Currently the AI system identifies drainage canals as rivers and creates a flooded situation in regions with these types of canals, such as South Florida.
This type of water should constrain to narrow straight shapes, have stagnant appearance, and be darker in color.
Looking at it from an algorithm perspective, looking at an aerial photograph, how would you go about telling a river, and a canal, apart?
The real question is what version of blackshark ai are we using in MSFS. Ai gets updated and grows fast. Look at how many versions already came out from chatgpt, and the others. Im assuming we are using the same version of blackshark ai that MSFS released with. The more AI grows the more itll be able to tell more details apart from sceneries. This could be a simple fix, which is just updating blackshark ai. My personal opinion is update evrrything and then if it doesnt start getting into how it can determine differences in details such as drainage canals and rivers, and even highways lower than the base of communities. Etc
You may be oversimplifying it a tad. Look at the images below. All three of these are rivers, but if you look at these sections, they could easily be mistaken for a canal, or even a dyke.
Here’s a dyke for comparison:
Here is a canal:
Simply put, the water shader doesn’t support narrow water bodies,
In my experience a minimum width of 10 meters is supported
And there is no Blackshark here, all (or pretty much) vector data is from OSM (and there a good mapper should have already tagged the right kind of water body)
When water masking, I’ve usually just removed the smaller canals as they look more natural with just the imagery.
Otherwise they flood the streets and neighborhoods with a post hurricane look.
Given the scenery compatibility with 2024, I suspect we may not see a major overhaul of the AI scenery system. Microsoft has actually been focusing on manual masking work for WUs. I doubt they would be making that effort if something better was on the way.
