Open data for "Predicting dominant terrestrial biomes at a Global Scale: Assessments of machine learning algorithms, climate variables indexing, and extreme climate"

______________________________________________________
This page contains public-domain data required to reconstruct simulation results in the manuscript "Predicting dominant terrestrial biomes at a Global Scale: Assessments of machine learning algorithms, climate variables indexing, and extreme climate," submitted by the following author.

Author: Hisashi SATO (JAMSTEC) 
email : hsatoscb_(at)_gmail.com

______________________________________________________
1. Folder "Code"
Detailed descriptions are available on the code. 

1-1. MachineLearningComparison.R
Machine learning programs using random forest (RF), naive Bayes classifier (NV), and support vector machine (SVM) algorithms.

1-2. Analyse_MapSimilarity.R
Calculate coincidences of simulated potential natural vegetation (PNV) maps simulated by different models.

1-3. Visualize_VCE.R
Generating VCE (Visualize Climate Image) for training CNN models.

1-4. Visualize_Maps.R
Visualizing global PNV maps.

1-5. Visualize_ClimateHistgrams.R
Visualizing histograms of climate datasets.

______________________________________________________
2. Folder "Input"

2-1. Unified_BIOCLIM_WorldClim.csv
Input data for the current climate.
This file contains the following variables.
   lon           Longitude at the center of the grid
   lat           Latitude  at the center of the grid
   bio1~19       Average climate indices from BIOCLIM (AveI)
   CDD~WSDI      Extreme climate indices (CEI)
   c1~c16        Fraction of PNV from MODIS data
   tavg01~tavg12 Monthly mean air temperature from January to December (Ave)
   prec01~prec12 Monthly precipitation from January to December (Ave)

2-2. Unified_BIOCLIM_WorldClimFutureRCP85.csv
Input data for future climate (@RCP8.5)
Including variables are the same as Unified_BIOCLIM_WorldClim.csv

2-3. BIOCLIM_RefNo.csv
This CSV file contains the following information for each grid.
   lat:    Latitude  at the center of the grid
   lon:    Longitude at the center of the grid
   latNo:  Latitude  number corresponding to the image file name
   lonNo:  Longitude number corresponding to the image file name
   lineNo: No use. Don't mind.
   vegNo:  Most dominant PNV based on the Unified_BIOCLIM_WorldClim.csv

______________________________________________________
3. Folder "Output"

3-1. PNV_sim
3-2. PNV_sim_RCP85.csv
Current and future PNV maps from various models. These files are the main output files from the code MachineLearningComparison.R. For PNV maps from CNN models (m4p1~6) were supplemented. Detailed methods to build CNN models, please refer to the following manuscript.
Sato, H. & T. Ise (2022). "Predicting global terrestrial biomes with the LeNet convolutional neural network." Geoscientific Model Development 15(7): 3121-3132.

Labels indicate combinations of machine-learning-algorithm and dataset for training the model. For example, In case of "m1p1", that column shows the simulation result of models trained with randomForest (RF) algorithm and Ave dataset.
m1: randomForest (RF)
m2: Support vector machine (SVM)
m3: Naive Bayes (NB)
m4: Convolutional Neural Network (CNN), which is NOT analysed in this code
p1: Ave
p2: Ave + CEI
p3: Ave + CEIpart
p4: AveI 
p5: AveI + CEI
p6: AveI + CEIpart