Light level geolocation (LLG) is based on what I called "Australian paradox" when I was 8 years old - like, they have long summer days while we have short winter days. Being on different hemisphere than kangaroos make the difference very obvious, but smaller changes in latitude do have also impact on day length; though with smaller granularity. Changes of longitude are apparent as time zones. By measuring time of dawn and dusk against precise clock and date, one can determine the longitude (when dawn or dusk happens) as well as latitude (day length - difference between dawn and dusk).

This looks trivial with any light sensor, like photodiode or light dependent resistor and real-time clock, but devil is in details, as usual.

1. This way of geolocation is slow. GPS can get fix in order of seconds or minutes, LLG in order of days, no less than half a day. There is nothing we can do about it.
2. This way of geolocation is also not exactly accurate. GPS has accuracy of meters, LLG in kilometers, up to dozens of kilometers. There is something we can do about it - like having clear view to sky and not much artificial light pollution.

While GPS looks like no-brainer decision here, LLG offers some advantages too:

1. It can be really cheap and low power. It can run and log for years from single coin cell battery.
2. No other advantages, but having just one point in this list seemed inadequate.

These points made LLG to enter its use for animal tracking (birds, sea animals), where usually one location per day is enough and the accuracy is fine to observe migration across continent.

There are some scientific papers available, but as usual behind the paywall (thanks to Sci-hub one can retrieve it and take a look) and are more like information of what can be done, with little to no secret sauce of how to do it. LLG doesn't seem to be very popular outside bird tracking circles, though it has some potential as low-power backup tracking autonomous vehicles or ships, for example. I believe it can be interesting to try to make open-source implementation of this method.