Environmental and demographic determinants of Aedes albopictus seasonal activity in southern France: a modeling study (Taconet, Radici et al., 2026)

WHAT THIS REPOSITORY CONTAINS

The R scripts here enable simulations that generate the results of the publication "Environmental and demographic determinants of Aedes albopictus seasonal activity in southern France: a modeling study" (https://doi.org/10.1101/2025.10.10.681559), Taconet, Radici et al. (2026).

In this article, we couple Machine Learning model and a mechanistic model to infer dependencies between Aedes albopictus (the Asian tiger mosquito) presence and abundance and environmental, meteorological, and demographic drivers across 4 sites in southern France.

THE SCRIPTS IN DETAIL

The repository is organized into two independent workflows:

- Mechanistic module (scripts beginning with "1")
- Machine-learning (ML) module (scripts beginning with "2")

A common module is represented by scripts "00_meteo_data_extract.R", which downloads weather data from the MétéoFrance repository and has to be run first, and "30_models_evaluation.R", which compares the performances of the mechanistic and the ML modules, which has to be run last.

Within each module, scripts should be executed sequentially.

Mechanistic module:
11_mechanistic_meteo_data_preparation.R: preprocess weather data and save them into chunks (1 RData file = 1 year, 1 site)
12_mechanistic_analysis_weather_season.R: contains the statistical analysis of weather for 2023 and 2024.
13_mechanistic_model_run.R: takes the MeteoFrance preprocessed chunks and simulates the mosquito population/oviposition trajectories.
14_mechanistic_model_results_plots.R: plots comparison between simulated and observed oviposition. Save results in "table_mechanistic.RData". 
15_mechanistic_Wilkinson_test.R: performs the Wilkinson tests for all interannual variables (oviposition; temperature and precipitation; mosquito life traits).

Statistical module:

21_ml_meteo_data_preparation.R: Imports, cleans, and harmonizes the raw meteorological observations from Meteo-France. Outputs a single daily weather dataset used throughout the machine-learning analyses.
22_ml_variables_extraction.R: Links weather observations to ovitrap collections and generates all candidate meteorological predictors, including cumulative lag-window variables, used for model development.
23_ml_compute_ccm.R: Computes correlation maps (CCMs) to identify the most informative meteorological variables and lag windows for predicting mosquito presence and abundance.
24_ml_model_fitting.R: Fits the Random Forest models for oviposition presence and abundance, performs leave-one-site-out cross-validation, and saves the fitted models and predictions.
25_ml_interpretation.R: Interprets the fitted machine-learning models using permutation variable importance and partial dependence plots to quantify the effects of the selected predictors.

Shared modules: 

00_meteo_data_extract.R: Downloads MeteoFrance data and saves them locally. To be run first.
30_models_evaluation.R: Compares the machine-learning and mechanistic models by reconstructing expected abundance predictions, generating comparison figures, and computing the evaluation metrics reported in the manuscript. To be run last.

THE DATA IN DETAIL (to be unzipped)

MeteoFrance/
Contains the daily meteorological observations downloaded from Meteo-France for the study period.
These files can either be generated automatically by running 1_A_meteo_data_extract.R, or used directly if already present.

Ovitrap/
Contains both raw ovitrap observations for each site in 2023-2024 and averaged across weeks, as data.frames.