Antoine Delepoulle

If you are just interested to compute statistic on lifetime you could use `nb_obs_by_track` property of your [Tracking observation object](https://py-eddy-tracker.readthedocs.io/en/latest/_autosummary/py_eddy_tracker.observations.tracking.TrackEddiesObservations.html#py-eddy-tracker-observations-tracking-trackeddiesobservations). This property is an array give number of eddy observation...

To get isoline with use matplotlib contour function, and this [method change in last version](https://matplotlib.org/stable/api/prev_api_changes/api_changes_3.8.0.html#contourset-is-now-a-single-collection). Use a version of matplotlib < 3.8.0, until we manage this issue.

could you share ncdump -h of your file?

First `data.get_demo_path` method is only to get demonstration sample, you don't need it for your own data. You need to specify height and speed unit in [eddy_identification method](https://py-eddy-tracker.readthedocs.io/en/stable/_autosummary/py_eddy_tracker.dataset.grid.GridDataset.html#py_eddy_tracker.dataset.grid.GridDataset.eddy_identification) if there...

You must cast date in datetime python format : https://docs.python.org/fr/3/library/datetime.html#datetime-objects

Hi Federica, take look at your timeserie, it seems that some daily identification are missing. ``` >>> 15811200 / 86400 183.0 ``` I guess there is 183 daily missing identification...

Several things come to me, you need to apply a large scale filter to remove large scale structure like in [this example](https://py-eddy-tracker.readthedocs.io/en/latest/python_module/02_eddy_identification/pet_eddy_detection_ACC.html#figures). Input grid have which resolution?

My first guess. Contour detection algorithm could be high sensitive to high frequency data or noise. You could try to apply a low pass filter to smoth/simplify a little bit.

If you divide domain, result are different? i don't like this algotrithm behaviour., There is something to dig.

Could you share script use to run eddy identification