Innovation Sprint: Fires per week prediction

**As a** Predictive Services Specialist

**I need** technology to perform and visualize advanced data analytics

**So That** I can identify patterns and trends in current fire season activity, identify historical analog years, and predict fire activity over the next two weeks.

**Acceptance Criteria**
- [ ] Given (Context), When (action carried out), Then (expected outcome)
- [ ] Given (Context), When (action carried out), Then (expected outcome)

**Additional Context**
- An example graph showing weekly fire detections for 2023 over percentiles from the past 20 years can be viewed here: https://bcgov.sharepoint.com/:i:/r/teams/01324/Shared%20Documents/General/_Innovation%20Sprints/2023-Jul/weekly_fires_over_20-yr_percentiles.png?csf=1&web=1&e=TlkTid
- R code and the supporting data used to create the example graph can be found here: https://bcgov.sharepoint.com/:f:/r/teams/01324/Shared%20Documents/General/_Innovation%20Sprints/2023-Jul?csf=1&web=1&e=HY4ADM
- Most of the code deals with formatting the CSV file exported from WF1 Business Intelligence (JasperSoft). The CSV export repeats the header every 30 rows. There's also lots of code devoted to plotting the data on a graph. The calculation of historical percentiles and identification of analog years can be found between lines 100 and 213. Lines 126 to 144 computes statistics for reporting purposes but these are not actually used in the weekly fire detections graph.
- I use a rudimentary approach to determine analog years on lines 202 to 213. For all historic years, I calculate the weekly difference in detected fires compares to the current year and add up the absolute values. The years with the smallest week-to-week difference are identified as closer analogs. There is no comparison to environmental or fire weather variables in the historical years as compared to the current year.
- I don't have the ability to predict weekly fire detections two week s into the future but would like the ability to do so. If that were possible, I would also like to predict the portion of new fires that will escape initial attack. This would likely need to be a regional analysis that gets rolled up to a provincial summary.
- Current year fire incident locations can be found here: https://catalogue.data.gov.bc.ca/dataset/fire-locations-current
- Historical fire incident locations can be found here: https://catalogue.data.gov.bc.ca/dataset/fire-incident-locations-historical
- Historical and current year fire weather data can be found here: https://www.for.gov.bc.ca/ftp/HPR/external/!publish/BCWS_DATA_MART/
- Fire weather data can also be accessed through a public API, but large queries are not permitted. Instructions for the public API can be found here: https://bcgov.sharepoint.com/:b:/r/teams/01324/Shared%20Documents/General/_Innovation%20Sprints/2023-Jul/BCWS%20Datamart%20and%20API%20%20V2.pdf?csf=1&web=1&e=MR6HiL

**Definition of Done**
https://github.com/bcgov/wps/wiki/Definition-of-Done