Start building an ML model

[iwandejong]

Preprocessing

• [x] NDVI (image pentadal)
• [x] PET (image pentadal)
• [x] CHIRPS (image pentadal)
• [x] Cloud Cover (monthly)
• [x] Ground Frost Frequency (monthly)
• [x] Precipitation (monthly)
• [x] Temp Min (monthly)
• [x] Temp Max (monthly)
• [x] Temp Mean (monthly)
• [x] Vapour (monthly)
• [x] Raining Days (monthly)
• [x] Growing Season (Seasonal)
• [x] Crop Stage Index (generated based on department research)
• [x] Historic Crop Yields (various crops)

Combining everything:

• [x] Make a table in Supabase to combine all the statistics into one dataset (means and stdev for each data point)
• [x] Interpolate all the data to be pentadal
• [x] Correlation Tests
• [x] Statistical Feature Selection (R/Python)
• [ ] Make a document to describe the model so it can easily be worked on
• [x] Start building the model (regression, forest, gradient boosting, neural network - in research)

Progress: https://lucid.app/lucidchart/9f71ed4a-68f7-4a86-8a67-66fbcb5485c8/edit?viewport_loc=-770%2C-21%2C2820%2C1362%2C0_0&invitationId=inv_9de32da7-f419-411b-a439-5f48bd857737

[iwandejong]

All preprocessing facilities are available. Problem is time it takes to analyse all of them.

• PET: 40 minutes
• CHIRPS: 40 minutes
• NDVI: 14 hours

This all needs to be done uninterrupted to ensure data validity on Supabase.

[iwandejong]

Built a basic random forest model on wheat. Seems to work pretty well, but will fine-tune as time progresses to ensure that there's no bias or other factors affecting the model