Differences in modelled FES elevations between pyTMD=2.1.4 and pyTMD==2.1.5

[robbibt]

I recently noticed that some of our FES2014 integration tests are failing, and realised that pyTMD is producing different results in 2.1.5 vs 2.1.4, sometimes by over 10 cm:

!pip install pyTMD==2.1.4 --quiet

from pyTMD.compute import tide_elevations
import pandas as pd
import numpy as np

tide_elevations(
    x=[122.269294],
    y=[-18.057430],
    delta_time=pd.date_range("2022", "2023", periods=5),
    DIRECTORY="/gdata1/data/tide_models_clipped/",
    MODEL="FES2014",
    EPSG=4326,
    TIME="datetime",
    EXTRAPOLATE=True,
    CUTOFF=np.inf,
)

>>> [1.6378203716778157, 1.4584970325019366, -2.682039596186077, 3.4367875303700806, -0.07229756252779537]

!pip install pyTMD==2.1.5 --quiet

from pyTMD.compute import tide_elevations
import pandas as pd
import numpy as np

tide_elevations(
    x=[122.269294],
    y=[-18.057430],
    delta_time=pd.date_range("2022", "2023", periods=5),
    DIRECTORY="/gdata1/data/tide_models_clipped/",
    MODEL="FES2014",
    EPSG=4326,
    TIME="datetime",
    EXTRAPOLATE=True,
    CUTOFF=np.inf,
)

>>> [1.627442336610703, 1.545339524166975, -2.744346618968223, 3.3916417087573367, 0.03555559012406023]

Longer comparison scatterplot:

I'm not yet sure if the 2.1.5 results are better or worse - just that they are different. I haven't tested this exhaustively across all the models, but so far FES models are the only ones that seem different (EOT20 and TPXO9 seem unchanged).

[tsutterley]

That's not good. I'll try to track it down.

[tsutterley]

Hey @robbibt,
A few separate things happening, all of which are related to when I updated arguments.py to include (a lot) more constituents. Some of these changes are permanent as they're fixes to old bugs. Some are new bugs that will be fixed in #336.

1. The Doodson arguments for eta2 have been corrected. This only affected when inferring from minor constituents.
2. I am inferring L2b when L2 is known. In 2.1.4, I didn't infer this constituent if L2 was available.
3. I didn't pass a keyword argument for the recursive calculation of compound tides. So the arguments were the default ones, and not the arguments for FES models. This affected several constituents, and should be fixed in the PR.
4. I was overwriting the argument for J1 to use the default and not the one for FES models. This should be fixed with the PR.

Thanks for pointing out the problem. I don't think I would have caught it without the nudge. I have a few things I need to check, but I think with the bug fixes I'll put out a release sooner rather than later.

[robbibt]

Fantastic, thanks @tsutterley - this wasn't a problem for us, just unexpected so I thought it was worth raising. Glad to hear it helped clear out some hidden bugs!