Global Gridded Crop Model (GGCM) Feature Attribution

This repository provides the pipeline code for reproducing the comparative feature attribution study presented in the reference paper below.

Pipeline

The pipeline processes climate and yield data of GGCMs, and combines it with auxiliary data from different sources (soil, site, Köppen-Geiger, harvested area, etc.).

flowchart TD;
    subgraph Processing;
    1a[1a_process_climate.py]-->2[2_run_analysis.py];
    1b[1b_process_yields.py]-->2;
    end;
    subgraph Visualization;
    2-->2a[results.p]
    3a[3a_plot_importance.py]-->2a;
    3b[3b_cluster.py]-->2a;
    end;

Loading

A description of the proposed folder structure and auxiliary input data can be found here.

Example

[1a] Process climate data (per GGCM)

The first step converts binary¹ climate data files into aggregated features used for attribution. pdayfile and mdayfile can be passed as paths to planting and maturity day NetCDFs. When omitted, the pipeline uses a biophysical growth model to estimate the growing season length.

python pipeline/1a_process_climate.py 
--mdayfile /data/input/epic-iiasa_gswp3-w5e5_obsclim_2015soc_default_matyday-mai-firr_global_annual_1901_2016.nc 
--pdayfile /data/input/epic-iiasa_gswp3-w5e5_obsclim_2015soc_default_plantday-mai-firr_global_annual_1901_2016.nc 
--climdir /data/input/climate 
--locfile /data/input/CLIMATE_LAT_PD_HD_PHU_ELEV_PRMT74_mai_noirr_fH_v3c.csv  
--co2file /data/input/co2_historical_annual_1765_2014.txt 
--out /data/generated/climate_EPIC-IIASA_maize.h5
--year_from 1971 
--year_to 2015 
--shift 0 
--clim_year_from 1901 
--clim_year_to 2016 

[1b] Process yield (per CCGM)

This step reads end-of-season yields per year, applies LOESS detrending and calculates relative yields.

python pipeline/1b_process_yield.py 
--yieldfile /data/input/yield/ epic-iiasa_gswp3-w5e5_obsclim_2015soc_default_yield-mai-noirr_global_annual_1901_2016.nc
--mdayfile /data/input/epic-iiasa_gswp3-w5e5_obsclim_2015soc_default_matyday-mai-firr_global_annual_1901_2016.nc 
--pdayfile /data/input/epic-iiasa_gswp3-w5e5_obsclim_2015soc_default_plantday-mai-firr_global_annual_1901_2016.nc 
--out /data/generated/yield_EPIC-IIASA_maize.h5
--year_from 1971 
--year_to 2015 
--shift 0 

[2] Analysis (multiple GGCMs)

The comparative feature attribution is performed in this step.

python pipeline/2_run_analysis.py 
--datadir /data/generated 
--inputdir /data/input 
--out /data/output 
--crop maize 
--irr rf
--features default
--models EPIC-IIASA GGCM2 GGCM2 ...

Note that the analysis uses additional, auxiliary data. Filenames, if changed, can be adapted in the data loading section in 2_run_analysis.py:

data = analysis.load_data(
    file_climate=data_dir / f'climate_{model}_{crop}_{irr}.h5', 
    file_yield=data_dir / f'yield_{model}_{crop}_{irr}.h5',
    file_kg=input_dir / 'Beck_KG_V1_present_0p5.tif',
    file_spam=input_dir / f'spam2010V2r0_global_H_{crop.upper()}_R_30mn.tif',
    file_soil=input_dir / 'HWSD_soil_data_on_cropland_v2.3.nc',
    file_site=input_dir / 'CLIMATEID_SLP_GGCMI_LATLON.txt',   
)

Results for the selected models are stored in /data/output/results.p for further analysis and visualization.

[3a] Importance score plots

python pipeline/3a_plot_importance.py 
--results /data/output/results.p 
--out /data/output/importance.svg 
--crop corn 
--irr rf

[3b] Clustering and dendrograms

python pipeline/3b_cluster.py 
--results /data/output/results.p 
--out /data/output 
--crop corn 
--irr rf

Reference Paper

Oberleitner, T., Folberth, C., et al., 2025. Explaining climatic drivers of yield anomalies in global crop models through metamodel-based attribution. Authorea Preprints. DOI: 10.22541/essoar.176071967.73925210/v1

License

Licensed under the MIT license.

This project uses open-source software packages released under permissive licenses (BSD, MIT, Apache 2.0, MPL 2.0, PSF). Full license texts are available in the respective package distributions.

Footnotes

1. Folberth, C., Baklanov, A., Khabarov, N., Oberleitner, T., Balkovič, J. and Skalský, R., 2025. CROMES v1. 0: a flexible CROp Model Emulator Suite for climate impact assessment. Geoscientific Model Development, 18(17), pp.5759-5779. ↩