Optical sensor data can be used to determine changes in anthocyanins, chlorophyll and soluble solids content (SSC) in apple production. In this study, visible and near-infrared spectra (729 to 975 nm) were transformed to SSC values by advanced multivariate calibration models i.e., partial least square regression (PLSR) in order to test the substitution of destructive chemical analyses through non-destructive optical measurements. Spectral field scans were carried out from 2016 to 2018 on marked ‘Braeburn’ apples in Southwest Germany. The study combines an in-depth statistical analyses of longitudinal SSC values with horticultural knowledge to set guidelines for further applied use of SSC predictions in the orchard to gain insights into apple carbohydrate physiology. The PLSR models were investigated with respect to sample size, seasonal variation, laboratory errors and the explanatory power of PLSR models when applied to independent samples. As a result of Monte Carlo simulations, PLSR modelled SSC only depended to a minor extent on the absolute number and accuracy of the wet chemistry laboratory calibration measurements. The comparison between non-destructive SSC determinations in the orchard with standard destructive lab testing at harvest on an independent sample showed mean differences of 0.5% SSC over all study years. SSC modelling with longitudinal linear mixed-effect models linked high crop loads to lower SSC values at harvest and higher SSC values for fruit from the top part of a tree.

The characterization of light as photosynthetically active radiation (PAR) is crucial for modeling and monitoring plant metabolism and development. Existing intercepted PAR sensing solutions are limited by the requirement to average many below-canopy measurements to account for heterogeneity in the plant canopy, which requires a human operator. In this work, we address this issue by introducing a novel sensing concept to measure the light coming from different areas of the tree canopy with a single sensor by dynamically varying the sensor’s field of view in the zenith and azimuth directions. The sensing system consists of (1) an imaging lens that collects the incoming light, (2) a programmable Liquid-Crystal Display (LCD), which acts as an optical shutter, (3) a diffuser, and (4) the AS7341 multispectral sensor. In this work, the PAR model required to calibrate the sensor is derived and the sensing system is calibrated and characterized. A sensor prototype is then assembled, calibrated, and characterized, achieving competitive performance with a mean average percentage error of 6.78% and a normalized root mean square error of 4.82% on PAR measurements. Tests using a custom LED test sphere show that the sensor has a spatial resolution of 7° and 15° in the θ and ϕ directions, respectively. With a power consumption of 19.54mW mainly attributable to the LCD, our new sensing system could be used for more than two years at one hour of measurements per day, which paves the way towards long-term continuous monitoring of PAR.