Identify input requirements for initial geocoders for address push - UW, PostGIS, Degauss, Nominatim

[rtmill]

Identify input requirements for initial geocoders for address push - UW, PostGIS, Degauss, Nominatim

[tibbben]

I have some detailed notes on my effort to stand this up in a containerized local environment ... Kubernetes. Please reach out: [email protected].

[kzollove]

- ID input requirements for geocoder testing
	○ Jim's analyst is running these tests, modified DeGauss, all testing in a jupyter notebook
	○ Benchmarking project: Where do these geocoders perform well, where they are off, where they totally fail
- Test dataset for geocoding:
	○ 104K public service locations, ~30K rural, ~3K Rural-Tribal and 26K non-tribal Rural
	○ Next step: parsing of address strings for "junk" addresses… will keep them in as negative control.
	○ How do we have an address standardization software
		§ Address standardization: formatting, and tests is this a real address
		§ Tim: no one address standardizer will work for all four of those geocoders
			□ The PostGIS geocoder has a standardizer in it
			□ Wrote one for Nominatim
			□ TODO (where do we start this"grid"/catalog of standardizers) :Andrew suggests a catalog of standardizers to suit everyone's needs, at least start collecting them, understanding their performance, but not fully described on our end.
		§ This is a pre-geocoder workflow step… standardized and then fed in to the geocoder.
		§ The three geocoders do not have an ingrained standardizer
		§ ArcGIS streetmap was recommended… not OSS, not free, requires license.
	○ Cybersecurity vulnerability assessment of the geocoders:
		§ Each are deployed on docker, which is fine
		§ However, the docker dependencies are >5years old and have some vulnerability elements
		§ Will reach out to Cole Brokamp (DeGauss), who to reach out to for Nominatim?
		§ Jimmy thinks they will maintain docker image locally and then pass over to Palantir
		§ Have decided on the upgraded version of the linux for the image, then will pass it back to Palantir
		§ CLAD project goal is the get the best geocode across any location, using multiple geocoders… the more geocoders the harder to maintain
		§ Tim suggests: If youre ready to pay for ArcGIS, go there. If not use OSM.

Tims approach is to do whatever is possible with OSM and then fall back on ArcGIS

[jaygee-on-github]

@kzollove, @jphuong, @AEW0330: it looks like from Jim's presentation and Kyle's write up that longitude and latitude will have a provenance. Where are you thinking we might capture this provenance? It could be both a workflow and its execution environment(s). Given Andrew's suggestion that we might use RO-Crate (which describes provenance and workflows on top of schema.org and JSON-LD), where might these knowledge graphs go? And will we be able to author them with the same authoring tool we are exploring at #324?