ohw23_proj_argo_ml

1. Get argo data

Argopy: https://argopy.readthedocs.io/en/latest/

• Rob: I'll help work on this

• Define which data we are going to use Temperature and salinity

• Define limits for Argo data: locations, time

Rob: ArgoPy can select for a time period and in a rectangular area using lat/long bounds and pressure bounds. Alternatively can select specific floats or profiles.

North Atlantic

Last year of data

2. Prepare data for the Machine Learning Model

Prepare the data vertically to be equal gridded. Use of numpy,interp?

The input data needs to have a format similar to each:

• a dataframe or a csv file
• each row represents a profile
• each column represents a data value in a specific depth
• you can add two more columns on the data related to the position of the profiles

Need to turn full Argo data DF to two dataframes, one for temp and one for salinity, with the following columns: LATITUDE, LONGITUDE, and each unique value of PRES

• Each row is a unique lat/long pair, and has the TEMP or PSAL at the depth corresponding to each PRES column

3. Generate spikes on the data

Add some random noise to the GDAC data

4. Apply IOOS QC on the "fake" data

If we have time, can use the ioos_qc module's qartod.spike_test function.

Try to find the thresholds for each test

• https://www.ukargo.net/data/quality_control/

• https://cdn.ioos.noaa.gov/media/2020/03/QARTOD_TS_Manual_Update2_200324_final.pdf

• https://ioos.github.io/ioos_qc/

• Spike Test -> IOOS Threshold is 3 C for suspect, 8 for fail

• Gradient (Argo) / Rate of Change (IOOS) -> Threshold of 3*StnDev

• Global Range / Gross Range -> Min/max for all ocean. Will need "unrealistic" spikes to properly test.

• Regional Range / Gross Range -> Min/max for region. Will need "unrealistic" spikes to properly test.

• Frozen / Flat Line -> 'Very small deviations', we would need to come up with a value. May not be necessary to run in addition to Gradient

• Stuck Value / Flat Line -> Threshold of 0. May not be necessary to run in addition to Gradient.

Flat Line and Rate of Change Tests are for a time series of data - not really appropriate to compare data within a single profile.

5. Define some configurations on the ML model

09AUG - talk about that

6. Apply the ML model

09AUG - talk about that

7. Prepare the final result

Jupyter notebook with all the steps

(Optional) Prepare the data for the Dense Neural Network (DNN) model

(Optional) Apply the DNN model