Hyperspectral assessment of wheat lodging: From field to EnMAP satellite observations

Résumé fourni par la source

Crop lodging, the permanent displacement of crop stems from their vertical position, causes substantial yield and quality losses in wheat production. Early and accurate detection of lodging and its severity is therefore essential for improving harvest management and reducing economic risk. This study, for the first time, examines hyperspectral data from the Environmental Mapping and Analysis Program (EnMAP) satellite in conjunction with field hyperspectral measurements and machine learning algorithms to detect wheat lodging and its severity and to identify spectral regions important for lodging detection. The study was conducted at Bonifiche Ferraresi Farm in Italy, where wheat biophysical measurements were collected alongside spectral measurements acquired using an Analytical Spectral Device (ASD) spectroradiometer, concurrent with EnMAP data acquisition. Wheat spectral reflectance derived from both field and EnMAP data was analyzed to determine how lodging alters wheat spectral characteristics and to identify sensitive wavelengths. Following spectral preprocessing, lodging severity was quantified using a lodging score and modeled with Principal Component Analysis (PCA) based Gaussian Process Regression (GPR), Partial Least Squares Regression (PLSR), Multilayer Perceptron (MLP), and Explainable Boosting Machine (EBM). Model performances and spectral relevance were evaluated through PCA loadings, Variable Importance in Projection (VIP), SHapley Additive exPlanations (SHAP) values, and EBM feature contributions. The results indicate that EBM achieved the highest predictive performance ( R 2 = 0 . 66 and ɛ rmse = 0 . 15 for ASD and R 2 = 0 . 60 and ɛ rmse = 0 . 18 for EnMAP), while interpretability analyses consistently highlighted six key spectral regions (550, 670, 720–740, 865, 1650, and 2130–2190 nm) as being sensitive to lodging-caused structural changes in wheat canopies. These findings demonstrate the potential of hyperspectral modeling for satellite-based lodging assessment under cloud-free acquisition conditions, while highlighting current constraints on operational deployment related to revisit frequency, atmospheric effects, and transferability beyond the study site and growth stage.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.