Metodological viability: Downscaling population projections from coarse administrative units to grid level for Urban Nature Access model

Software Support
1

Hello everyone,

I am currently conducting research on socio-environmental justice and the distribution of urban ecosystem services. I am using the Urban Nature Access model to analyze future scenarios for 2030 and 2035.

I have encountered a challenge regarding data resolution: official population projections for those years are only available at a macro-administrative level (large districts/localities). However, my LULC (Land Use/Land Cover) maps and the required analysis for the model demand a much finer spatial grain (at the grid/pixel or block level), which is unavailable for future dates.

I would like to ask the community’s opinion on the following:

  1. Is it methodologically sound within the InVEST framework to distribute the projected macro-unit population proportionally, based on the current population density of the LULC pixels?

  2. If I assume that the built-up areas in my 2030/2035 LULC maps will absorb this growth, is a dasymetric mapping approach (weighting population distribution by urban land-use classes) considered a robust proxy for the model’s demand requirements?

  3. Are there any known statistical biases or “red flags” I should be aware of when performing this type of downscaling from large administrative units to a high-resolution grid for future scenarios?

I want to ensure that the “demand” side of the model remains consistent with the “supply” side, even if the population data starts at a much coarser scale than the biophysical data.

Thank you very much for your guidance and suggestions.

2

The WorldPop released a new dataset with annual population projection to 2030. Though 2035 is not covered but you might find ways to adapt their approach for further years WorldPop :: Population Counts

3

Hi.

Yes, we have done a lot of population downscaling/disaggregating. A dasymetric mapping approach is a robust approach and that’s what we used in this paper to map population from the census district level. We’ve also used the NOAA nighttime lights dataset to map population at a grid level (based on light intensity) when we had larger census/administrative districts. As Yingjie said there’s also the gridded modeled pop’n datasets, these datasets perform fairly well and are another option - but I know in many cases people want to use their local census data.

With that said, and I’m not sure I would consider this a red flag per se, but more a concern you should consider is the level of precision you need to have for your analysis in an urban environment which is obv way more heterogeneous in terms of population density than a LULC is going to be able to capture categorically. Sometimes there’s building footprint data that could be used to even further refine a dasymetric mapping approach.

So in summary yes these pop’n mapping approaches are robust and we’ve used them in many different applications, but I would do a little reading about how folks do it in cities specifically to see if you can get some guidance.

4

Thank you for pointing me to the WorldPop annual population projections up to 2030. I was not aware of this recently updated dataset, and it is indeed a promising alternative to performing full downscaling from coarse administrative units from scratch

Your suggestion provides a robust benchmark for my own downscaling exercise and saves considerable effort.

5

I am especially grateful for your detailed response, as it comes from direct, hands‑on experience with population disaggregation in applied InVEST projects. Your confirmation that dasymetric mapping is a robust approach gives me confidence, and I will cite your paper in my methodological framework.

Your warning about intra‑urban heterogeneity is a key insight. In my study area, population density varies enormously within the same LULC “urban” class – ranging from high‑rise districts to informal settlements with very low density. Using only a binary urban/non‑urban mask would indeed introduce severe bias.

Regarding your recommendation to read city‑specific literature – I have already begun reviewing dasymetric studies in other Latin American megacities (e.g., São Paulo, Mexico City, Santiago). I will incorporate their findings on how to handle informal settlements and verticalisation, which are highly relevant to my case