Mapping Covid-19 Vulnerability
This post covers the mapping of South Island communities and their vulnerability to Covid-19 reinfection as part of an ERST202 class project.
While semester 1 feels like a long time ago now, the threats from Covid-19 are still remarkably present. Recent incidents around returning Kiwis and their efforts to break out of quarantine illustrate just how vulnerable our borders are to reinfection and a dreaded “second wave”. The team of 5 million did an awesome job of taking NZ cases down to zero, but it’s easy to see that even one case in the wrong place at the wrong time could take us right back to level 4. There are certainly some spatial components to people being at threat, and there are also some spatial components behind what might make some areas more vulnerable than others, in the event that community transmission rears its ugly head. Things like the number of older people, or the medical resources in an area. At the end of last semester, the good folk of ERST202 took on mapping some of these factors for several South Island communities as part of a class project.
This is not a new idea. The US Centres for Disease Control (CDC) have developed a Social Vulnerability Index (SVI) with an interactive map that visualises how vulnerable different areas are to a range of events (you may need to run this in FireFox or Edge [never thought I’d recommend someone using Edge...]).
The colour codings allow you to tell quickly how at risk a particular area (US counties in this image) is compared to others, given the legend.
Underneath the bonnet of this map is a fairly complex set of demographic factors (15 in all across four themes) that contribute to this:
There are some challenges to doing something like this, not least of which is how do you set up these factors so that they are comparable in a logical and defensible way. (We hit on some similar issues here.) The CDC’s approach was to rank each area unit based on the number of people (or things) fitting each criterion. These were then divided by the largest number of people (or things) across the whole area so that each factor is rescaled to a value between 0 and 1. Combining them all and rescaling them again reflects the influence of each factor and makes each area comparable. This also means that different numbers can occur depending on whether the analysis is nation-wide, or state-wide, or county-wide.
For our class project, we adopted a similar approach though simplified things down to just four factors:
- Population Density – higher population density increases the likelihood of community transmission
- % of population over 65 years of age – older people are at higher risk of death from Covid-19 and associated complications
- Socio-economic status – people of lower socio-economic status have fewer resources to allow them to remain isolated from exposure
- Access to hospitals and clinics – as distance away from hospitals and clinics increases, health care opportunities decrease.
We did our analysis at the meshblock level within different census area units. Two of these factors come from or can be derived from census data (population density and % of population over 65). For the socio-economic status, we used the NZ Deprivation Index, available as a spatial data layer. The locations of hospitals came from a Ministry of Health spreadsheet with street addresses that were then geocoded to give us points on the map. Distances were calculated from the centre of each meshblock to the nearest public hospital.
For this analysis, each student worked with a census area unit. Within that area unit they worked at the meshblock scale and added in data related to each of the four factors. I provided them with the maximum value across the whole study area which they could use to rescale the factors across their particular area unit. In the end, we could stitch together all the area units into one larger study area – the rescaling meant that each meshblock was easily comparable against all the others.
To keep things consistent, in terms of symbology, everyone used a layer file on their final vulnerability score. And here’s our final result:
Values closer to 1 are more vulnerable than those closer to 0. Here’s the web app if you’ve like to take a closer look.
So now we can get a sense of, given our factors, how vulnerability varies across the study area. For instance, we can look more closely at each meshblock and see how its vulnerability scores rack up:
This particular one has low population density and % of people over 65 (0 for both, in fact), but is high on the deprivation scale and far from the nearest hospital.
But there are a few caveats:
- Each meshblock’s vulnerability is only comparable to other meshblocks in the study area and not to any outside the study area. If we did this on a national scale the numbers would most likely change;
- The demographic data are based on the 2013 census, as it was felt that the 2018 results were unreliable without some further study;
- Meshblocks are the finest scale we have access to for mapping census data – any value we map gets spread homogeneously across the area of the meshblock so we can’t easily take into account how, say, population might vary within a meshblock. This becomes more important as the meshblocks get larger;
- The hospital vulnerability used straight line distances from the meshblock centre to the nearest hospital. This is quite different from the actual distance that might need to be travelled on the state highway system to get to the nearest hospital. If we’d had more time, we would have used some network analysis to determine the actual road distances
In the end, this is a very simple analysis compared to the SVI, but it does serve to demonstrate (I hope!) the usefulness of spatial indices to better prepare for eventualities. While things are (mostly) fine and dandy, Covid-wise, right now, we’ll have to remain vigilant if and when the borders begin to open up. A national scale vulnerability index like this could be a part of the planning and help government focus their efforts..