at Google ScholarTitle data
Fraga, Helder ; Guimarães, Nathalie ; Candiago, Sebastian ; Yang, Chenyao ; Sosa, Eduardo ; Serrano-Notivoli, Roberto ; Dalezios, Nicolas R. ; de Souza Rolim, Glauco ; Santos, João A.:
Agroclimatic drivers and suitability niches of major European crops revealed by explainable machine learning.
In: Ecological Informatics.
Vol. 99
(September 2026)
.
- 104065.
ISSN 1574-9541
DOI: https://doi.org/10.1016/j.ecoinf.2026.104065
Abstract in another language
Anticipating how European agriculture will shift under future climates demands a clear understanding of which climatic conditions govern crops today. Here, we present an explainable machine learning (ML) framework that moves beyond conventional methods to reveal the structure of crop-climate relationships across the continent. By combining ML ensemble classifiers (Random Forest, XGBoost, CatBoost and LightGBM in a VotingClassifier), with SHAP (SHapley Additive exPlanations) dependence analysis and k-Shapley interaction networks, we characterise the agroclimatic niches for 40 crops derived from ∼13 million agricultural grid cells, using bioclimatic indices from state-of-the-art CERRA climate reanalysis data. The framework allowed to accurately assess the agroclimatic niches of all 40 crops, with moderate-to-strong cross-validation performance, ranging from Matthews Correlation Coefficient (MCC) 0.5 (apple) to 0.9 (olive). More importantly, the methodology allowed to identify the most important bioclimatic factors for crop distribution. The Latitude-Temperature Index was found to be the most influential driver for all 40 crops, followed by isothermality, temperature annual range, temperature seasonality, and minimum temperature of the coldest month, demonstrating that the north-south thermal gradient and continentality-oceanicity axis are the primary structuring dimensions of European agriculture. SHAP dependence curves expose crop-specific non-linearities, including long suitability ranges, unimodal optima, and bimodal response patterns. Crucially, interaction networks reveal that cool-season crops exhibit predominantly synergistic driver interactions, while warm-season crops show antagonistic ones, a structural distinction invisible to standard importance methods. The result is a European-wide atlas of crop agroclimatic niches and suitability ranges, offering a quantitative and interpretable foundation for adaptation planning and climate impact assessment.
Further data
| Item Type: | Article in a journal |
|---|---|
| Refereed: | Yes |
| Keywords: | Ensemble classifier; Machine learning interpretability; AI in agriculture; Agroclimatic indices; Europe |
| Institutions of the University: | Faculties > Faculty of Biology, Chemistry and Earth Sciences > Department of Earth Sciences > Professor Ecological Services > Professor Ecological Services - Univ.-Prof. Dr. Thomas Koellner Research Institutions > Central research institutes > Bayreuth Center of Ecology and Environmental Research- BayCEER Graduate Schools > Bayreuth Graduate School of Mathematical and Natural Sciences (BayNAT) > PEER Ökologie und Umweltwissenschaften |
| Result of work at the UBT: | Yes |
| DDC Subjects: | 500 Science > 550 Earth sciences, geology |
| Date Deposited: | 30 Sep 2026 06:55 |
| Last Modified: | 30 Sep 2026 06:55 |
| URI: | https://eref.uni-bayreuth.de/id/eprint/99508 |