marcosrbenso/ClimateImpactML

Random Forest Algorithms to predict climate impact-drivers (CID), a.k.a., climate extreme indices for impact studies, in crop yields of soybean maize using Random Forest and XGBoost in a SHAP (SHapley Additive exPlanations) framework

Machine Learning (ML) models for explaining the impact of climate risks on crop yield

ClimateImpactML is a GitHub repository that contains instructions to reproduce results from "A data-driven framework for assessing climatic impact-drivers in the context of food security". which is under review in the journal Natural Hazards and Earth Systems Science (NHESS)

Code references

01_crop_data.R

This script was written to process annual crop yield data from different sources to remove trends and heteroscedasticity from data. This procedure aims to reduce the impacts of technological changes over the years on yields.

The following packages are required to run this code.

• blockCV
• geobr
• sp
• tidyverse

02.rf_model.R

This script was written to select the most relevant CIDs for climate impacts studies on crop yields. The same approach can be applied to other sectors.

The following packages are required to run this code.

• CAST
• caret
• doParallel
• varImp
• data.table

03_shap_model.R

The objective of this script is to run RF model for explaining the impact of climate extremes on crop yields The following packages are required to run this code.

• tidyverse
• ggpubr
• ranger
• shapviz
• treeshap
• xgboost

Data references

[available soon!]

The data used in this repo can be found in