RE: Update on boxplots

[brenwin1]

I manage to extract for each grid cell, the number of fire ignitions per year in the historical data (). I have done up a dot plot of count against each year for each cell (can be found in historical data boxplots) section.

[brenwin1]

Hi everyone,

Wanted to provide the latest updates and findings and some suggestions for next steps we are working on. To summarise the findings from the meeting:

(I) comparing historical (2000-2018) vs. satellite ignitions (2019-2020)

• blue cells: historical data > satellite ignitions
• red cells: satellite ignitions > historical data

(II) Plotting yearly ignitions in historical data (correspond to dot plot above)

• plotting the historical fire ignitions from 2000 to 2018

  

• comparing to satellite fire ignitions from 2019 to 2020

  

*both plots are filtered to months 10 (Oct) to 3 (Mar) for comparability

findings

• most cells have no/few ignitions i.e. low level of probability

• notable outlier: cell 208

  • from 2000 to 2007 (one of event)
  • arsonist in Bendigo

• where historical ignitions > satellite ignitions

  • particularly cells around 251, 252, 271, 272 area (slightly NorthEast of Melbourne CBD)
  • there was a spike. in ignitions from 2014 to 2018
  • ??? more consistent personnels at fire towers?

• where satellite ignitions > historical ignitions

  • two remote areas identified
  • one in the SouthEast (cells 276 & 277 area) and North of that (cells 196, 216 area)
  • some historical fires occured but unable far less than what was in satellite data
  • one in SouthWest (cells 305, 306)
  • NO historical fires occured (found that there were no fire towers around that area)

(III) Suggested Next steps

• Use median of the historical data (potentially) *need to compare now that we got satellite data from 2016 to 2018

  • below plot of median historical data (points not overlaid to avoid overcrowding)

• Run simulation

  • simulation is ran to produce probability of bushfire in each cell.

  • Before adjustment (simply based on historical bushfire counts). These are simply in each cell.

  • After adjustment (made adjustment with with more information)

• before adjustment

  • mesh median historical data in simulation with satellite data
  • find a way to properly inflate/deflate ignitions counts

• adjustments *research

  • ask questions like:
  • What is the consistency of personnels at fire towers? (e.g. NorthEast area; very close to Melbourne, more personnel indicating higher accuracy so increase probability)
  • What is the fuel type? (e.g. flammable fuel increase probability// not flammable; decrease probability)
  • outliers (e.g. decrease probability in upcoming years; after arson perpetrator caught)