From Forum: Deterministic SCC Result

[lrennels]

moved from forum

https://forum.mimiframework.org/t/first-time-user-calibration-of-scc-estimates/313

I am a first-time user and trying to get acquainted with both Julia and MimiGIVE. I was able to install Juia, add packages and run the default model. However, when I run the default model, I got result that seemed far too low. I was expecting to get something closer to the $185/ton that is found the SCC explorer tool. Any tips for calibrating my model to make sure it’s running correctly? I was following sections 2a (fully coupled model) and 4a (deterministic scc calculation) on the Github repository.

scc = MimiGIVE.compute_scc(year=2020)
26.166523152489944

scc = MimiGIVE.compute_scc(year=2030)
38.34888259268095

[lrennels]

Good question! Our deterministic SCC calculation calculates the SCC based on one set of parameters, but it is not a central value of the full distribution, since many of the key uncertainty parameters, like the socioeconomic pathway and the climate model parameter set, do not have a "central" run. What you're doing is correct, and your confusion is fair! In order to get the reported value (within the standard error) you'll need to run a Monte Carlo. The deterministic calculation can be helpful for testing, debugging, etc. but will not give a sense for the central tendency of the full distribution.

[afoxwolfe]

Hi Lisa,

Thanks so much for your reply here and the follow-up on the forum! It's incredibly helpful to get your directional feedback in the last response.

I have taken some time to play around with the monte carlo simulations. When I run the basic monte carlo model using:

• RFF SSPs
• n = 1,000
• prpt of 0.002 and eta =1.24 (in line with a 2% discount rate using the Ramsey Discounting method)

I get a central SCC estimate of ~$146 with a standard error of ~$3. Does this seem right to you? I'm trying to piece together why my social costs using these parameters might be approximately $40 lower than those found in the SCC explorer and the Nature paper. Would I need to couple with MimiCIAM? I saw some documentation on that in the README.

> julia> discount_rates = [(label="Ramsey", prtp=0.002, eta=1.24)]
> julia> result = MimiGIVE.compute_scc(year = 2020, discount_rates = discount_rates, n = 1000)
> julia> for (k,v) in result[:scc]
>            println("Specification: $(k.region) SCC in $(k.sector) sector, using discount rate $(k.dr_label) specified by prtp = $(k.prtp) 
>            and eta = $(k.eta):")
>            println(" --> Expected SCC = $(v.expected_scc) with standard error $(v.se_expected_scc)")
>        end
> Specification: globe SCC in total sector, using discount rate Ramsey specified by prtp = 0.002 and eta = 1.24:
>  --> Expected SCC = 145.98150071009536 with standard error 3.148607640363834

Additionally, I eventually would love to calculate damages on a country-by-country basis. If you have guidance on how to do that, I would appreciate that immensely! To that end I'm curious if the damages are calculated on a country-by-country basis or if damages are calculated globally and then disaggregated. If MimiGIVE does the latter, would plugging in MimiRICE allow me to do the former? I am helping a professor at Yale School of the Environment develop a course on Climate Ethics and Modelling and we are putting together quantitative examples on inter- and intra- generational equity. I'm guessing this question might be better suited for the forum and can post there if you agree!

[lrennels]

Hi @afoxwolfe

Your work looks great here, and the code looks correct to me. I think the pesky difference you are seeing is that results are reported in USD $2005. In the Nature paper code you will see a file called price_level_inflator.jl with

const pricelevel_2005_to_2020 = 113.648 / 87.504

plugging that in you'd have an expected SCC of about $188.30 for an n of 1,000 (which is a good n value to do your iterations, we use n of 10k for final runs but 1k tends to be enough for iterative work).

MimiCIAM is automatically coupled in for the compute_scc function, so you're all good there. It isn't when you do a more simple get_model --> run_model call.

[lrennels]

Per the country-by-country damages, by damage function GIVE produces

• Agriculture at the FUND region level (16 regions) (Moore et al., 2017)
• Mortality and Energy at the country level (184 countries) (Cromar et al., 2022 and Clarke et al., 2018)
• Sea-level rise at the segment level (~12k coastal segments from the DIVA database) (Diaz 2016)

You can get a general sense for the distribution of impacts from the individual papers for each, if you don't want to rerun the model, since they all contain some description of the distribution. If you do want to get the actual damage values from the GIVE model there is definitely some support for that that I can point you too -- it is just a bit memory intensive (large files are produced and not aggregated over time or discounted for you, so there's just a bit of work to do) so if it's for a class it could be easier to use the visuals from the respective papers -- but I'm happy to point you to the MimiGIVE functionality if you'd like to do that!

[lrennels]

@afoxwolfe Assuming your questions here have been answered I'll close this for now, but be in touch with follow ups -- either with a new issue or if it's linked to these questions just make sure to @ me so I see it

[sleepguyjin]

Hello both @lrennels , thank you so much for developing a very helpful interaction here. I was wondering if I may jump in, as I am also interested in the analysis on a country-by-country basis, and the motivation seems quite related. Please let me know if I should start a new thread.

When playing around with the MCS SCC, I added the compute_disaggregated_values flag, but I got an error message that indicates that this flag is unsupported (yet). This seems related to the fact that this is in a beta version, as suggested in the readme, and also related to your comment that country-level damage values are quite memory-intensive, in particular for many MCS runs. Would it be possible to access the country-level SCCs ---along with their RFF-SP- emissions--- by adding some tweaks to these lines? (N.B. Without this particular flag, the code seems to be working.)

using Mimi
using MimiGIVE

discount_rates = [(label="Ramsey", prtp=0.015, eta=1.45), (label="Constant 2%", prtp=0.02, eta=0.)]

result = MimiGIVE.compute_scc(year = 2020, discount_rates = discount_rates, n = 5, compute_disaggregated_values = true)