The Eudat algorithm was designed while working for UbiEst around 2003-2004. It produces discrete population density maps indirectly from punctual data and nearby road graph analysis.
“Population density” information as provided from international vendors mostly comes in two forms: loose shapes (usually isolines) providing a rough population density for the area, or punctual data (usually related to cities as a whole) describing a density and maybe a radius.
Punctual data however can be reconstructed by using both existing data and road graphs from international GIS vendors such as NavTeq. The immediate advantage of such approach is very fine granularity (usually below 1 square kilometer) and quarterly updated content.
In eudat, the existing punctual data is first augmented by distributing the population count and position onto the real underlying road graph (using information such as civic numbers, road length, administrative status, etc). Then, punctual data is discretized, a statistical model is built and used to balance under-covered areas. This process brings the initial 190k to almost 1 million of samples just for the western Europe.
Unfortunately this method is still not perfect and shows several artifacts:
Most of these issues could probably be worked around by keeping a manual list of adjustments to be made when the statistical model is applied, but doing so required too much manpower than what we had at the time. As such, the approach has been abandoned, leaving us with the pretty pictures.