Eddy tracking about py-eddy-tracker

[AntSimi]

First tracking use only detection file, it doesn't care of grid format

General information

area_tracker associate observations only if there is an overlap greater than 20 % between contour of observations (use effective/outter contour)
Code for area tracker here
Match function
overlap function

default tracker is based on distance between observations center, and check if amplitude and radius are close (maximal factor of 2.5 between the two observations)

Now, i use only AreaTracker which are more able to follow eddies during merging/spliting events.

General questions:

1. Which tracker did you use it for those figures?
2. How many days are display? 5 or more ?

Case

I guess yaml was set like that

PATHS:
  # Files produces with EddyIdentification
  FILES_PATTERN: MY_IDENTIFICATION_PATH/Anticyclonic*.nc #(5 files)
  SAVE_DIR: MY_OUTPUT_PATH
# Number of timestep for missing detection
VIRTUAL_LENGTH_MAX: 0
# Minimal time to consider as a full track
TRACK_DURATION_MIN: 3

• Case 3 : i didn't see your identification file, but i think there are days between the untracked observation and the short track, maybe if you use VIRTUAL_LENGTH_MAX option you will join those observations, this options allow missing detection between few days
• Case 1 : How did you do display? Currently i wonder why i saw only one contour on track subplot(top right)
• Case 2 : Like case 1 i need more informations to understand, maybe you could share identification files and tracking configuration and tracking atlas?

You have some general informations about tracking here.

[Ashwitaa]

I used 'default tracker' with display of only 5 days for the figures I have attached before.
Yes, the yaml file is set like the way you've mentioned above.

How did I do the display?:--- By using effective contours in lat and lon

I also have made few tests using different regions and in some cases I see results like Case01 and Case02.

I have attached the identification and tracking files, and the tracking atlas.

files.zip

[AntSimi]

Before to look at tracking, i do a quick look on identifcation files:

from matplotlib import pyplot as plt
from py_eddy_tracker.observations.observation import EddiesObservations
from glob import glob
from os import path
import pylook
fig = plt.figure(figsize=(10,11))
ax = fig.add_axes([.05,.05,.9,.9], projection='plat_carre')
ax.set_xlim(-100,30), ax.set_ylim(-60,75), ax.grid()
anticyclones = glob('files/identification_files/A*.nc')
cyclones = glob('files/identification_files/C*.nc')
for f in anticyclones:
    EddiesObservations.load_file(f).display(ax, color='r', ref=-100, label=path.basename(f))
for f in cyclones:
    EddiesObservations.load_file(f).display(ax, color='b', ref=-100, label=path.basename(f))
ax.legend()
fig.savefig('display.png')

I am surprised to see so few eddies in atlantic ocean. I will look at tracking step

[Ashwitaa]

I forgot to inform, that the eddies what we see here are at 1500 m depth.

[AntSimi]

• Case 1 :
  There is a jump between two structure (due to weakness of default tracker) which are a mix of distance, radius similarity and amplitude similarity, default cost fuction code
  
  There are two missing detection (9/7/2007 and 11/7/2007)
  But only one observation was used in a path and the other go in untracked: My advice change tracker and use AreaTracker

from matplotlib import pyplot as plt
from py_eddy_tracker.observations.tracking import TrackEddiesObservations
import pylook
fig = plt.figure(figsize=(10,10))
ax = fig.add_axes([.075,.075,.85,.85], projection='plat_carre')
ax.set_xlim(-12,-9), ax.set_ylim(37,40), ax.grid()

d = TrackEddiesObservations.load_file('files/tracking_files/Cyclonic.nc')
d.plot(ax, ref=-100, label='Cyclonic path')
d.display(ax, ref=-100, label='Cyclonic contour', extern_only=True)
m = d.scatter(ax, d.time - 21000, ref=-100)
c=plt.colorbar(m, cax=fig.add_axes([0.94, 0.05, 0.01, 0.9]))
ax.legend()
fig.savefig('display_case1.png')

• Case 2 :
  Similar to case 1 but with more complexity, again change tracker to areaTracker will avoid this type of problem

from matplotlib import pyplot as plt
from py_eddy_tracker.observations.tracking import TrackEddiesObservations
import pylook
fig = plt.figure(figsize=(10,10))
ax = fig.add_axes([.075,.075,.85,.85], projection='plat_carre')
ax.set_xlim(-12.5,-9.5), ax.set_ylim(34,37), ax.grid()

d = TrackEddiesObservations.load_file('files/tracking_files/Anticyclonic.nc')
d.plot(ax, ref=-100, label='Cyclonic path')
d.display(ax, ref=-100, label='Cyclonic contour', extern_only=True)
m = d.scatter(ax, d.time - 21000, ref=-100)
c=plt.colorbar(m, cax=fig.add_axes([0.94, 0.05, 0.01, 0.9]))
ax.legend()
fig.savefig('display_case2.png')

• Case 3 :
  there are a missing detection on 9/7/2007, use virtual will allow to link all the observations together
  

[AntSimi]

In summary maybe use AreaTracker and virtual (set maybe between 1 to 5 days, to avoid missing detection)