search

CIGLR-ai-lab / GreatLakes-TempSensors

Collaborative repository for optimizing the placement of temperature sensors in the Great Lakes using the DeepSensor machine learning framework. Aiming to enhance the quantitative understanding of surface temperature variability for better environmental monitoring and decision-making.
MIT License
0 stars 0 forks source link

Implement Density Channel Masking Approach for DeepSensor #35

Closed DaniJonesOcean closed 1 month ago

DaniJonesOcean commented 2 months ago

Task Description:

Try out a new density channel masking approach in DeepSensor by creating a field where the value is 1.0 at buoy locations and 0.0 elsewhere. This field should be formatted as an xarray Dataset, like the other fields, and then used as a new context set in DeepSensor. The goal is to analyze the representation of the covariance structure using only values from buoy locations.

Checklist:

  1. Create the Density Channel Mask:

    • [ ] Identify the buoy locations and obtain their coordinates.
    • [ ] Generate a 2D field where the value is 1.0 at buoy locations and 0.0 elsewhere.
    • [ ] Format this field as an xarray Dataset, ensuring it is consistent with other input fields used by DeepSensor.

  2. Integrate the Density Channel Mask into DeepSensor:

    • [ ] Configure the DeepSensor environment to include the new density channel mask as part of the context set.

  3. Train and Evaluate DeepSensor with the New Context Set:

    • [ ] Train the DeepSensor model using the updated context set that includes the density channel mask.
    • [ ] Monitor the training process and document any improvements or challenges observed.
    • [ ] Evaluate the model's performance with the new context set and compare it to the baseline model to assess the impact of using buoy location values exclusively.

  4. Document the Process and Results:

    • [ ] Create a Jupyter notebook to document the process of creating the density channel mask, integrating it into DeepSensor, and analyzing the results.
    • [ ] Save the trained model and relevant output files.
DaniJonesOcean commented 1 month ago

NOTE: Task is to convert a set of (lat,lon) positions into a 2D mask, with 1.0 at the grid cells with buoys and 0.0 otherwise. This will contain some implicit representation of the grid (grid size). [Possibly look for alternative methods?]

DaniJonesOcean commented 1 month ago

@eredding02 Could you direct me to the files with the buoy locations?

DaniJonesOcean commented 1 month ago

I'm going to suggest making this "inactive" by closing it, as to cut down on the number of things in front of us in this final week of analysis. We'll keep it in mind for possible future work.