Re: [Flightgear-devel] Traffic 2020: Towards a new Development Model for FlightGear AI Traffic

[Thorsten R. <tho...@sc...>]

> I’d be curious to hear from Torsten D and Thorsten R about the weather  
> simulations, how they might cope with this.

Speaking for AW: I guess the main thing is lots of stuff is a lot easier  
if you can assume it happens in the 100 km around a certain spot.  
Basically you can assume Earth is reasonably flat, you can expand  
coordinates into a local Cartesian system of equally-sized tiles, you  
never need to put up with spherical trig, you can put enough 'looseness'  
into the weather grid such that it gradually deforms to adapt to  
prevailing winds and traces the fronts (for instance weather at fronts is  
oriented - there's a cold and a warm side to a front with quite different  
properties).

If you have to cover the whole globe with about equally sized weather  
tiles, the form of the grid and continuity quickly becomes issues. If you  
allow tile sizes to shrink with latitude, then the larger clouds may not  
fit into a tile any more... You have to use spherical trigonometry to set  
up everything, it's not clear where the 'looseness' to deal with frontal  
weather or higher latitudes should come from...

Don't you wish sometimes we'd live on Discworld?

Personally I think the idea of a local 'loose' grid gradually adapting to  
what's needed is pretty neat because it solves a lot. How you'd MP that  
depends on whether you want to have exact sync (i.e. see the identical  
arrangement of clouds at identical places) or approximate sync (see  
optically similar weather) across instances.

Exact sync means either giving up the loose grid (without a clear idea on  
my side of what should replace it) or as in the P2P mode, allow the first  
plane in the region of interest to define the grid and make all others use  
it as long as they're close and transit to their own as they depart  
(should be unproblematic as long as Earth is sparsely covered by MP  
planes, i.e. most of the planet would not have to explicitly generate  
weather tiles - which I believe is the case).

Approximate sync means that each instance fetches weather tile meta-data  
 from the server and build from that (rather than its own random number  
driven tile generator), but that will not have the grids for two planes  
precisely aligned with each other, so they will see the same type of  
clouds and weather, but not necessarily in the same spot. Needless to say,  
that's easier.

(Note that whether you see a cloud and fog etc. also depends on rendering  
settings, so we're unable to guarantee identical weather visuals for the  
simple reason that we have different renderers and selectable LOD and  
density ranges for clouds).

That's at least my two cents. It wasn't called 'local weather' for nothing  
;-)

* Thorsten