Shouldn't that be sorting by the world.y position? Ie, northness, rather than northness + altitude = world.y + world.z = screen.y.
I'm not sure about how much it matters for your game particularly (you don't seem to have world.z coordinates), but in general screen-space operations are a recipe for that sort of bugs, and should be avoided whenever practical.
My game is an "old fashioned" isometric game, meaning it is a 2D game through and through. All assets are square (for top down) or rectangle (for isometric) sprites. Of course, since it's pretending to be a 3D game, you have to draw/place the isometric tiles oddly so the inner rhombus (or hexagon, which is what an isometric cube is without shading) are drawn flush, since the graphics engine is working with rectangles. Thus the sprite's rectangles overlap like this: https://i.imgur.com/NTNHvbL.png
In order to simulate a camera looking into a 3D world, I need to draw everything from the top of the screen to the bottom, in that order, so assets with height are drawn over everything behind it (in world space)
For simulating world height (e.g. something flying one tile off the ground), you just need to move the sprite's screen.y position up the sprite's height (i.e. the height of the rectangle) per tile of height.
I have yet to implement terrain height into my game, but I think there are a couple ways to accomplish this.
1.) I can store the part of the asset touching the ground for all assets stacking upwards (like a multistory building). Thus I can keep the depth sorting algorithm I have.
2.) Or I can add the asset's world height multiplied by asset's (rectangle) height to the screen.y position (again, just for the depth sorting part)
For both cases, I may need to add micro adjustments if I need a specific order within the tile stack itself (seems unlikely). edit: this wont be needed unless I have overlap in the stacking, and I wanted to e.g. have something lower down on the stack be drawn over something above the stack, like an antenna or w/e.
Lastly, this only works if all assets are the same size (thus larger ones are a combination of smaller sprites).
Hope I explained this clearly.
To be fair, I was pilfering loanwords like "screen-space"(adj) from the 3D dialect.
> In order to simulate a camera looking into a 3D world, I need to draw everything from the top of the screen to the bottom
Right, the point I was trying to make was that, technically, you need to draw things from the north edge of the world to the south. If you have a asset with height, you need to draw it at a higher screen position, but also a later time, than a low/short asset that's at norther world position but lower screen position. (Presumably you know that already; I was mostly asking/objecting about confusing the screen position versus the world position.)
> I can add the asset's world height multiplied by asset's (rectangle) height to the screen.y position (again, just for the depth sorting part)
My point was that you'd add[0] the world height plus the world.y position, in order to get the screen.y, but depth sort based only on world.y, independently of height or screen position. That way things that are vertically on top of each other have the same world x and y coordinates (differing only in z), even though they're draw at different screen.y coordinates to provide isometric simulation of height.
0: give or take factors of -1 depending on which way the various coordinate axes go
Most 2d engines use shader system underneath, it easier to write a shader to solved all the z-sorting problems. Only minimal 3d maths and billboarding(tile) are all you need.