Esca Epidemiology — Supplementary Material

Supplementary analysis code for the Esca epidemiology paper. Reproduces the machine-learning analyses, FORD variable selection, and manuscript report from the data in this repository, using a data set hosted on Zenodo.

---

Quick Start

Note that the original data set is hosted on Zenodo:

https://zenodo.org/records/21704548

and that a description of the variables is available both on Zenodo and in the file description dataset.md

Run the three steps below from the repository root in an R session:

# 1. Run all ML configurations (8 scenarios × 3 models)
source("run_all_configurations.R")

# 2. Run FORD variable selection
source("ford_analysis.R")

# 3. Knit the manuscript report
rmarkdown::render("Esca_Epidemiology.Rmd")

This reproduces the main performance tables, variable-importance rankings, and FORD sections of the manuscript. Map rendering requires external GIS assets (see External Dependencies); without them, the report knits with a fallback message.

---

Repository Structure

Once the different scripts have been run, the structure of the folder is the following:

├── data/                       # Tracked input data (disease, weather, soil, climate)
├── results/                    # ML pipeline outputs per configuration
└── plots/                      # Report-generated plots

Core Analysis Scripts

Script	Role
run_all_configurations.R	Orchestrates the ML pipeline across 8 configurations
esca_ml_config.R	Defines configuration scenarios (filters, feature sets)
esca_ml_pipeline.R	Core ML pipeline (tidymodels: Random Forest, XGBoost, LASSO)
combined_variable_importance.R	Aggregates variable importance across configurations
ford_analysis.R	FORD forward variable selection analysis

Report

File	Description
Esca_Epidemiology.Rmd	Main manuscript report (HTML output)

---

Data Requirements

Tracked Input Data

All files below are tracked in the repository — no external download is needed:

File	Description
data/RU_data.csv	Soil water reserve capacity per plot
data/plot_station_associations.csv	Plot ↔ Vitimeteo station spatial associations
data/plot_meteoswiss_associations.csv	Plot ↔ MeteoSwiss station spatial associations
data/weather_data_Agrometeo_Esca_plots_monthly.csv	Monthly Vitimeteo weather data
data/meteoswiss_weather_data_1995_2024.csv	Monthly MeteoSwiss weather data
data/VPD_data_Agrometeo_monthly.csv	Monthly VPD from Vitimeteo (optional)
data/VPD_data_MeteoSwiss_monthly.csv	Monthly VPD from MeteoSwiss (optional)
data/regional_climate_P_ET_summary.csv	Regional climate summary (optional, for report)
data/regional_climate_P_ET_yearly.csv	Regional climate yearly data (optional, for report)

Generated Outputs

The report (Esca_Epidemiology.Rmd) reads outputs produced by steps 1 and 2:

• results/combined_performance_summary.csv
• results/combined_variable_importance.csv
• results/ford_variable_selection.csv
• results/ford_variable_selection_seasonal.csv

---

Dependency Graph

run_all_configurations.R
 ├── sources: esca_ml_config.R, esca_ml_pipeline.R
 ├── reads:   Esca_CH_2001-2022.csv (on Zenodo), data/RU_data.csv,
 │            data/plot_station_associations.csv, data/plot_meteoswiss_associations.csv,
 │            data/weather_data_Agrometeo_Esca_plots_monthly.csv,
 │            data/meteoswiss_weather_data_1995_2024.csv,
 │            (optional) data/VPD_data_*.csv
 ├── writes:  results/<config>/ (per-config outputs)
 │            results/combined_performance_summary.csv
 └── triggers: combined_variable_importance.R

ford_analysis.R
 ├── sources: esca_ml_config.R, esca_ml_pipeline.R
 └── writes:  results/ford_variable_selection.csv
              results/ford_variable_selection_seasonal.csv

Esca_Epidemiology.Rmd
 ├── reads:   Esca_CH_2001-2022.csv (on Zenodo),
 │            results/combined_performance_summary.csv,
 │            results/combined_variable_importance.csv,
 │            results/ford_variable_selection*.csv
 └── writes:  plots/*.png, Esca_Epidemiology.html

---

Analysis Configurations

run_all_configurations.R executes 8 scenarios (4 variety filters × 2 feature sets):

Configuration	Variety Filter	Feature Set
Chasselas Only	chasselas	Monthly
Exclude Sauvignon	All except sauvignon blanc	Monthly
Stable Chasselas	chasselas, plots observed 2004–2022	Monthly
Main Varieties	chasselas, gamay, pinot noir, pinot gris	Monthly
Chasselas Seasonal	chasselas	Seasonal aggregates
Exclude Sauvignon Seasonal	All except sauvignon blanc	Seasonal aggregates
Stable Chasselas Seasonal	chasselas, plots observed 2004–2022	Seasonal aggregates
Main Varieties Seasonal	chasselas, gamay, pinot noir, pinot gris	Seasonal aggregates

Three models are tuned via 10-fold cross-validation per configuration: Random Forest (ranger), Gradient Boosting (XGBoost), LASSO (glmnet).

---

External Dependencies

Optional components that affect only map rendering in the report:

• map_Esca_sampling_plots.R (local script, not tracked)
• Swiss GIS assets expected under ../swiss-maps/

If absent, map chunks in Esca_Epidemiology.Rmd fall back gracefully and the report can still be knitted.

---

R Package Dependencies

• Core: tidymodels, tidyverse, readxl, lubridate, xgboost, ranger, glmnet, vip, finetune, doParallel
• Utilities: corrplot, probably, httr, data.table

---

Notes

• The seasonal main-varieties output folder is named main_varieties_seasonnal (legacy typo preserved for consistency).
• prepare_ru_data.R regenerates data/RU_data.csv from the original soil workbook when needed.