Window size for the eddy network (EddyNetworkGroup & EddyNetworkBuildPath)

[Ashwitaa]

Hi Antoine,

I had a few issues and questions for the network files, basically how the "window" is working for the network files… It's natural that with every different window size (for grouping and building path) there're different segements and merging/splitting networks (in terms of count) but, it's not clear which value we should pick and how they are acting in the alogrithm for creating the network

Below I have following questions. Could you please try helping me in understanding these?

--- The window issue:

• When does the window parameter refers to FILES and when does it refers to TIMESTEPS (ts)? (e.g., daily ts == number files, 12h ts==2*files, 5-day ts==files/5)
• How to better choose a window size (when you have output like e.g., 12h or 5 days). Does it depends on the eddy characteristics (size, lifetime...)?
• What are the consequences of choosing a too big or too small window for building a group / network and for merge/split/segmentation

-- extra

• what are the segments that are part of a network but don't participate in any merge nor split (not connected to other segments)?

Thanks…

[AntSimi]

I understand it could be confusing.

• When you use EddyNetworkGroup window argument must be given in number of files.
• When you use EddyNetworkBuildPath window argument must be given in time.

Window argument is given for all observations. It must be a compromise, i prefer value under 7 days. Too tiny window could cut track/network in several part. Too large will agreggate eddies with no interactions. There is no obvious choice ... But window isn't the only feature to manage interaction there is also overlap ratio which could be modified in mode minimal/union/hybrid and also you could play with min_overlap.

We try to validate our choices with lagragian particles like show in OSTST poster and presentation

When you divide your Group/Network in Path you could get lonely segment, with no spliting or merging identified. One of observation of those segments have an overlap with network observations but this overlap ratio was not the first candidate. There is like a touch between two segment.