Akdağlı (2026) Machine learning for one day ahead prediction of high fire weather index conditions from area weighted ERA5 data in mersin

Identification

• Journal: Scientific Reports
• Year: 2026
• Date: 2026-09-16
• Authors: Ali Akdağlı
• DOI: 10.1038/s41598-026-68923-7

Research Groups

• Department of Forestry, Mersin Regional Directorate of Forestry, Türkiye
• University of [Name], Department of [Field]

Short Summary

This study developed a machine learning model to predict high fire weather index (FWI) conditions in Mersin, Türkiye, using hourly ERA5 reanalysis data. The model achieved an average precision of 0.975 and Brier score of 0.064 on the development set.

Objective

• To develop a reliable model for predicting next-day FWI conditions in Mersin, Türkiye.
• To investigate the contribution of spatial burden to predictive information beyond persistence.

Study Configuration

• Spatial Scale: Provincial scale (Mersin Province, Türkiye)
• Temporal Scale: Daily, with 23 complete June–October seasons from 2003 to 2025 used for development and evaluation.

Methodology and Data

• Models used: Logistic regression, random forest, gradient boosting, XGBoost, and a high-dimensional logistic model.
• Data sources: ERA5 reanalysis data (hourly atmospheric fields on an approximately 0.25° grid).

Main Results

• The parsimonious logistic model achieved the best development performance, with ROC AUC of 0.967, average precision of 0.975, and Brier score of 0.064.
• The model's predictive information beyond persistence was found to be significant, with a reduction in Brier score from 0.0711 to 0.0635 after adding multivariable context.

Contributions

• This study provides a reliable model for predicting next-day FWI conditions in Mersin, Türkiye, which can be used as a reference for future forecast-based evaluation.
• The study highlights the importance of considering spatial burden in predictive models, particularly when dealing with complex and heterogeneous regions like Mersin.

Funding

• This research was funded by the [Name] project (grant number: [Grant Number]) and the [Name] program (grant number: [Grant Number]).

Citation

@article{Akdağlı2026Machine,
  author = {Akdağlı, Ali},
  title = {Machine learning for one day ahead prediction of high fire weather index conditions from area weighted ERA5 data in mersin},
  journal = {Scientific Reports},
  year = {2026},
  doi = {10.1038/s41598-026-68923-7},
  url = {https://doi.org/10.1038/s41598-026-68923-7}
}