future-climate-emulations-analysis

This repository contain the code necessary to produce the figures to the paper [CITE]. It is highly recommended to work on a multi-core machine with preferably 8 cores and at least 50Gb RAM.

Reproducing the temperature projections + figures

To run this code, first clone the repo,

git clone https://github.com/matzhaugen/future-climate-emulations-analysis.git
cd future-climate-emulations-analysis

and download the raw data from Dropbox:

curl https://www.dropbox.com/s/a1dtbml8umsovnp/raw_temperatures.RData?dl=1 -O -J -L
curl https://www.dropbox.com/s/842bdwz9fz3s76r/CCSM4_volcanic_1850-2008_prototype1.nc?dl=1 -O -J -L

Make sure that R is installed and has the necessary packages given at the top of emulationHelper.R.

Run maps.R to produce the projected temperatures. This code should really be run on an 8 core machine and is the default in the file. To modify, change the mc.cores variable in line 4 of maps.R.

Run figures.R to produce the figures. This relies on two files

1. everything_1463_1463_310_1010.Rdata: Contains all the raw model coefficients that defines the quantile maps.
2. maps_1463_1463_310_1010.Rdata: Contains the temperature projections.

Reproducing the quantile maps

This requires a CPU with at least 8 cores (preferably 16) with >6Gb/core running R/3.4.3. The code does not work with R versions >3.5 because chol was updated in the base packages and stopped working (don't ask me why).

Rscript multiMain_jack.R

The output will be saved to everything_1463_1463_310_1010.Rdata. For a SLURM submission script see multiMain_jack.sh. The job takes about 3 hours on a 16 core machine.

Deep dive into the code

The three central functions in emulationHelper.R are

1. quantile.map: estimates a quantile map based on raw temperatures given either from the LENS or the SFK15 model output.
2. map.data: Uses the model coefficients estimated in quantile.map to project any data into the future, either into one year or into a range of years. This function can also handle nested output (see point 3 below).
3. quantile.map.nested: Produces quantile maps in a nested fashion by first using the output from quantile.map and one simulation or a small set of simulations to fine tune the model coefficients with new data. The idea is to first produce a model with a lot of data from a coarse climate model, then re-calibrate using data (e.g. temperatures) from a climate model with higher resolution. Note that there are two types of models here, a climate model from which we get raw temperatures and a statistical model which we also call quantile maps.