Temporal accuracy validator

[geohacker]

@awright @smit1678 do you have a sense of what we can define as our recency benchmark? For instance, how old is too old/stale data for the malaria campaign?

We can build a generic OSMLint validator that takes a timestamp and gives out all the buildings that are older or newer than that point in time.

[awright]

Good question. Defer to @smit1678, but these "grades" from the CHAI presentations, might be useful. (And then somehow this became a ticket full of screenshots)

Of the 5 countries, Temporal Accuracy was only measured in 2.

Botswana

More Detailed Deck

Namibia

More Detailed Deck

[smit1678]

I think temporal accuracy isn't just the timestamp for when it was added to OSM, but needs to capture somehow the temporal accuracy of the data sourced used to generate it.

Here are a couple ways I started thinking about temporal accuracy:

1. when was the building first added to OSM and then when was the building last edited? For example, some of the data was first remotely added to Botswana in 2016, but then attribute information will be added to the buildings in 2018.
2. if the building was remotely enumerated with satellite imagery at all. So one check would be: what was the source of the data and was imagery used?
3. if the building was recently added but used really old imagery (say 10+ years old). This one is harder to get because only Bing and OAM report imagery dates per tile at the moment that I know of.

In short, I think timestamp is the first check (satisfies part of point 1), but then we need another check that helps further define temporal accuracy. Do we also want to add a check for if imagery was used when the building was added?

[geohacker]

@smit1678 @awright - thank you! This is helpful.

some of the data was first remotely added to Botswana in 2016, but then attribute information will be added to the buildings in 2018.

This is going to be really hard to do at the moment - we don't have a reliable history query infrastructure (I don't want to advocate building one more tool based on Overpass until we know what we'll do to get around the pitfalls.)

In short, I think timestamp is the first check (satisfies part of point 1), but then we need another check that helps further define temporal accuracy.

Looks like this is a good first pass. We'll be able visualise the recency of buildings in the report. We can perhaps show the satellite imagery based on the source tag, but that wouldn't have a timestamp associated.

[awright]

@geohacker cool. This works for me. @ascalamogna good to close the ticket and question then?

[geohacker]

@awright great!

Let's close after we ship the osmlint/vectortile piece of this.

[kamicut]

@tyrasd since this is a very similar approach to what OSMA already does, I was wondering if we could get the tiles already processed there for our target AOI and calculate a similar chart on the frontend of osma-health. What do you think?

[smit1678]

@kamicut Should we be concerned with the risk of being out of sync if one is ahead of the other? Also, if the longer term goal is to expand beyond most recent edit with more history information, do we lose those goals by using the other osma tiles?

[kamicut]

We are going to use the vector tiles from the building filter because they contain timestamps. I'm closing this issue because we will start with edit recency @smit1678 using the timestamp and then we can create a new issue if we want to expand this.