How many people live here?

Buildings can be used as proxies in estimating population. Using census data on household size, if you know the number of buildings (and specifically residential ones), you can estimate how many people there are in particular location and, given PLACE's high resolution aerial imagery (5cm is the size of large post-it note) you can easily count buildings.

Using new imagery for Zomba, Malawi (acquired by PLACE in December 2023) and available for analysis just a couple of weeks after collection, we wanted to show how PLACE imagery coupled with machine learning could give the Council new insights into their city.

Our analysis began with a positionally accurate  true ortho , generated from an image collection comprising several hundred  geo-tagged  JPEGs (use the interactive map below to see the result):

And this is what we did:

• Using Esri's  pre-trained models  (by pre-trained we mean models that have been developed on generic image data, not image data from Malawi), the true ortho provided the input for building detection (or more precisely building roof detection) in ArcGIS Pro. Even using generic models the results were impressive with buildings represented as discrete, smoothed polygons. Using  Text_SAM  you simply type in what you want to detect (segment) e.g. buildings, trees, cars and away you go.

Below is a short video that shows Text_SAM in action (this just represents a small area in Zomba demonstration purposes):

• Once buildings were detected, and within this small area 651 buildings were, we used ArcGIS Pro to calculate each building's area (in square meters). The average area was 89 square meters. This figure can be referenced back against prevailing housing guidelines, for example Habitat for Humanity uses a floor plan for rural Malawi of  30 square meters .

• Next we moved onto estimating population. According to the 2018  Malawi Census  the average household size in Zomba is 4.2. We used a slightly higher figure of 4.4 to account for population growth between 2018 and 2023. The calculation is the number of buildings * 4.4, so, 651 * 4.4 which gives an estimated population of 2,684 (by comparison the  WorldPop population estimate  for this area in 2020 was 2,158). Council planners can use these numbers to plan for service provision, whether waste collection, clinics or schools.

Mapping is also critical in identifying vulnerable and high priority assets for risk reduction investments addressing  SDG 11.5  which seeks to significantly reduce the number of deaths and the number of people affected and substantially decrease the direct economic losses caused by disasters.

Using PLACE imagery acquired for Dansoman, Accra in Ghana in 2020, the World Bank assessed housing resilience using deep learning models that it had developed for the Caribbean. These models were used to map roof condition and roof material.

The results from this study can provide input into risk models used to assess the financial and human impact of natural calamities:

The results of the study showed that of over 25,000 dwellings detected, more than 15,000 (60%) had a metal roof. Slightly less than 15,000 (60%) had a roof deemed to be of fair condition and just over 7,000 (30%) had a roof of poor condition.