Helping The London Fire Brigade Identify High-Rise Residential Buildings
Qbiz UK have been working closely with the London Fire Brigade (LFB) over the past few years to assist in making London safer by leveraging data for better decision making and planning in the fighting and prevention of fires.
This post details one of our projects for the LFB lead by one of our senior data scientists, Riccardo Scott.
The Context
Following the tragedy of the Grenfell Tower fire in 2017 the LFB set the goal of identifying and inspecting all high-rise residential buildings in London.
A building is defined ‘high-rise residential’ if (i) it contains at least one residential address, and (ii) it has more than 6 storeys or the floor of the last storey is over 18 meters above the ground. As the Grenfell Tower fire demonstrated so clearly, fires in high-rise residential buildings may become a major challenge for the fire department to extinguish, as external fire-fighting and rescue operations may not always be feasible.
For this reason, pre-emptively identifying high-rise residential buildings and ensuring LFB crews are familiar with building layouts, as well as the fire-fighting facilities and fire-engineered solutions installed to assist in executing safe and timely interventions, are essential to saving lives and protecting people’s homes.
Working With Topographical and Address Data
This task requires two types of data: topographical data — for identifying the height of built structures, as well as address data — for locating the structures and administering inspections.
Topographical data for Londons buildings is provided by Ordnance Survey’s (OS) MasterMap product, wherein buildings consist of TOIDs (TOpographic IDentifiers) — unique identifiers for almost every topographical feature in Britain. TOIDs are represented by polygons which have, among other properties, height values.
OS’s AddressBase Premium product provides up-to-date, accurate information about addresses, properties and land areas. AddressBase revolves around the Unique Property Reference Number (UPRN). Assigned by the government and OS, and maintained by GeoPlace, UPRNs serve as unique identifiers for every addressable entity in Great Britain. Almost every geographical object can have a UPRN, for example; a parking lot, a bus shelter, the advertising space on that bus shelter, an electricity sub-station, a shop, a flat, or a house — there are over 7 million UPRNs in London alone.
If we know the UPRN of a specific geographical object we can access almost any address-related attributes specific to that object. In addition to street name, number, postcode and their precise location, UPRNs are assigned to one of over 500 different categories. There are, for example, over 300 commercial and over 20 residential sub-categories.
UPRNs can be related to one another in parent-child relationships, for example, a parking space can be the ‘child’ of a flat, and that flat the ‘child’ of the estate or house it is in. This adds rich information over the structure of addresses with ‘families’ that can span over several generations consisting of hundreds of addresses (see Fig. 1 for an example).
Of course, OS provides a lookup table between the topographical data (TOIDs) and the address data (UPRNs). With this, our job should be easy, right? Find TOIDs with heights above 18 meters, and link them to their corresponding UPRNs, and we’re done! Unfortunately, addresses and buildings in the real world (especially in London) are not as simple as one might imagine.
Fig. 1 shows two representations of a three-building complex in Hackney (Note: we’re using ‘building’ here in the common, non-technical sense of the word — more on this below). It illustrates the complex way in which ‘child’ UPRNs, ‘parent’ UPRNs, and TOIDs relate to one another, and to the physical structures with which they are associated: each physical structure we would commonly call a ‘building’ is associated with at least one TOID and one UPRN, but may be associated with several of each, depending on the nature of the structure.
Fig. 1 (a) gives a topographical view of the complex, which contains the tall blue building on the right, and the two smaller beige buildings on the left. The yellow arcs between the blue and beige buildings indicate that some (but not all) of the UPRNs associated with each of the beige buildings share a common parent in the tall blue building.
Fig. 1 (b) shows a detailed network graph representation of the complex — UPRNs (orange nodes) may be connected to both TOIDs (blue nodes), in which they are located, as well as to their parent UPRNs. The brown node (on the right) is a a parent TOID to the 6 child TOIDs (blue nodes) connected to it.
To better understand the complexity of this data, consider the blue node corresponding to the TOID associated with the tall blue building (indicated by the right-most arrow). Notice that this (blue) TOID node is connected to 81 (orange) UPRN nodes, which correspond to geographical objects in the tall blue building (e.g. a flat). These in turn are connected to a single parent UPRN. Furthermore, notice that this parent UPRN is also the parent to 6 UPRNs located in the TOID belonging to the left-most beige building, represented by the leftmost arrow, and that the 6 remaining UPRNs located in that TOID do not have a parent UPRN at all. Lastly, the TOID containing the 81 UPRNs is connected to the 5 TOIDs (the right-most cluster of blue nodes) via the (brown) TOID parent node, and that these 5 TOIDs do not contain any UPRNs at all — in short, things can get quite complicated when dealing with topographical and address data.
