load one image, and compare frames one at a time. I wanted to see the connections directly, across a whole area at once.

The first person I watched use it lit up at the comparison: "Oh, this is it. Now it makes sense." Search and directions interested them too, but the comparison got the biggest reaction.
It also had the least visible way in: a small icon with no label. After that session I added a "Compare" label to the button. On phones the toolbar has no room for the word, so there it is still the icon.
The hardest problem was alignment. Dragging photo corners into place left small errors I couldn’t fix by hand, and they compounded. Each photo had been bent to agree with its neighbors, so whole layers sat 10 to 40 meters off the streets they showed, and two 1927 layers over Beverly Hills disagreed by as much as 95 meters. So I borrowed methods from my hyperspectral camera project. Instead of warping each photo to fit, I modeled the camera that took it: its position, height, heading and tilt, focal length and lens distortion. Thousands of points shared between overlapping prints tied each flight together so it could be solved as one piece. Every print was held rigid, because letting them stretch let photos drift 65 to 130 meters. Each photo was then projected onto a USGS elevation model so hillsides land where they belong.

To pin the flights to the ground, I matched them against today’s streets. Each photo was rendered from its current camera and compared, patch by patch, with OpenStreetMap street outlines and satellite imagery. Every match became a control point, weighted by how far it could be trusted. The highest weight went where the street map and the satellite agreed on a major road or a rail line, since those section-line roads mostly existed in the 1920s. Ordinary streets counted for less, then field lines and canals, then a lone satellite match, and anything near water least of all. After each solve I measured again from the improved position, because matches taken from a bad start lock onto the wrong street. On the 1938 flight, 93 percent of the re-measured points agreed with the first solve within 10 meters.

Streets only work where streets survived. On 1920s farmland, dunes and marsh, street matches were noise, and so were they at Irvine, where today’s grid has nothing to do with the 1931 ranch roads. One 1946 print read 330 meters off by its street matches, yet it sat within a meter of the 1938 photos. In those places I anchored new photos to blocks I already trusted, and to fixed landmarks like the ends of piers. The piers caught an error nothing else would. The first placement of a 25-kilometer strip of the 1927 coast, bent by its lens, sat 350 to 470 meters off at Huntington Beach, Newport and Balboa, yet it looked right along the shoreline. Each run was scored on check points left out of the fit, usually 2 to 4 meters off. Then I checked it by eye: I drew today’s street outlines over the photos and laid overlapping layers in red and cyan, where any offset shows as colored fringes.


Exposure was its own problem. Every print is brighter where the sun’s hot spot falls and darker toward the corners. On the 1931 flight the centers ran about 1.2 times the average and the corners 0.4 to 0.7. I measured that falloff from the 16-bit scans, so the bright centers wouldn’t clip to white. On the 1947 flight each print got its own tilt, because the sunny side flips with the flight direction. Brightness was matched on land only, since sun glitter on the ocean changes from photo to photo. On the 1947 layer, that cut the typical brightness step at photo joins from 13 grey levels to about 5.