Hi everyone, it's Calandiel again.

Today I'd like to talk about climate and the release date for 0.3.

And since I imagine the latter is what most people are interested in, let's start with it. Progress has been steady, we've been working on displaying buildings in the city view and letting AI create new buildings. Characters can also give birth and die now, though, the numbers need tuning, similar to the first iteration of the nutrition model few months ago.

As for the release date itself, the main issue is that bugs with rivers we found last months require more work to fix than I initially expected. I'm not sure if it's worth delaying 0.3 to fix them (while foragers do use rivers for fishing and water, they're not as essential as they'd be for agriculturalists). If we don't, the late December release will stay as the target (though, probably after Christmas, not before it). If we do, we may need another month to fix everything up. One possibility would be releasing a preview of the update as a part of our monthly test builds and releasing it publicly later, when rivers are fixed. Let me know what you guys think.

For now, though, here are some of the buildings I mentioned:

They're still rough around the edges but I think they'll look acceptable in a months time.

In other news, our old climate repository is seeing some work again: https://github.com/Calandiel/climatenet

I'd like to improve the climate model after 0.3 is out and over the last weekend I explored an approach that should do just that. If you've been in the community for a while, you may remember that the climate model in the first release of SotE was a general circulation model. And while it arguably produced more realistic dynamics, it failed short as far as predicting Koppen climate zones goes. Which shouldn't be surprising as comparable scientific models that used comparable computing resources, such as Model II also fail to do so. Because of that, we changed the model from a prognostic to a diagnostic one, trading dynamics, and (in some situations) underestimation of inland deserts for higher variety in Koppen climates.

While it's better for a video game than the previous model, it still has enough issues to make me look for improvements. The approach I've been considering lately is training a neural network on handcrafted Koppen climate maps for SotE heightmaps. I did some initial testing and I think with enough training data it should resolve issues with the current model.

That, however, will require help of people who have experience worldbuilding climates. If you do (or want to learn it) and you'd like to help, send me a message on Discord and I'll explain the rest of the process.

In a week or two, when this months test build is released, I'll release another dev diary describing it to non-Patrons.