seafloor_action='previous' allows z to drift below seafloor

[jordanasevigny]

Hello,

I am running particles backwards in time in ROMS using the OceanDrift model and have a question about how the seafloor_action='previous' setting works.

When my particles hit the seafloor, it appears that latitude and longitude stay fixed at the last position before contact with the seafloor, which makes sense. However, z continues to change, and w, u, and v are still being recorded (even below the seafloor). As a result, if I plot a time series for a particle that hits the seafloor, z continues responding to changes in w, even though lat and lon are static.

Because z does not remain fixed at its last valid value before contact, I believe this causes an issue: if w remains negative for several steps, z can become unrealistically deep (e.g., 10,000 m along the coast). Eventually, when w becomes positive, the particle appears to “unstick” — but only once enough upward movement has accumulated such that z returns above the seafloor depth (total w since hitting the seafloor must be positive).

I had assumed that with seafloor_action='previous', the entire position vector (lon, lat, and z) would be held constant until the particle is allowed to move again. That is, I had expected the particle would begin moving again as soon as w becomes positive — but that doesn’t appear to be the case if z continues to drift downward.

Additionally, I’m puzzled by the fact that positive w values are still interpolated and assigned while the particle is well below the seafloor. I suspect this might be related to how the velocity interpolation clips z values to the ROMS grid (e.g., a very deep z gets snapped to the deepest layer), but I’m not entirely sure.

Is my understanding of how seafloor_action='previous' works correct? And if so, is there a recommended way to keep z fixed as well until the particle re-enters the water column?

Thank you for your time and insights,
Jordana

[knutfrode]

I believe your interpretation is very right: the vertical coordinate has hitherto been neglected when moving particles back.
See the actual code here:
https://github.com/OpenDrift/opendrift/blob/master/opendrift/models/basemodel/__init__.py#L740

And yes: ocean model 3D fields are extrapolated downwards, and also beyond actual seafloor depth. This in itself I believe makes sense, to avoid having no values and a possible gap between the lowest model layer and the actual seafloor depth where particles could be trapped.

But the issue with also resetting z to previous value should be fixed.
There are some changes in the pipeline regarding the "previous" mechanism, but it might take some weeks until this is out.
In the meantime, you could try to copy this line:
https://github.com/OpenDrift/opendrift/blob/master/opendrift/models/basemodel/__init__.py#L739
(self.elements.z[below] = -(sea_floor_depth + sea_surface_height)[below])
and insert after line 747.
This should fix the particle depth to the seafloor depth of the interaction point - which is not strictly correct - but should be quite close - and at least it should remain fixed.

[jordanasevigny]

Thank you very much! Your recommended temporary fix appears to be working for me.