I want to preface this by saying I am 100% glad this feature is finally in TSW. I have wanted this feature since day 1 and have asked for it multiple times over the course of 5 years However What I am noticing is a very set pattern of the weather always progressing through the same stages in the same order ie Clear > Foggy > Cloudy > Rain/Snow > Storm The only variation seems to be how long the weather stays on one of the stages before progressing to the next. Personally I have yet to play a service where the pattern goes the other way or randomises it a bit more than it currently is. I would like to see possibily for example a couple of minutes of rain and return back to clear (how weather quite often works in the real world) It just seems once the weather starts to change it follows the same pattern each and every time which makes it feel very scripted. I would like to see this feature improved upon in future updates and hope it wasn't just a one and done inclusion. Thanks for reading
I think the dynamic weather is rather nice and frequently have quick showers happen along with storms building up like they would in real life but also from storms/heavy rain to much lighter rain. It is a neat effect.
The weather is part scripted part dynamic - the scripted bit is to make sure that transitions make sense, the dynamic is that it's random how it navigates it. Let me share you a dev diagram - the percentages on here may not match the ones in the game exactly but it should give an idea of how the system works. Note: these images are coder-art and auto generated by a tool. This is summer. This is winter. There is one for each season. So if you tell the game to start on "light cloud" in a winter month, you can se you have a 60% chance of going to a clear sky, 20% chance of going to foggy and 40% of a light snow fall. This will happen after a random time. Let's say you go to snow light, after a random time there is then a 40% chance of going to a snow storm, 80% of the snow stopping and it getting foggy and a 20% chance of it simply returning to clear. Hope this helps explain it a bit. If so - i'll ask someone in the Creative Services team nicely to pretty these images up (sent to me by a dev so I could see what had been done) and then we can make it the subject of an article or something. Matt.
It is based on presets with a choice of things that can happen with probabilities assigned to those outcomes (Ah, Matt has replied with a diagram as I type). These are different at different times of the year so more unsettled weather will happen if playing in October than would do in June etc. It is the first iteration of it and I think it is okay but one thing that needs sorting is having rain when there are hardly any clouds in the sky. That shouldn’t happen but does a lot and that’s one reason why I don’t use it as often as I might. Another thing is that fog instantly disappears instead of fading which is like some magic trick.
At least the 12-year-olds complaining about the previously Faraday cage-like lightning being "nerfed" seem to have gone away now...
How are those percentages to be read, since they don't add up to 100% for each bubble? Especially when they add up to more than 100%?
Ah, this is probably because the arrows are missing in the winter diagram... Maybe it is a logic puzzle.. add arrows in a way that outgoing arrows from each bubble add up to 100%.. maybe like this:
Thankyou for the detailed response DTG Matt That gives me a bit more a clearer picture of what is going on under the hood. I think maybe my problem is that I was starting services in the summer months (note I am in Australia and services use your system time to set the default season -- So July/August for me is Winter but Summer where the routes are generally set) I will play around with the seasons a bit more to see if I can ge a bit more variation...as the winter months look like they see a little more switching between the patterns compared to summer
I still don't agree with lightening storms in winter personally. Yes I know they happen every so often but it's a rarity in my experience. I think if I amputated 3 fingers on my left hand I'd still be able to count the number of times I've seen it in 43 years on that hand.
Here in Australia (at least in the Eastern states) we have had 3 years of solid rain and storms due to La Nina. I can count on one hand the amount of sunny days we've had
Ironically, as I stare out my window at the office at probably the clearest, sunniest, warmest day we've had all year...well and truly sick of the rain but at least it kept a lid on temperatures, now that things are starting to clear Spring will be making up for lost time I'm sure.
Reports are suggesting that La Nina is hanging around until at least early next year....so keep the brolley handy. Covid/lockdowns have also been a bit more tolerable due to the rubbish weather at least
I've run a number of services in which the weather either clears up, or deteriorates and then improves, on SEHS and CJP, for example. So, I think the DW feature, though not perfect by any means, is better than it was initially and worth keeping as an option in the settings.
Nah I was referring to the lockdown we had in 2021. Alot easier to not go out when its bucketing down with rain
Remember I said those percentages probably weren't correct, I think this was the developers first pass before entering it into the game so it will have been adjusted on going in to data files but he didn't then come back and update these to reflect how it ended up - the 'wiring' is accurate ro the game but the percentages may well be off. Just provided as a guide to explain is all
Thanks for the information. Bearing in mind you've mentioned the percentage being possibly different in the game it still seems the lightning storm should be somewhat uncommon. But they do appear often. As mentioned previously, the weather mostly goes from bad to worse almost like the game treated the x% more like (100-x)%. Dynamic weather is a cool addition, but it definitely needs some tuning and probably a bit slower changes (or speed setting).
I also think the propabilities should be tweaked a little. Like a heavy thunderstorm should really be a rare occurence, even in autumn, but right now it feels like most of the time it starts to rain, it will end in thunderstorm. Also for winter, it would great if you could add an additional clear initial preset - clear with snow cover.
