WHAT THE EVIDENCE SAYS
A PRISMA review of 127 studies found high reported accuracies for some models, with CNN methods averaging R² near 0.85. Studies concentrated in China and the United States; chlorophyll was well represented, whereas CDOM and Secchi-depth retrieval were less developed.
WHY IT MATTERS / OUR READ
A strong published R² is not a substitute for independent local validation in an unfamiliar water body.
CAVEATS / UNCERTAINTY
Heterogeneous study validation protocols limit comparisons, and an average performance measure is not proof of global operational reliability.
SOURCE
Remote Sensing in Earth Systems Sciences — Rathupetsane et al. (29 Oct 2026)
View original evidence ↗