Ground-based RGB imaging to determine the leaf water potential of potato plants

The determination of plant water status from leaf water potential (Psi L) data obtained by conventional methods is impractical for meeting real time irrigation monitoring requirements. This research, undertaken first, in a greenhouse and then in the field, examined the use of artificial neural network (ANN) modeling of RGB (red green blue) images, captured by a ground-based, five mega pixel digital camera, to predict the leaf water potential of potato (Solanum tuberosum L). The greenhouse study examined cv. Russet Burbank, while the field study examined cv. Sangre. The protocol was similar in both studies: (1) images were acquired over different soil nitrate (N) and volumetric water content levels, (2) images were radiometrically calibrated, (3) green foliage was classified and extracted from the images, and (4) image transformations, and vegetation indices were calculated and transformed using principal components analysis (PCA). The findings from both studies were similar: (1) the R and G bands were more important than the B image band in the classification of green leaf pigment, (2) soil N showed an inverse linear relationship against leaf reflectance in the G image band, (3) the ANN model input neuron weights with more separation between soil N and PsiL were more important than other input neurons in predicting PsiL, and (4) the measured and predicted PsiL validation datasets were normally distributed with equal variances and means that were not significantly different. Based on these research findings, the ground-based digital camera proved to be an adequate sensor for image acquisition and a practical tool for acquiring data for predicting the PsiL of potato plants.;Keywords: nitrogen, IHS transformation, chromaticity transformation, principal components, vegetation indices, remote sensing, artificial neural network, digital camera.