CIAM Uncertainty Propagation

This repository accompanies a study by Wong, Ledna, Rennels, Sheets, Errickson, and Anthoff (2021) using the MimiCIAM implementation of the CIAM model described in Diaz (2016) ("Original CIAM").

Relative to Original CIAM, MimiCIAM:

• uses updated forcing via the Shared Socioeconomic Pathways (SSPs),
• uses updated sea-level rise projections,
• includes a bug fix that does not allow flood defense heights to be lowered over time,
• and implements a limited decision-maker foresight regarding future sea-level rise.

Getting started

Obtain the git repository for this project

First, clone or download the CIAM_uncertainty_propagation repository (that is, this repository). Once this is on your computer, open the Julia console and set this folder as your working directory in the Julia console. You can also obtain a snapshot of this repository that includes all relevant results files and older CIAM results files from Zenodo here: https://zenodo.org/record/5218888.

Installing required packages

This code was created using Julia v1.5 and requires several Julia packages.

To install these packages, first enter the package manager by pushing the ] key in the Julia console. Then, run the following code.

add CSV
add DataDeps
add DataFrames
add Dates
add DelimitedFiles
add Distributions
add Mimi
add Missings
add NetCDF
add Query
add RData
add Statistics
add StatsBase

You will need to add the MimiCIAM model package itself as well. To do this, run the following line while still in the Julia package manager.

add https://github.com/raddleverse/MimiCIAM.jl.git

Then, press the Backspace key to return to Julia.

Grab the large data files

Sea-level projections

If you obtained this repository via Zenodo, then the large data file containing BRICK sea-level projections from Vega-Westhoff et al. (2019) is in this repository at ciam-code/data/lslr/BRICK_projections.RData.

If you obtained this repository via GitHub, then this large data file was not present. You can download it by running the following code from the Julia console, in the top-level directory of this project.

if !isfile("./ciam-code/data/lslr/BRICK_projections.RData")
    url = "https://zenodo.org/record/3628215/files/sample_projections.RData"
    download(url,  "./ciam-code/data/lslr/BRICK_projections.RData")
end

Original CIAM GAMS outputs for benchmarking

You will need a directory full of results from the original implementation of CIAM in GAMS from Diaz (2016) in order to complete the baseline comparisons (Figure 4 in the manuscript). If you obtained this repository via Zenodo, then those files should be in ciam-code/output/originalCIAM_gams_outputs.

If you obtained this repository via GitHub, then you can obtain a zipped folder full of the Original CIAM (GAMS version) results files from the Zenodo repository accompanying this work.

• This link provides only the Original CIAM GAMS results (for baseline comparisons)
• This link provides the entire software repository snapshot, including the Original CIAM GAMS results

Run the baseline MimiCIAM simulations

This set of simulations corresponds to Figures 4 and 5 in the manuscript.

• Figure 4 is based on simulations that check the MimiCIAM implementation against the Original CIAM results from the GAMS model version. This includes a version of MimiCIAM that should match the Original CIAM results (correcting for perfect decision-maker foresight for sea-level rise and the construction height decrease fix) and the new "baseline" version of MimiCIAM that uses the Original CIAM forcing data.
• Figure 5 is based on simulations that implement updates to the GDP and population forcing (SSP scenarios), the sea-level rise forcing, and both. Each of these three cases is compared to the baseline MimiCIAM simulation from Figure 4.

To run these baseline comparison simulations, change directories into the ciam-code/src directory of the CIAM_uncertainty_propagation project. If you were already in the top-level CIAM_uncertainty_propagation directory, then you can do this in the Julia console using:

cd("ciam-code/src")

Then, before running the baseline_comparisons.jl script, you will need to change Line 43 where the brickfile is set to point to wherever you placed the BRICK sea-level projections locally on your computer. Once you have made that change, run the script either interactively or by running in the Julia console:

include("baseline_comparisons.jl")

By default, output files will be saved in ciam-code/output/baseline_comparisons. Analyzing this output and generating figures/output data files is described below under Analysis.

Run the Monte Carlo ensembles

To run the Monte Carlo ensembles under SSP5-RCP8.5 (main text) and SSP1-RCP2.6 (Supporting Material), change directories into the ciam-code/src directory of the CIAM_uncertainty_propagation project. Then, run the montecarlo_driver.jl script. It will give you an error about missing the BRICK_projections.RData file if you do not have it in the ciam-code/data/lslr directory.

The script is set up to run by default ensembles of 1000 simulations, varying sea-level rise scenarios, varying CIAM socioeconomic parameters, and varying both. This is done for each of SSP5-RCP8.5 and SSP1-RCP2.6.

The script will also run single simulations using the sea-level rise ensemble members that give the 5th, 50th, and 95th percentiles of global mean sea-level rise in 2150. CIAM parameters are held fixed at their default central values. The point here is to examine how the anticipated coastal adaptation costs differ between:

• Case 1: you compute these percentiles of the sea-level rise ensemble, and assume they are representative of the same percentiles of the distribution of adaptation costs; and
• Case 2: you compute the adaptation costs for all sea-level rise ensemble members, then compute the percentiles of the distribution of adaptation costs.

Analysis

To run the analysis for the baseline comparisons, run through the Jupyter notebook that is at work_baseline_comparisons/plotsAndAnalysis_baselineComparisons.ipynb. As the notebook mentions, some of those validation steps where the MimiCIAM results are compared against the Original CIAM GAMS results will take a long time. For example, changing the MimiCIAM segments to adjust for perfect foresight (so they'll match the old GAMS results) takes about 30 hours on a desktop workstation. (In the MimiCIAM code moving forward, only a random subset of segments are checked, so it is much faster.)

To run the analysis for the Monte Carlo uncertainty propagation experiments, run through the Jupyter notebooks that are at work_uncertainty_propagation/SSP5-RCP85.ipynb and work_uncertainty_propagation/SSP1-RCP26.ipynb. These are much faster and should not take more than a few minutes to run interactively.

Both Jupyter notebooks will yield CSV files containing the numbers used to generate figures.

---

Questions? Feedback? Tony Wong (aewsma at rit.edu) We are also interested in helping folks use other sea-level projections with MimiCIAM. So, please also feel free to ask questions and/or open issues in the MimiCIAM.jl GitHub repository.