osmod

Estimating origin-destination flows from OSM data

TODO List

1. Choose a test location - Briston, U.K.?

2. Use osmdata to extract all building polygons and building heights where given. For residential buildings:

   • Construct some kind of simple statistical model to estimate heights of buildings lacking OSM heights - Polygon area times height ~ residential population density, with census data as ground truth and calibration.
   • Use global population density data to down-sample to OSM building polygons.

   For commercial and industrial buildings:

   • Repeat same procedure for density of employments (likely using different models for commercial and industrial).
   • Think of a smarter way to estimate employment densities?

3. Define a distance decay function for proportion of people that cycle to work as a function of distance to workplace.

4. Define some kind of function defining for a residential location some distance from the city centre the probability of that person working in some location a specific distance from both the city centre and their residential location. This will be some kind of two-parameter distance decay function.

5. Use output of Step#4 to route population density through the street network using a probabilistic router to convert that to relative densities of cycle traffic. This can be done separately for travel both to and from work.

Refinements

1. Include travel not related to work, through connecting residential densities with leisure and commercial locations, to reflect recreation and shopping activities.
2. Improve the model suggested in Step#4 to reflect not necessarily radial urban structure.
3. Use google earth engine to automatically identify bicycles - that could well be possible? - and individual people on pavements. Feed that in as an extra parameter to reflect the possibility that those locations where more people are out and about without cars are also more likely to see more cycling activity.
4. Several spatially-defined variables will ultimately be able to be both automatically and globally derived, and the analysis could likely be fed into a neural network rather than hard-coding a fixed statistical algorithm. OSMODMAI? ODOSMAI?