How the reception figures are worked out

And, more usefully, how far they can be trusted.

Where the numbers come from

Ofcom publish, for every transmitter, its position, its aerial height, and the effective radiated power of each service. Nobody publishes coverage as reusable data — Freeview's checker exists but its predictions are not licensed for reuse, and are not used here — so coverage has to be modelled from those inputs.

The model is ITU-R P.1812-8, the International Telecommunication Union's point-to-area method for VHF and UHF. It is the same family of method used for broadcast planning. Our implementation agrees with the ITU's own validation data to 4.8×10⁻⁹ dB across 63 test paths, so any error here is in what we feed it, not in the arithmetic. [V]

The ground between transmitter and receiver comes from OS Terrain 50 across Great Britain and Copernicus GLO-30 for Northern Ireland, the Channel Islands and the Isle of Man, sampled every 50 m along the path. Land cover comes from ESA WorldCover, which is what distinguishes a path over a town from one over a field, and sea from land.

What counts as good reception

A prediction is a field strength in dB(µV/m); a verdict needs a threshold to compare it against. Those are the ITU's published planning figures, not ours:

Where it is wrong, and by how much

Three known optimisms, in rough order of size:

1. Television transmitters are treated as omnidirectional at full power. Most are not. Ofcom publish radiation patterns for radio but, for television, only for a set of mostly small relays — a 2009 Freedom of Information release covers 70% of transmitter records but only a few per cent of households, and not one of the 93 stations above 100 kW. So off the main beam, a television prediction here can be substantially high. [D] FM and DAB do not have this problem: patterns are published for 99.4% and 98.9% of radiated power respectively.

2. Local clutter is not modelled. The path over the ground is modelled, including the height of buildings and trees along it. What is not modelled is the last hundred metres — your neighbour's roof, the tree in the way, which side of the house the aerial is on. Those routinely cost more than everything else on this list.

3. Interference is not modelled at all. The prediction is a field strength against a noise-limited threshold. Co-channel interference from a distant transmitter on the same channel, which is what actually limits reception in much of lowland England, is not in it.

How far it has been checked

Ofcom published predicted coverage for 81 named sites. Against that table, this model gets the order right and the size wrong:

The third row is the one worth pausing on: nothing in the model is told that the commercial multiplexes radiate less power than the public service ones. It falls out of the published figures, on every site. [V]

Read that as: if this site says transmitter A is a better bet than transmitter B, that ranking is supported. If it says a transmitter covers a particular place, that is a weaker claim, and the further out you go the weaker it gets. It is why transmitter pages give coverage as a share and never as a number of households.

A second model, and the thing it agrees about

Ofcom's table is nearly twenty years old, so the model is checked a second time against a predictor that is still maintained. Compared against the Wolfbane Cybernetic DTT predictor — sampled at 300 locations, one request every ten seconds — the two agree about the thing that matters most here and disagree about the level. Both are models, so a disagreement locates a difference between them and says nothing on its own about which is closer to a real aerial.

The agreement that matters is on the decision. Across the 294 locations where both name more than one transmitter, the strongest is the same mast at 83%, and ours is inside their top three at 99%. [V]

On the level they disagree, and this site is the optimistic one: a median +28.3 dB across 1,699 paths. That is not one number, which is the part worth knowing:

It grows with path length — +20.7 dB inside 15 km against +34.4 dB beyond 60 km. A difference in what the two models mean by a figure would be flat; a difference in how much the ground is taken to block a path grows exactly like this. [I]

Everything in more detail

The full working — every parameter, every trap in the source data, and the measured cost of each step — is kept alongside the code rather than summarised here, so that it cannot quietly diverge from what actually runs.