I think a long representation of time and space is best with ancillary geometry table.
Implementation consideration:
Attribute of all used geometry as an sf object with a column for geometry identifier and the respective geometry. The geometry attribute should contain neighbors and weights (potentially stored as a listed object for spdep compatibility)
The main table contains a column for the geometry identifier, the time period, and any other values that would be desired.
All time periods must have have all regions.
If there are 2 time periods with 10 regions we would have 2*10 = 20 observation in our table.
Users can then create any lisas they'd like and then summarize with Mann Kendall to identify changing temporal patterns.
There is an older implementation of this from edzer pebesma
https://cran.r-project.org/web/packages/spacetime/vignettes/jss816.pdf
I think a long representation of time and space is best with ancillary geometry table.
Implementation consideration:
Attribute of all used geometry as an sf object with a column for geometry identifier and the respective geometry. The geometry attribute should contain neighbors and weights (potentially stored as a listed object for spdep compatibility)
The main table contains a column for the geometry identifier, the time period, and any other values that would be desired.
All time periods must have have all regions.
If there are 2 time periods with 10 regions we would have 2*10 = 20 observation in our table.
Users can then create any lisas they'd like and then summarize with Mann Kendall to identify changing temporal patterns.