Automatic BLM calibration

[jeffreyhanson]

The Marxan Good Practices Handbook outlines a method for automatically calibrating boundary length penalties (https://pacmara.org/wp-content/uploads/2010/07/Marxan-Good-Practices-Handbook-v2-2010.pdf, see section 8.3.5). It would be good to investigate this further, and add functionality to prioritizr. Admittedly, it does sound almost too good to be true.

[IsaakBM]

Hi @jeffreyhanson. I've created some R code to deal with the calibration issue that it's based on the Marxan Good Practices Handbook The only problem that I see is that in order to run a calibration procedure you need to have first the output for each "run" (e.g., BLM == 0 & BLM == 1). Then, calculating the "sweet spot" is not a problem (it works actually)

[jeffreyhanson]

Hi @IsaakBM - that's awesome! What has your experience been with this method? Did the automatic BLM value result in a reasonably clumped prioritization? Or was it not clumped enough? Was the automatic BLM really good such that you didn't need to manually iterate over a range of BLMs?

Yeah, the trick with integrating this into prioritizr is that we need some way of making it work automatically for other objective functions (e.g. for the max features obj function we need to automatically set features weights to zero) -- and that not necceasrily straightforward.

[jeffreyhanson]

Just to follow up, we recently wrote up a new tutorial for running calibration analyses (see https://prioritizr.net/articles/calibrating_trade-offs_tutorial.html). It provides R code for running the automatic method discussed here (see https://prioritizr.net/articles/calibrating_trade-offs_tutorial.html#cohon-et-al--1979-method).

[IsaakBM]

Hi @jeffreyhanson

Sorry for my slow (very very slow!) reply. This is great, I'm actually going to try your optimization code.

Thanks!
Isaac

[jeffreyhanson]

No worries at all! Awesome - thank you! Please let me know if you encounter any issues or have any suggestions for improving it?

[IsaakBM]

For sure!