May 25, 2022
Developing a heat pump installer proximity indicator
While we spend a lot of time thinking about how to drum up consumer demand for heat pumps, the supply side of the industry is equally important, as policy interventions to encourage uptake and other efforts intensify.
What happens when there is demand but a lack of installers in proximity to households? Is there a way to measure local proximity to get a sense of where installers install heat pumps to anticipate supply? In an effort to investigate these questions, we’ve developed a baseline installer proximity indicator using data from MCS and Energy Performance Certificate (EPC) ratings.
The code for this indicator sits within the enriching pipeline of our data infrastructure repo on GitHub, a central location to harmonise our data fetching, testing and processing steps across our core MCS and EPC datasets.
A quick note on our data sources: MCS is an organisation that certifies heat pump installers, among other certifications. Their data includes the name and location of heat pump installer companies in addition to the type and location of heat pump installations. Installers do not need to be MCS certified and as a result, the data does not represent all heat pump installations across the UK. Meanwhile, we also have access to housing information across GB via EPC records, a review of a property’s energy efficiency. While there are noted data quality issues with EPC records, it’s an open, free and rich source of property information.
Beware that this blog is a technical one so if reading about a baseline indicator methodology sounds like a drag, give this medium post a miss. We’ll have plenty of interesting, non-technical content coming soon on a variety of topics related to UK home decarbonisation. Otherwise, buckle up for an indicator explainer and a few geography visuals!
What do we mean by installer proximity?
In order to generate an installer proximity indicator, we first had to develop a definition of proximity. In order to define what it means for a property to be “close” to an installer, consider the example below:
Even though the property is close to another property where an installation took place, it’s far away from the majority of installations performed by the company. We assumed that if a heat pump installation company installed many heat pumps in one area with a few outliers, they were more likely to install in the highly concentrated area as opposed to near an outlier. We therefore opted for the median point of installations as opposed to the installer company’s location because we found that, especially in the instance of larger installer companies, their company location could be far away from many installations. This assumption suggests that we could look at how far away properties are from the median location of the installed heat pumps to mitigate against the effect of outliers.
The geographic distribution of heat pump installations
Based on our working definition, we developed a proximity indicator that took into account the geographical distribution of heat pump installations per heat pump company. To capture distribution, we calculated the distance between the median point and all installation points per company. We were then able to generate a number of differently sized geodesic buffers (circle-shaped boundaries that take into account the earth’s shape) per company by calculating the distance between the median and the minimum, 25th percentile, 50th percentile, 75th percentile and maximum installation distance. Each installer company therefore had five buffers associated with them, ranging from small to large. Finally, we removed a few outlier companies that had unusually large maximum buffer areas so as not to capture companies with buffers so large that they captured all of our EPC properties.
Generating the indicator
Once we had buffers that captured the distribution of heat pump installations per company, we could then turn our attention to property characteristics of houses in the EPC data to investigate the composition of buffer boundaries that properties fell within. We did so by counting the number of different minimum, 25 percentile, 50 percentile, 75 percentile and maximum company buffers a given home fell within.
Finally, we converted these raw counts into an average weighted indicator that penalised high counts of maximum buffers and rewarded high counts of minimum buffers. This means that if an EPC property had a higher score, it fell within the bounds of many company minimum buffers. Meanwhile, if EPC homes had a lower score, they were not in as close proximity to installers.
We were able to apply this approach to a random sample of 50,000 properties from the EPC database across England and Wales to investigate areas of high and low installer proximity. Apologies to those interested in Scotland and Northern Ireland findings — it’s a bit of a data quirk as different countries in the UK release EPC data separately, as opposed to a UK-wide release. This method could easily be applied to different EPC datasets.
What we learned (and hope to learn!)
We can make a number of regional observations based on installer proximity scores across a random sample of EPC rated buildings in Wales and England. For example, by exploring density maps of installations and installer proximity scores, we observe that coastal regions across Wales and England are not in as close installer proximity and that there are a number of installer proximity hot spots across London and Northern England. While this is largely unsurprising given that there’s high installation density around densely populated areas in general, there are also some areas that appear to have higher installer density scores and lower household density, such as Harrogate.
We can also compare and explore scores in different areas. For example, the average indicator proximity score for homes in Canterbury is almost twice as high as in Reading, suggesting that there may be more homes in closer proximity in the former location than the latter. Meanwhile, although the average installer proximity score in Llandrindod Wells, a town in Wales with approximately 22,000 households, is slightly lower than the England and Wales average, it is over twice that of the minimum installer proximity score. It was also rated the 5th happiest place to live in Britain, perhaps due to proximity to heat pump installers!
Through investigating these scores, we could identify and target geographic regions of EPC properties that could be suitable for a heat pump and are in close installer proximity yet don’t currently have heat pumps. The main use case of this indicator is therefore to feed it into machine learning models for future data-driven projects as part of Nesta’s sustainable future mission.
Next steps
This indicator is simply a baseline and could definitely be improved in a myriad of different ways: we could adapt it to use road network data or travel time or we could explore more advanced algorithms that take into account neighbourhood density like kernel density estimation. In the meantime, we’re rooting for homes with high installer proximity scores to get heat pumps as more individuals retrain or enter the heat pump installer landscape!
thanks also to Liz Gallagher for the code review and guidance
