Arctic vegetation prediction and refugia analysis

Predicting how Arctic vegetation will look in 2050 based on CAVM and CMIP6 climate data based on Pearson et. al. (2013) methodology. This is code for the paper

M. R. Chacko, J. Oehri, E. Plekhanova, G. Schaepman-Strub Will current protected areas harbor refugia for threatened Arctic vegetation types until 2050? A first assessment (2023) Arctic, Antarctic, and Alpine Research https://doi.org/10.1080/15230430.2023.2203478

Data

Before running the code, please download the

• CAVM rasterfile https://data.mendeley.com/datasets/c4xj5rv6kv/1

• Substrate CAVM shapefile https://arcticatlas.geobotany.org/catalog/dataset/circumpolar-arctic-substrate-chemistry

• 19 bioclimatic variables for the 3 climate models (EC-Earth3-Veg, IPSL-CM6A-LR, MRI-ESM2-0) and SSPs 126 and 585 scenarios from 2041 to 2060 on 30sec resolution https://www.worldclim.org/data/cmip6/cmip6_clim30s.html#2041-2060

• GLC2000 land cover dataset https://forobs.jrc.ec.europa.eu/products/glc2000/glc2000.php

• Final predictions and data for the data analysis part: https://doi.org/10.5281/zenodo.6902764

• For detailed methods describing the analysis part: https://www.biorxiv.org/content/10.1101/2021.04.28.441764v1 [post-print link to follow]

Code

The code has the following structure

1. Preparing predictors

   1.0 Cropping bioclimate variables to the Arctic extent - above 60 latitude

   1.1a Masking bioclimate variables with CAVM

   1.1b Masking bioclimate variables with CAVM leaving 100 km buffer with trees

   1.2 Stacking 19 bioclimate variables with CAVM and converting the raster to the dataframe

   1.3 Process substrate - converting to 1 and 0 for acidic and non-acidic soils and expanding to 100 km buffer, assigning random acidity there

   1.4 Preparing the vegetation file with trees - adding GLC2000 tree classes to 100km buffer of CAVM

   1.5 Combining substrate, bioclimatic variables and vegetation data in the final dataset for RF classification and prediction

2. Classification using Random Forest

   2a Training RF on substrate and bioclimatic variables to classify CAVM vegetation

   2b The same on the larger area with 100km buffer with trees

3. Prediction using Random Forest

   3a making prediction for 6 datasets (3 climate models, 2 emission scenarios)

   3b prediction for CAVM area only - which we use as an input in restricted scenarios

4. Restricted tree dispersal scenarios

   4 combining predictions made in 3a and 3b for restricted tree scenarios (5 km and 20 km disperal).

5. Analysis and figures

   5.0 using the results of the zonal histogram (see paper methods) to calculate a) abundance of vegetation types in the present circumpolar Arctic, within and outside of protected areas b) abundance of vegetation types in the future predictions of the circumpolar Arctic, within and outside of protected areas c) classify the vulnerability of Arctic vegetation types using the Red List of Ecosystems Criteria