I would give it a go. Water absorbs infra-red light very effectively to the extent that water appears black in the infra-red spectrum. Hence thermal cameras based on far infra-red are effectively detecting surface temperature of water. However, I doubt the camera's you are looking at operate exclusively at far infra-red. Near infra-red should work in shallow water, and there are a lot of cameras that cover that spectrum (anything with the IR-block removed). What might take more work is finding the right LED's for illumination. You probably don't want to spook the fish, so need to illuminate beyond their spectral range (>700nM?), but not too high, otherwise light absorption by the water will be an issue. Getting the angle right on those lights to minimize glare will take a bit of trial an error - but generally speaking locating the light right beside the camera is the worst spot. Try 45 degrees to surface as a start (or underwater illumination). Same applies during the day - mid angle sunlight generally gives you less glare (e.g. mid morning in summer). Polarized filters reduce glare, rather than eliminate it. Worth a try.

And this article will help Article Underwater infrared video system for behavioral studies in lakes

and some of the principles here

https://en.wikipedia.org/wiki/Electromagnetic_absorption_by_water

Recent development in image processing (e.g. machine learning) might help you with the many hours of video you end up with.

Did you consider hd echo sound imaging? There is some facinating technology available which results in precise images inc. shadows. Echo sound imaging is independent of light, so works perfectly at night or in milky water. However, images are black-white only.