Multiple Methods for Predicting Strawberry Powdery Mildew Severity from Field Canopy Reflectance Data

Sensor-based techniques have demonstrated potential as alternatives to visual rating techniques of plant diseases in numerous horticultural crops. Our previous study showed that canopy reflectance data could significantly improve the genomic prediction of powdery mildew resistance in a strawberry breeding program. In this study, we evaluated multiple methods for canopy reflectance as a phenotyping approach that can be applied across many research contexts. We tested stepwise multiple linear regression (SMLR), partial least squares regression (PLSR), and hyperspectral best linear unbiased prediction (HBLUP) using canopy reflectance to predict strawberry powdery mildew severity. Visual rating and canopy reflectance measurements were conducted using seedlings from two different crosses from the University of Florida strawberry breeding program evaluated in 2018 to 2019 (T1) and 2019 to 2020 (T2) field trials. SMLR analysis showed that as few as five wavebands were highly correlated with disease severity, with coefficients of determination of 0.94 and 0.71 and root mean square error (RMSE) values of 0.20 and 0.44 in the T1 and T2 trials, respectively. Significant wavebands were found in the UVA region. A PLSR model using 10 variables also showed high predictive abilities of 0.94 and 0.92, respectively, with an RMSE of 0.32 within the T1 and T2 trials, whereas HBLUP showed slightly lower accuracy, with respective accuracy levels of 0.84 and 0.82 and RMSEs of 0.49 and 0.51. In PLSR, the accuracy substantially decreased by 25 to 35%, whereas in HBLUP, it decreased by 12 to 29% after validation across datasets, and moderate predictive ability was achieved. Overall, the canopy reflectance-based foliar disease prediction methods presented in this study demonstrate potential for field-based high-throughput phenotyping of strawberry powdery mildew. [Formula: see text] Copyright © 2025 The Author(s). This is an open access article distributed under the CC BY-NC-ND 4.0 International license .