He did mention it was in opti mode, which means it was relying on downward sensors. I'm not sure if that holds well in wind. But being over water, now the waves/ripples, if it even catches them and overrides winds, are thought to be stationary patterns which they are not. Think of the AC tracking a pattern on a moving treadmill or people mover. The AC would follow the movement of the treadmill/peoplemover belt. Same thing over rippling water. And again, that's if it doesn't get overtaken by winds.
I was flying about 30ft or so above water. Several 100ft out, my M2 seems to drift like crazy, even doing a toilet bowl dance. I thought my IMU or something else was starting to fail so I reeled it in via RTH which worked perfectly.
A few days later it occurred to me that the VPS in Mavics is very good, particularly compared to a P3 and it was the VPS that was overtaking the GPS at least to some extent.
I bet the toilet bowl effect was VPS tracking water currents causing drift, and GPS correction taking affect when the VPS was found to be excessive.