Error in the online momentum trend diagnostic

[Robin_Waldman]

Context

I am using the online 3D momentum trend diagnostic (ln_dyn_trd = .true.) to compute offline vorticity budgets. The model version is NEMO v3.6, downloaded a few months ago from NEMO website. I am working on a Mediterranean configuration (NEMOMED12).

Analysis

I have first checked on 1 month of simulation that the total trend stored (utrd_tot and vtrd_tot) corresponded to the sum of individual trend terms (which is the case) and was equal to the actual trend deduced from restart files (un, vn in the restart of the end minus beginning of month). This latter point is not verified. The error is typically low (a few percents) but it can reach very high values at localized points on the map, especially at intermediate depths (~70-300m). The error is even larger when computing the curl of the momentum trend in order to retreive the vertical relative vorticity trend.

Fix

Changeset r8102 (by Dave Storkey) has recently debugued the online tracer trend diagnostic, in particular by separating between trend contributions in even and odd timesteps. I am thinking that maybe the error on the momentum trend diagnostic also comes from that.

[Robin_Waldman]

I have also saved the instantaneous velocities at the end of January and February instead of extracting un,vn from the restarts. In this case I get a similar (although not same) result: the online trend diagnostic (utrd_tot+utrd_atf) has a typical error of a few percents as compared to the u difference between beginning and end of month.