Resolution Matters, Part 1: Varna Lake.

Why should it matter? Weather systems are such enormous creatures, sometimes capable of wiping out entire village or even a town! Why should a resolution of 9km be so different from that of 2km? There’s hardly a topic in the world more relevant to the old saying:

The devil is in the detail.

We know the wind could be such a devil! And the detail here is the terrain. Sometimes old fishermen, or rangers can predict the local weather better than any supercomputer, without knowing the satellite imagery or radar data, or even the local radio broadcast! They do so, simply because the local topography of the terrain influences the weather quite enough to “limit” it to a bunch of patterns that can be learnt from the locals. Especially if they have a good amount of years behind them to do so!

We’ve recently participated in a sailing event — The traditional “Third of March Regatta” in Varna Lake, Bulgaria. Everything in this regatta is rather old, and … traditional — rigs are wooden, boats are slow and heavy, people know each other for years. One of them is particularly well-known for winning the regatta for the last 21 (yes, twenty one) years. Many ingredients asking for good preparation on the tactics. Wind shifts are notorious here — few years ago, we’ve witnessed here two smokes going in the opposite directions at the same time!

Two things, however, were new for us — one was the success from the previous year, when we’ve managed to secure a steady second place, even challenge the winner, and the other was the intention to rely solely on Windyweek’s forecast. And, of course, compare it to the other products out there. But, first — here’s how the Varna Lake looks like:

Source: Varna, 43°11'55.64"N, 27°52'21.99"E, Google Earth, March 13th, 2017.

The Black Sea is to the East, the Varna City is on the hill to the North and North-East. There are small hills on the South shore, which get higher if you follow them to the East. The wind usually comes from either the openned West and North-West sides or from the two channels, i.e. South-East. This cozy bay hosts the sailing event for the last 30 or 40 years (the event is 60+ years old, but in the beginning it was held in the open sea — just outside the channels).

So, there were several details to consider in chasing the devil — (i) hills from both sides of the lake do induce a funnel effect; (ii) the height of the shores to the North and to the South makes it hardly possible for the wind to blow from there, and (iii) due to the cold winter, the sea and the lake were unusually cold. All of the softwares we’ve used showed generally similar predictions, including Windyweek. None of them nailed it perfectly, including Windyweek.

However, one of them consistently different forecasts, closer to the reality that followed — Windyweek!

All of the days, Windyweek’s forecasts were 2–3 knots higher, which came to be true. The reason is that the higher resolution manages to take into account the funnel effect, especially when the wind is blowing from the channels, which was the case. On the first day all platforms predicted drop of the wind around 5pm, but only Windyweek suggested it will not be that quick, and indeed — we’ve managed to practice another extra hour.

On the second day all predicted a dramatic wind shift, as can be seen:

Snapshot from

In reality, it happened around 11:30am, but for Windyweek’s credit — other platforms has predicted it to happen around 3pm!

And, on the third and last day, other platforms predicted wind from South and South-East, which we’ve already seen is not quite possible, so the actual direction was more like Windyweek’s prediction — East, South-East, and — as expected — stronger, because of the channels.

Photo courtesy of Svetlana Mihova

Yes, there was fog on the racecourse, and yes — Windyweek did predict it, and no — not all others did! Why? Many possible reasons — Windyweek’s ingestion of the most detailed topography data for the region, meaning it knows the elevation of the terrain at 90 meters resolution; or the use of the best available data for land classification, meaning it knows what‘s below — a lake, forest, land, type of soil, color, etc.; or the fact that the WRF model used, properly takes into account the interaction between water, land and atmosphere, something completely missing from the more common GFS model. But, that’s another story. Let’s get back to the resolution — one could ask:

Why doesn’t everybody use higher resolution for weather prediction?

Well, because it doesn’t come for free. Having a 2km resolution instead of 9 means ≈20 times more calculations, i.e. that much more processing power required. Not to mention that some platforms rely entirely on the freely provided low-resolution models, essentially meaning they don’t make additional calculations, meaning they don’t add new value to the results. On the other hand, it is just not feasible to cover the whole Earth with such high resolution.

We keep trying, though. Stay tuned!

Photo courtesy of Svetlana Mihova.