Project SalmonSpy

Background:

The world's largest salmon run occurs in Bristol Bay, Alaska every summer. The exact timing and magnitude of the salmon run is unknown, but predictions are made by ADFG at the start of every year. In 2018 a record 62.3 million salmon returned, 21% more than predicted by ADFG. The goal of this project is the develop a more accurate prediction of the Bristol Bay salmon run then what's currently provided by ADFG.

Work Items:

1. Data fetch/scrape
   1. fishing boat activity = 'http://globalfishingwatch.org/'
      • Docs and SDK = https://github.com/GlobalFishingWatch?language=python
      • More technical docs: https://github.com/GlobalFishingWatch/paper-global-footprint-of-fisheries/blob/master/data_documentation/fishing_effort.md
   2. historical salmon catches = 'http://www.adfg.alaska.gov/index.cfm?adfg=commercialbyareabristolbay.harvestsummary'
   3. historical salmon counts (what escapes upriver) = 'http://www.adfg.alaska.gov/index.cfm?adfg=commercialbyareabristolbay.salmon#fishcounts'
   4. historical salmon catches in "test" fishery = 'https://www.bbsri.org/'
   5. stream and lake levels = 'https://waterdata.usgs.gov'
   6. tide activity = 'http://kapadia.github.io/usgs/reference/api.html'
   7. land weather = 'https://graphical.weather.gov/xml/'
   8. marine weather = https://www.ncdc.noaa.gov/data-access/marineocean-data
   9. more marina data to checkout = https://www.ncdc.noaa.gov/cdo-web/datasets
   10. NOAA web data services guide = https://www.ndbc.noaa.gov/docs/ndbc_web_data_guide.pdf
   11. tides & currents = https://opendap.co-ops.nos.noaa.gov/

Station IDs

Port Moller         9463502
Adak Island         9461380
Unalaska            9462620
St Paul Island      9464212 
> more here:    https://opendap.co-ops.nos.noaa.gov/stations/index.jsp

1. Data clean and transform

   1. Verify data integrity
   2. Standardize column names
   3. Convert all files to CSV

2. AWS setup

   1. Spin up micro EC2
   2. Create an S3 bucket
   3. Setup Sagemaker

3. Train the model

   1. Use built-in algorithms with Sagemaker
   2. Pre-process with Jupyter

4. Evaluate the model