|Ph.D Student||Orlova Yevgeniya|
|Subject||Vis-NIR Spectroscopy for Non-Destructive in-Situ Monitoring|
of Apple Fruitlet Physiological State
|Department||Department of Civil and Environmental Engineering||Supervisors||PROF. Raphael Linker|
|DR. Boris Spektor|
|Full Thesis text|
Apple trees tend to bear an excess of young fruits (fruitlets). Chemical thinning with an application of bio-regulators early in the season is currently the most common solution to adjust fruit load. However, most thinners are effective only in the first few weeks following bloom and the results of this approach are highly inconsistent between seasons. Forecasting the expected fruitlet drop in response to an initial thinner application would be useful to support grower decisions regarding follow-up applications. The purpose of this research was to forecast fruitlet drop using in-situ spectroscopy in the visible and near-infrared (Vis-NIR) range.
Preliminary experiments were performed on control and treated ‘Golden Delicious’ trees at Matitiahu research station orchards in 2015 and 2016. These experiments revealed issues with sunlight interference and set-up configuration. These issues were addressed by designing a new measurement probe and developing an appropriate measurement protocol together with an algorithm that removes most of the sunlight contribution in the measurements.
The main measurement campaigns were conducted in 2017 and 2018, focusing on mature ‘Golden Delicious’ trees treated with synthetic auxins according to standard commercial practice. Fruitlets were tagged and monitored in situ every 2-4 days by measuring reflectance over the 400-1000 nm range. Measurements at 4-12 days after treatment (DAT) were used to forecast fruitlet drop taking as reference the fruitlets status on 20-26 DAT. Principal component analysis was carried out, followed by a classification algorithm - linear or quadratic discriminant analysis. Performing measurements on 4 DAT proved too early to predict fruitlet drop with adequate reliability, resulting in forecast accuracy of 64%. On the other hand, measurements at 6-12 DAT resulted in satisfactory accuracy of 80-97%, depending on the selected dates.
In addition, the most informative bands in the Vis-NIR range for forecasting fruitlet destiny were identified. Simple wavebands, band differences, and band ratios were investigated as classifying features, and the optimal features and appropriate threshold values were determined via the receiver operating characteristic curve. The best classification accuracies, ranging from 66% to 87%, were obtained with the band difference of 973 nm and 404 nm for the earliest monitoring dates on 4-6 DAT. For the later monitoring dates, 7-12 DAT, the best accuracies, ranging from 76% to 95%, were obtained with the ratio of the bands at 693 nm and 674 nm. Overall, the results for all dates except 4 DAT can be considered as satisfactory.
Complimentary measurements of intact vs. denuded fruitlets showed that the differences observed in spectral reflectance of dropping and retaining fruitlets can be explained by a substantial change in trichome density during fruitlet development, which in turn affected apparent chlorophyll absorption.
To conclude, the work presented here demonstrates that in-situ spectroscopy is suitable for forecasting apple fruitlet drop rate. It could lead to the development of a portable device relying on measurements at a few wavelengths for predicting fruitlet drop. Such forecasts would help growers manage chemical thinning, which in turn would lead to optimizing fruit load and size at harvest.