The Problem(s)
In order to tackle the problem of identifying high-rise residential buildings for inspection, we need to consider two key questions in terms of the data we have: “What is a building?” and “Which address best describes a building?”.
What is a building?
While TOIDs may represent what one would commonly consider a ‘building’, they may also represent very detailed features of ‘buildings’, too. For example, a house with four balconies may well be represented by 5 TOIDs. As a further example, we may have a large composite structure (e.g. a complex like the one shown above) which contains a 40 meter high outer ‘building’ represented by a single TOID, but containing several flats who’s UPRNs are associated with the TOID representing the10 meter high ‘building’ at the centre of the structure.
Imagine many high-rise TOIDs (we will use TOIDs in place of ‘building’ to avoid confusion) next to each other, all contain addresses but common sense may say they belong to the same ‘building’. How many inspections should the LFB administer in such a ‘building’? Where does the ‘building’ in terraced housing begin and end?
The amount of exceptions by far exceeds the number we’d like in the ideal case where for the majority of structures, one address in an isolated house happens to be represented by one TOID. The same holds for the way addresses are structured.
Which address best describes a building?
Does your flat have a number, a letter, or both, or a name? Flat 1, Flat A, Unit 1a, First flat top floor, Top flat first floor? Are you living in a house with a number, or a name, or both? Is your flat number also the street number and the house only has a name?
There are many ways of addressing a property and many ways for properties to be part of a building — and vice versa. After all, addresses contain information required to deliver a letter; front doors, delivery points and P.O. boxes. They were not designed to describe or identify buildings.
What about the parent addresses mentioned above? Parent-child relationships are an abstract concept and not necessarily designed to represent objects the physical world. UPRN families can span over hundreds of meters and several buildings and are thus not always an apt proxy to use as the description of a ‘building’.
Furthermore, historical data and a dynamically developing real estate market pose a challenge to maintain an accurate, up-to-date representation of physical buildings and addresses, and OS and GeoPlace are doing a fantastic job.
The Solution
With TOIDs, street numbers, house names and parent addresses neither always, nor intuitively representing a building, the problem boils down to one question:
What is a building and how can one best aggregate the addresses in it?
There are many ways of solving this problem, and each one leads to a slightly different number of ‘buildings’, but ultimately it is only important that the right premise, building or building complex is inspected once the LFB’s inspectors arrive on site. (Note the ambiguous use of site, premise, building and building complex?). Here we present one way of solving the problem using geo-spatial operations, network theory, and domain knowledge for aggregating TOIDs and UPRNs to more comprehensible ‘buildings’ and addresses.
In a first step, we address the granularity of TOIDs. By merging TOIDs that touch each other we group them to form aggregated units, reducing the total number. If one can jump 50cm from roof to roof (TOID to TOID), the TOIDs are considered part of one ‘building’ — see Fig. 2 for an illustration. If one of these ‘buildings’ contains a TOID which has a height of 18 meters or more, as well as a TOID associated with a UPRN with a residential address, it is considered a high-rise residential building.
The second step tackles the question of which address to use for which building and how we may aggregate them as much as possible to a new ‘building’ level. First, all UPRNs having a parent UPRN in the same building are removed, the parent UPRN, which is usually less specific (e.g. a reference to multiple flats, but having no flat number of its own), is kept. Next, accounting for entrances from different streets, the addresses per building are aggregated by street. Where possible, any additional identifier the addresses have in common (e.g. house name) is extracted, as well as the highest and lowest street/flat number. These street level addresses are completed by postcode, locality, administrative area and town name.
The Result
As a result of our TOID and address (UPRN) aggregation (See Fig. 3) we now have a more intuitive representation of a ‘building’, having as few addresses as possible describing it, with a good deal of superfluous information removed, and making the geospatial data easier to interpret. This gives the LFB a good overview of how many inspections have to be performed, as well as where the inspection sites are located. The number of residential addresses per inspection site also helps with, for example, prioritising inspections and assessing risk.
We’ve seen that the physical world and addresses are not as ideal as one might think. However, together with local authorities, GeoPlace and OS do an incredible job of maintaining full and rich data sets of topographical and address data in the UK. In fact, it is so granular that for goals such as the LFB’s identification of high-rise residential buildings with human readable addresses, it needs to be aggregated. Here we have used geo-spatial data, network theory, and domain knowledge to provide a solution adapted to the problem.
If you are interested in finding out how Qbiz UK can assist you or your company in crafting data solutions specific to your needs, get in touch with Andrew Spry, CEO at andrew.spry@qbizuk.com.
