Towards precise grape management based on multispectral and hyperspectral remote sensing : a thesis presented in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Agriculture and Horticulture at Massey University, Manawatū, New Zealand

Abstract

Non-destructive and rapidly measurements of grape quality allow vine growers to make effective decisions for selective harvesting and vineyard management, which results in more economic and sustainable wine production. The objective of this dissertation was to investigate the potential of using non-destructive techniques at pre- and post-harvest stage to predict ‘Pinot Noir’ grape quality parameters (total soluble solids, TSS; and titratable acidity, TA). Three nondestructive techniques unmanned aerial vehicle (UAV) based on multispectral camera, visiblenear infrared-shortwave infrared (VNIR-SWIR) spectroscopy, and hyperspectral imaging (HSI) were applied on commercial vineyards. In the multispectral study, the spectral reflectance measurements of vine canopy were acquired using an UAV during flowering stage. The acquired spectral data were combined with environmental ancillary data to predict grape TSS. UAV combined with ancillary data showed a moderate prediction performance, indicating grape growers can manage large vineyard areas efficiently, reducing labor and operational costs based on this technique. In the hyperspectral study, the spectral reflectance measurements of grape berries were acquired using VNIR-SWIR spectroscopy and HSI during pre- and postharvest stages. VNIR-SWIR spectroscopy was used directly in the vineyard condition with a contact probe to ensure stable light conditions during pre-harvest stage. The acquired spectra data were pre-processed by various methods including principal component analysis and spectra pre-processing. The ability of using different spectral regions to predict grape TSS was validated with an external dataset. The spectroscopy under VNIR spectrum is sufficiency for accurate prediction, suggesting cost-effectiveness over VNIR-SWIR. HSI was used under laboratory conditions during post-harvest stage to predict and classify grape quality parameters (TSS and TA). The high level of classification accuracy achieved indicating HSI was adequate for grape TSS sorting purposes. However, classification of TA remained unsuccessful for industry application. Further study should consider to using different sensors to determine grape TA. This study demonstrated the ability of using different non-destructive techniques to predict grape quality parameters in ‘Pinot Noir’ grape variety, which would allow more accurate and real-time assessment of grape quality, aiding in the optimal timing of harvest to enhance wine quality.