I think we need a "no extreme weather" option. So in clear winter it might change to clouds or light snow but never to snow storm or thunder. Or during summer there might be a change from clear to rain but not clear to thunderstorms or clear to extreme fog.
Well, it would be interesting to see the real percentages. In my case,I always start with clear sky using the actual date. So I drove only in summer and autumn since TSW3 is out. In 100% of the cases, I had rain (either light, shower or storm) after 30-45 minutes. 100% of the cases. I may be really unlucky, but I think the rain chance percentage is way too high. Don't take me wrong, I like the dynamic weather, but I think it needs fine tuning.
Just about every time I have used dynamic weather since it was introduced, it has end with a thunder storm. This even happened this past Sunday on the German route even though the date was the actual date, 23 October. Most of the time I do not use it and select my own weather for that run. One of the other things that I find a little unrealistic is, the number of cycles it does in the short period of time at these runs take.
It would also be very interesting to know how often the "does-the-weather-change-test" is performed in a specific length of time. If we knew the probability (percentage) of "clear" weather changing to "any rain" (p) and how often the "does-the-weather-change-test" is performed in a specific length of time (n), then we could use the formulae for a Bernoulli process to calculate the probability (percentage) of the weather NEVER changing to rain in this length of time (k = 0), no? With k = 0 this should be quite easy, since the binomial coefficient of n choose 0 = 1 and p^0 = 1 as well, so all you have to calculate is (1 - p)^n. So if the probability of "clear" weather changing to "any rain" is only, say, 10 %, then 1 - p is 90 % or .9, so after let's, say, 20 tests the probability that the weather NEVER changes to rain throughout the trip drops to .9^20 = .121.. or about 12 %, after 35 tests even to .025... or about 2,5 %. So no big surprise that the weather almost always changes to rain eventually, especially on longer routes. Or am I getting anything wrong here?
The question is rather - does it take place or not? Assuming you choose a short shift, the weather will not change much over time, remaining fairly static. It's also interesting here to experience a change in the weather, otherwise you wouldn't need it. Assuming you choose a long shift, the weather may change several times. Depending on the season, geology and topology, this is not unrealistic. But that can happen if the weather changes three times within 90 minutes. Dynamic weather doesn't mean to me that it can change, it will! It is also logical that it follows a meteorological sense. And that often feels like a repetition or seems predictable. You can influence that a little if you change the start condition of the dyn. weather changes at the beginning. But the fact is, you go through a certain chain, just at a different point. I could also imagine that if you make this more random, it will seem less logical and therefore less realistic. But weather is also largely predictable. If it's friendly at first, it will get worse. Or the other way around. So you can prepare for it. I agree that a little fine-tuning would do well here. But basically I like these weather capers. Because whenever I activate this feature, I know what to expect and that's how I want it. We all must not forget that all of this is also the immediate consequences of climate change...
Given that 0 = 1, and p^0 also = 1, I think you're forgetting the most important question: would you like a toasted teacake? (This comment was brought to you by Talkie Toaster, with apologies.)
We don't want teacake, no toast, buns, baps, bagets or bagels, no croissants, no crumpets, no muffins, no potato cakes and no hot cross buns! And definitely no smegging flapjacks!
My own experience has been positive. On a long SPG run, for instance, I started with light clouds and got occasional showers with clearing repeating over the entire service, appearing very realistic as to actual weather in western PA in fall. I have also not encountered the problem that others have where services always seem to end in the worst weather possible, although it has happened perhaps more often than I would consider realistic. If it were up to me, one change I might consider would be an "algorithm stop" probability that would cancel any further changes to the weather, perhaps 10% at the start of a service and 5% after each algorithmic change. This is to reflect the fact that sometimes the weather doesn't change. I'm the current situation, if you drive long enough the weather is guaranteed to change. The drawback is that some players might then notice a lack in dynamism. Even with no change I like the effect and if I really don't want the weather to change I can set it myself. It is probably the best feature of TSW3.
I really like to see realistic weather changes on a route per route basis. For example NTP, in reality the weather is usually worst west of the Pennines and better on the east side. It's common to have heavy rain in Manchester and bright sunshine in Leeds (Manchester is the wettest city in England due to the Pennines trapping clouds over it). Id also like to see clouds hang around mountains and weather deteriorating as we gain altitude.
I agree to maybe ease up on the heaviest storms, they should be pretty rare. I mean, if you WANT to drive in a storm just set that as your starting weather. I'd also love to see a more "static" dynamic weather, where it may change slightly, but it tends to stay the same. Or, even better, a slider for variability.
Pls make it a three-dimensional slider for x: general volatility (high–low) y: extremity (likely–unlikely) z: chaoticity (insane–realistic) Well, no kidding, I am so happy that you implemented the dynamic weather, whatever you do with it, I will enjoy!
