Generate map based visualizations

[brunosan]

OSRM can create isochrone polylines, but QGIS makes a better work. We should find a way to generate these on the UI

[olafveerman]

We have started looking into map based visualizations of the ETA data generated by RRA. To experiment and see what works well, we set up a playground that uses results from three projects from the staging environment.

The challenge will be to design a visualization that works well across projects with very different characteristics. The most important ones that will drive the design choices are:

• variation of ETA - eg. Argentina has some very big outliers
• variation of population - the population per origin is more evenly distributed in Argentina than Morocco
• density of the origins - the origins on some projects are a lot more disperse than others

At this moment, the playground styles each point of origin according to their ETA (color) and population (radius). We've added some options to the map that allow you to change the scale. This will not be a feature of the final visualization, but allows us to make better informed design decisions:

• bucket - ETA and population results are bucketed. The advantage is that the user can easily related to the results, and that we can match them with the buckets that will be used in the charts (10, 20, 30, 60 minutes, etc). The disadvantage is that buckets that work for one scenario, may not be appropriate for the other.
• normalized - ETA and population are normalized on a 0-1 scale.

Next steps

Over the next couple of days we will be doing some more experiments and working on the style. In the meantime, your feedback would be greatly appreciated. Specifically:

• does this visualization provide you with better insight into the accessibility of a region? Is it helpful to communicate results to the people you work with?
• are there other types of visualizations that we should be looking at?

cc @Holly-Transport @qli1205 @xijielu @stvno @brunosan

[brunosan]

• I might too geeky, but a log10 would work well for me. It gives detail close by but also far away.

• Another option would be using capped percentiles (color range from 5th-95th so you get a color change for most of the data, and you also account for outliers. I wouldn't use too many color classes.

• Human natural breaks could work too: e.g. for Hospitals, schools, ... I'm sure there is a relation of effectiveness with distance in time. ... No idea if those numbers exist or case-base color coding is doable.

For the perspective of the user, I can imagine the money is on the difference of ETA before/after the planned change.