Parsimonious machine learning for the global mapping of aboveground biomass density

This repository hosts data and code presented in the article "Parsimonious machine learning for the global mapping of aboveground biomass potential". The repository contains a compressed file containing all the code needed to reproduce the methodology that we developed and to analyse its results. We did not upload all the temporary and intermediate data files that are created during the execution of the method. We rather uploaded "milestone" data, i.e. final results or important intermediate ones. This includes the final training dataset, model calibration data,  the final trained model, the global data for prediction, the final global map of potential aboveground biomass density (AGBD) at present times (raster files at 1km2 and 10km2 resolution), maps depicting regions where climatic conditions are outside of the training range of positive AGBD instances and maps depicting world regions without trees. 

Files:

code.zip : Compressed directory with all the code needed to reproduce the methodology presented in the manuscript. Contains a README file. Also contains temporary data generated in the process, the training dataset, the trained model, and model calibration data.

potential_AGBD_Mgha_1km_present_climate_1980_2010.tif : the predicted global potential AGBD under contemporary climate conditions and at a resolution of 1 squared kilometer.

potential_AGBD_Mgha_10km_present_climate_1980_2010.tif : the predicted global potential AGBD under contemporary climate conditions downsampled at a resolution of 10 squared kilometers.

potential_AGBD_Mgha_10km_model_difference.tif : the difference between our prediction of potential AGBD and the prediction from a complex state-of-the-art model from Walker et al. (2022). 

potential_AGB_Mg_1km_present_climate_1980_2010.tif : the predicted global potential pixel-level AGB under contemporary climate conditions downsampled at a resolution of 1 squared kilometers.

number_predictors_out_of_range.zip : tiled maps representing the number of climatic predictors outside of the training range before including 0 AGBD instances in the training dataset. 

tree_absence_map.zip : tiled maps representing world regions without trees. Based on Crowther et al. (2015) (https://elischolar.library.yale.edu/yale_fes_data/1/).

inference_pipeline_potential_agbd_Mgha_climate.pkl : Calibrated model for the prediction of potential AGBD given bioclimatic conditions. 

predictors_data_global.zip : Global predictors data to apply the model on.