Look up Markov chains. You can build a probability matrix from the diagram (although the RainShower node seems to lack the data required for this). The matrix will be 8x8 so good luck solving it by hand but Matlab is good for this purpose. You start with a 8-dimensional vector that has all the initial conditions, presumably the transpose of the row vector in the probability matrix for your initial weather pattern. So for example if you start in Clear Sky the vector would be: (0.75, 0.05,0,0,0.2,0,0,0), but really any vector will do as long as it is a probability vector (sum of the components equals one). Then you multiply by the probability matrix for each time the dynamic weather system "ticks". There's a lot of really cool theory about Markov chains, for example: you can end up in "dead ends" where it's not possible to go back. Also there always exists a set of probabilities (or perhaps a "state" is a better term, an eigenvector for those who know linear algebra) that is unchanged by the probability matrix. So no matter how many times you run the "tick" you will always end up with the same probabilities.
Wow awesome! There's the theory for solving the whole thing if probabilities change between tests (as opposed to working with a constant probability throughout the whole chain of tests, like in Bernoulli processes)! I actually tried to calculate the no-rain probability for the summer weather diagram "by hand", but it gets very complex even after only five tests (but at least the same figures, 0.75, 0.2, 0.05, show up as in your vector): Maybe this is the reason why you can get stuck in rain or a thunderstorm.. Rain shower seems to miss the probability for "staying at rain shower" (maybe 85%), I guess.
So... luckily I found a website that can solve the 8x8 matrix. I built the matrix, guessing the missing data for RainShower (0.15 to LC, 0.85 to RSH) and Lightning (0.4 to RL, 0.2 to RST, 0.4 to L), punched it in, calculated 50 transitions and added up the probabilities for all the "rainy" states (RL, RSH, RST, L) What you can see is that with the data given in the summer weather diagram the chance of ending up in a "rainy" state after a test climbs from 0 (at the first test) to a more or less stable 27.69% (starting after about 30 tests). What I haven't found out yet is how to calculate the probability of NEVER ending up in a "rainy" state after a certain number of tests.. Anyway, maybe this game should be known furtheron as "Rain Sim World".. I love rain (irl and in the game), btw, in case anyone wondered..
Well maybe like this: Calculating the "no-rain" probability for every row and then multiply down the rows... If this is correct, the probability of it starting to rain at some point in the journey after starting with clear skies is higher than 90 % after test #13, higher than 99% after test #20, higher than 99.9% after test #28 and higher than 99.99% after test #37. cwf.green , does that make sense?
Interestingly enough, yesterday I ran a high speed service on SWHS and started with fog and it stayed foggy for the entire 40 minute run.
Yeah. You can predict exactly how it’s going to go. The weather changes at exactly the same point of a route.
Not sure exactly what you are doing here but a better way to solve this problem of "did we ever reach a rain state" is to make it impossible to go back to a non-rainy state and run the simulation again. For example, here is your original matrix: And here is a matrix with states 5 and 6 (RL and RSH) modified such that it is impossible to return to a non-rain state once either of these states have been reached. The other rainy states cannot change back to sunny without passing through these states so they may be left unchanged. In other words, we have modified the Markov chain to look like this: Now when we run the simulation from the starting point of clear weather, we can see the percent chance that we ended in a non-rain state without EVER touching a rainy state by summing the probabilities for each non-rainy state: After 13 transitions (row 14) we have a 52.32% chance of ever having rain. The chance of ever having rain goes up with the number of transitions, eventually to 100%. I'm not sure how many ticks the average user will experience during a given service (remember that not all ticks change the weather state!) but it's probably not unreasonable to guess that there is up to a 25-40% chance of rain in a service of average length. You can only get "stuck" if the state has no way to leave it. All of the nodes on Matt's flowchart have escape routes to other nodes, so none of them are so-called "terminal" (or absorbing) states. Cheers
Awesome! I can definitely see the appeal of your approach! Don't know why my approach makes the probability of "any rain" escalate even faster, but it is obviously flawed.. I am not sure if I am qualified to discuss this given that I haven't heard about Markov chains before yesterday morning, but your changed matrix doesn't seem to agree with your diagram, though. In the matrix the RL state seems to always change to RST (100 % probability), while according to the diagram it should change to RST only by a percentage of 40 % and remain at RL by a percentage of 60 %: Nevertheless, it doesn't seem to make any difference how the state progresses within the rainy states once it is "caught" there, I think. I tried to verify this by reducing the system to a 5x5 matrix with just one absorbing "Rain" state Calculating this, I ended up with exactly the same values you did, so I think your approach is valid. According to this, the probability of rain increases beyond 80 % after tick #28 and beyond 90 % after tick #39.
You are absolutely correct, the digraph does not match the matrix and has a few errors. It was a bit late writing this up and knew I made some mistakes somewhere. Really what I was trying to demonstrate is that we have a 0% chance of going back to a non-rainy state, so the percentage chance of rain will still come out the same. I've updated my original post for clarity and posterity. Cheers