Robustness check for time series

so-wise / weddell_gyre_clusters

Unsupervised classification of Weddell Gyre profiles

MIT License

2 stars 1 forks source link

Robustness check for time series #95

Closed DaniJonesOcean closed 2 years ago

DaniJonesOcean commented 2 years ago

Looking at the time series, how does the location of the profiles year-on-year possibly bias the time series? The depth of Tmax might especially be biased/contaminated by the fact that the spatial distribution of the observations is not uniform over time.

Ways to check:

Plot the locations of the profiles, as a histogram
Plot the mean lat/lon and other statistical moments to see spatial biases (idea from Andrew)

DaniJonesOcean commented 2 years ago

The results don't actually look all that good. There's okay consistency in the mean latitude, but the mean longitude has some big variations with time, in the 20-30° range. I guess that's not so huge as far as longitudes go, but it's not ideal...

Shall we try the histogram approach too, just to check?

(In a way this simplifies things. If no robust time-variation is possible, it simplifies the paper)

Unless we...do an OA for each class separately? :O Nooo....

DaniJonesOcean commented 2 years ago

Yep, I'm going to call it - those variations are too large. The variations make the time series difficult-to-impossible to interpret. We need better data coverage to make this work. Closing this issue now.