kevinkuruc / DICEFARM

0 stars 1 forks source link

Double check stability of results with pulse size. #11

Open FrankErrickson opened 4 years ago

FrankErrickson commented 4 years ago

Just working with the base FAIR model, adding a ton of carbon (1e-9) would give weird impulse response behavior. But slightly increasing the pulse size made the results consistent. Just double check that there are no numerical stability issues with DICE-FARM (doesn't seem like it, but just an easy mistake to avoid).

kevinkuruc commented 4 years ago

It does happen in DICE-FARM. I just made the impulse bigger, and then scaled it back down when plotting/reporting social costs. That seems a fine workaround, yeah?

FrankErrickson commented 4 years ago

Yea that's the typical approach. For instance, just add a GtC, estimate the SCC as $/GtC, then scale it down to be $/ton C. It'd be good to just check a variety of pulse sizes to make sure it's stable. I'm not super sure what's making it unstable for a single ton, other models have been stable. There is a part of the carbon cycle where you approximate the roots of an equation, so maybe you just get outside of that algorithms precision.

Did fixing this in DICE-FARM change the results at all?

kevinkuruc commented 4 years ago

Not sure I ever went through with calculating SCC for very small changes when I saw how wonky temperature IRFs were for a one-tonne pulse. The model can't detect a welfare change for $1 anyway, so I just skipped right to using 1000 tonnes CO2 for numerator of SCC and 1000 dollars for the denominator (doing it the Nordhaus way, not discounted sum; though I checked I get the same answer in DICE2017 so this method seems fine to me).

I think I tried different scaling up (for values other than 1000) and kept getting same SCC, so it should be stable, but I'll verify that tomorrow.