"How should we treat sub-country urbanisation projections? Right now we use the national value for sub-country analyses."
After a bit of finagling I started googling. There is a new dataset:
GHS-WUP-DUC R2025A
https://human-settlement.emergency.copernicus.eu/ghs_wup_duc_r2025a.php
GADM 4.1 levels 0–2, 1975–2100 at 5-year steps.
It is however aligned to the UN WPP2024 not SSPs, so we still need to translate this new dataset into usable SSP data.
In any case back to the problem at hand: If we consider Kampala, its district run from 14 to 100% urban, clearly, a single "national" urbanicity factor will not represent Kampala''s district well.
How then to keep the spatial pattern.
The most simple fix is to fix urbanicity at baseline, but that doesn't allow for scenario's. Perhaps we could use it as a last resort backup. Wat we do know is the national urbanicity + the constraint that districts pop weighed average must equal national total.
Then the next simplest fix is to have each subnantional entity to follow in the national entities footsteps, but then we throw out "urban" growth due to clamping. We should therefore bound growth, bound it between 0 and 1 but not resort to clamping. An odds approach could work here (which are bound between 0 and 1), which is also called iterative proportional fitting.
see
https://cran.r-project.org/web/packages/ipfr/vignettes/using_ipfr.html
https://en.wikipedia.org/wiki/Iterative_proportional_fitting
https://pypi.org/project/ipfn/
lets think trough this. Lets take Uganda's least urban district, which is 13.9% urban. It odds of being urban are 0.139 / (1 - 0.139) = 0.161, which is about ~1 urban person per 6 rural persons. we need to figure out by what number to multiply those odds, which is what iterative proportional fitting is for. That does require several attempts at finding the correct factor. For Uganda, this factor is 3.71 by 2050. We multiply the odds 3.71 * 0.161 = 0.6 and then calculate back what the share should be, which is B / (B - 1) = 0.6, where B = 0.6 / (1 + 0.6) = 0.375, and we therefore have a 37% share of urban. This works for urban and non urban alike, and none will grow beyond 1. Nor will the total grow beyond the national total by construction.
We can use this to scale into the future baseline local urbanicity using predicted national urbanicity.
"How should we treat sub-country urbanisation projections? Right now we use the national value for sub-country analyses."
After a bit of finagling I started googling. There is a new dataset:
GHS-WUP-DUC R2025A
https://human-settlement.emergency.copernicus.eu/ghs_wup_duc_r2025a.php
GADM 4.1 levels 0–2, 1975–2100 at 5-year steps.
It is however aligned to the UN WPP2024 not SSPs, so we still need to translate this new dataset into usable SSP data.
In any case back to the problem at hand: If we consider Kampala, its district run from 14 to 100% urban, clearly, a single "national" urbanicity factor will not represent Kampala''s district well.
How then to keep the spatial pattern.
The most simple fix is to fix urbanicity at baseline, but that doesn't allow for scenario's. Perhaps we could use it as a last resort backup. Wat we do know is the national urbanicity + the constraint that districts pop weighed average must equal national total.
Then the next simplest fix is to have each subnantional entity to follow in the national entities footsteps, but then we throw out "urban" growth due to clamping. We should therefore bound growth, bound it between 0 and 1 but not resort to clamping. An odds approach could work here (which are bound between 0 and 1), which is also called iterative proportional fitting.
see
https://cran.r-project.org/web/packages/ipfr/vignettes/using_ipfr.html
https://en.wikipedia.org/wiki/Iterative_proportional_fitting
https://pypi.org/project/ipfn/
lets think trough this. Lets take Uganda's least urban district, which is 13.9% urban. It odds of being urban are 0.139 / (1 - 0.139) = 0.161, which is about ~1 urban person per 6 rural persons. we need to figure out by what number to multiply those odds, which is what iterative proportional fitting is for. That does require several attempts at finding the correct factor. For Uganda, this factor is 3.71 by 2050. We multiply the odds 3.71 * 0.161 = 0.6 and then calculate back what the share should be, which is B / (B - 1) = 0.6, where B = 0.6 / (1 + 0.6) = 0.375, and we therefore have a 37% share of urban. This works for urban and non urban alike, and none will grow beyond 1. Nor will the total grow beyond the national total by construction.
We can use this to scale into the future baseline local urbanicity using predicted national urbanicity.