Forecasting of Fusarium head blight spatial distribution in winter wheat using machine learning

Abstract

Predicting spatial distribution of Fusarium Head Blight (FHB) is essential for precision preventive site-specific fungicide application in winter wheat cultivation. The current study presents a novel approach for the prediction of the within-field spatial distribution of FHB of winter wheat using random forest (RF), least squares support vector machines (LS-SVM), and multilayer perceptron (MLP), using high resolution data on soil characteristics, meteorological data, and remote sensing derived crop growth indices. The predictive performance of the models was assessed using two distinct training approaches based on data collected from three fields in Lithuania; a cross-field validation approach (Approach 1) and a field-specific model approach (Approach 2), using data attained from three experimental fields in Lithuania. In approach 1, MLPs achieved the highest performance with coefficient of determination (R²) values reaching up to 0.71 and residual prediction deviation (RPD) values reaching 1.87. In Approach 2, MLPs have demonstrated high performances with R² values reaching 1.00 and RPD values up to 25.63. Field-specific models significantly outperformed cross-field models, achieving Kappa coefficient values ranging from 0.91 to 0.97 across all investigated fields. The above findings indicate the potential of the effective combination of machine learning models with remote sensing and soil data for the accurate FHB prediction and for the adoption of more sustainable and targeted crop protection practices.

Plant phenotyping relevance

冬小麦のFHB（病害状態）の圃場内空間分布を、リモートセンシングデータと機械学習で推定し、複数の検証設計で性能評価しているため、病害フェノタイピング手法の適用・検証が中心です。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.