Use FB's High-Res Pop Density Maps for finer-grain spatial population estimates (30m2 gridded)

[daveluo]

We currently have population data at the US county level from the US census, estimated in 2018, and broken out by demographics according to their definitions: https://www2.census.gov/programs-surveys/popest/technical-documentation/file-layouts/2010-2018/cc-est2018-alldata.pdf

If we want higher spatial resolution (30m x 30m gridded) population density estimates and break out demographics a different way, consider the open data (CC 4.0) here: https://registry.opendata.aws/dataforgood-fb-hrsl/

Another benefit is that this is standardized data for most countries of the world to facilitate if/when we or others want to apply our US analysis and workflow to new countries.

Initial notes from https://gitter.im/covid19-healthsystemcapacity/community?at=5e71167697371d57b58758bb:

later on, could add https://registry.opendata.aws/dataforgood-fb-hrsl/ which has finer-grain demographic breakouts (i.e. population over 60 of age) and higher spatial res than census tracts also easier to scale worldwide (not US-specific definitions or boundaries)

presumably, areas with higher proportion of elderly are at highest risk of being critically ill (going into ICUs) if infected due to age and comorbidities, and have poorer outcomes

so there could be disproportionate ICU demand coming from these finer-grained spatial areas where there's a lot of elderly (especially if there's an outbreak in a facility or community i.e. Kirkland nursing home in WA), this potential demand may get missed when population is spatially aggregated to county level

not necessary to do this on the first cut, but something to keep in mind for later

[daveluo]

Direct links to data by country: https://data.humdata.org/search?q=High%20Resolution%20Population%20Density%20%2B%20Demographic%20Estimates&ext_page_size=100&sort=score%20desc%2C%20if(gt(last_modified%2Creview_date)%2Clast_modified%2Creview_date)%20desc

[echeipesh]

• Data: HRPD USA is on s3://azavea-humdata/facebook/hrpd/ (requester pays)
• Process: Summaries were generated with https://github.com/echeipesh/geotrellis-hrpd
• Result: us-county-hrpd-v1.csv
• Verification: notebook

Verification is very light, I poked around lightly comparing the results to 2018 census and they appear plausible. I'd check for holes and large discrepancies before running with this.

The notebook can be copied into notebooks folder in this project and CSV file can go into data/external. If this becomes of use or there are questions/problems they can be refined further.