What would be the best way to normalize the gradients?

[vsilvafelipe]

Hi, everyone. @mayajakes and I briefly discussed some approaches to normalize the gradients in SSH (or ADT). We thought of dividing the max of the whole time series at each coordinate or in the max at each day. Does anyone have some ideas about other approaches?

[sophieclayton]

If you look at this paper, there is a great section on "Practical Definitions of Fronts" on page 4.

I think that you have a couple of options:

1. Define some threshold value for the SSH (SST, Chl, whatever you want) gradient. Then you just define points above that threshold as "frontal".
2. Use some statistical method to identify fronts, for example look at the PDF of SSH gradient, and define some subset of the most extreme values as "frontal".
3. Another method that has been used a lot is edge-detection (which I know nothing about), but @ocefpaf put this nice notebook together using OpenCV to detect fronts in SST.

[vsilvafelipe]

Hi, @sophieclayton. We thought of using a normalization to avoid using the threshold value. It would help the user to choose the sensibility for the choice of the position of the front. Anyway, at this moment, I imposed a threshold value of 2*1e-6 for the SSH gradient. It is in the arguments of the match_obs_alt function and can be changed by any user. In case we follow with this function, I will try to implement the statistical method (PDF) to define the most extreme values.