Construction of a potato moisture diagnosis model based on hyperspectral characteristic parameters

Proper water supply is crucial for high-yielding and high-quality potato tubers. Therefore, the accurate monitoring of potato water and precision irrigation based on water scarcity information has important practical significance for potato water-saving management. Hyperspectral remote sensing has unique advantages in diagnosing crop water stress. In this paper, we measured the canopy spectral reflectance and plant water content under five irrigation treatments. The characteristic spectral parameters that responded to plant water status were selected, and a hyperspectral moisture diagnosis model of the potato leaf water content (LWC) and aboveground water content (AGWC) was established. We found that both the potato LWC and AGWC significantly decreased with increasing water stress. The potato canopy hyperspectral reflectance peak appeared in the red edge region, and this area’s reflectance varied significantly under different water treatments and decreased with decreased irrigation. Six potato moisture monitoring models with the sensitive band, first derivative, and water spectral index were established. The R 2 values of the partial least squares regression (PLSR), support vector machine (SVM), and BP neural network (BP) models between the LWC and hyperspectral data were 0.8418, 0.9020, and 0.8926, respectively. The R 2 of the PLSR, SVM, and BP models between the AGWC and the hyperspectral data reached 0.8003, 0.8167, and 0.8671, respectively. All of the six models can realize the prediction of potato plant water content, but SVM was the best model for predicting the LWC of potato. These results will be highly significant in guiding the precision irrigation of different growth stages of potatoes.