CAMELS Hydrology

Prompt

I'm a watershed scientist, and I've read a bunch of papers about LSTMs in the context of hydrological forecasting. Everyone seems to use this dataset called CAMELS or international cousin called CARAVAN, but I've always thought CAMELS may be too coarse of a dataset, because it can be pretty "contaminated" with upstream dams, canals, etc... that change the discharge signal. What investigations have been done about just how "natural" Camels sites are? How can we overcome any of this contamination? Bonus, how does UpstreamTech handle these issues?