AbstractFruit splitting (or fruit cracking) refers to the physical failure of the rind, which appears in various fruits and vegetables. Fruit splitting in citrus first appears as a microcrack, usually on the styler-end of the flavedo layer, spreading towards the albedo (the internal, white-colored layer) and the center of the fruit. Splitting usually begins four months after fruit set, before reaching its peak in terms of splitting incidence between September - November. At the end of the splitting period, split fruits usually abscise and drop. Some citrus varieties are more sensitive to splitting (Nova, Valencia) than others. Some years, termed splitting years, see an increase in the scope of the phenomenon, during which less-sensitive varieties (such as 'Ori') display higher splitting incidence. In recent years, fruit splitting appears to be more common. It is likely to be caused by climate change, heatwaves during sensitive periods and extreme climatic events. The level and frequency of irrigation has a significant effect on the final fruit size and its diurnal fluctuations in diameter. Hence, a significant water flow to the pulp during sensitive periods of fruit development may apply excess force on the developing rind and cause the fruit to split. Research suggests that deficit irrigation during the splitting period can reduce splitting. On the other hand, deficit irrigation can reduce fruit size and even affect yield. Fruit splitting is also affected by fluctuations in soil water content. Hence, it was recently suggested that more frequent irrigation could reduce splitting incidence. The research aim is to assess the effects of different irrigation regimes on splitting incidence in 'Ori' and 'Nova' mandarins, while minimizing the reduction in fruit size and yield.
The research is based on four irrigation experiments in two commercial plots in the 'Mehadrin' orchards, one near Kibbutz Na'an and the other near Kfar-Chabad. The experiments in Na'an have started before the beginning of this research. The first experiment was held in 2018 in a 'Nova' plot near Na'an, it included two levels of regulated deficit irrigation (RDI). The second experiment was held in 2019, it comprised of an early RDI, a late RDI and an integrated RDI (late and early RDI). The third experiment was conducted in an 'Ori' plot near Na'an in 2019, it included a late RDI treatment. The fourth experiment was held in an 'Ori' plot near kfar-Chabad in 2020. The treatments were high irrigation frequency, low frequency with RDI and high frequency with RDI. Each experiment included a control treatment, replicating common local irrigation practices.
The research included trunk and fruit dendrometers to estimate tree water status by measuring MDS and FDV, and relating them to splitting incidence. Also, a machine learning (ML) prediction model was developed to assess the spatial distribution of tree water status throughout the plot during the 'Ori' 2020 experiment and the subsequent season. The model was developed using drone-based remote sensing and validated using leaf water potential (LWP) measurements by pressure chamber. The data for the prediction model was based on four flight missions between September 2020 - June 2021. Based on the data, 10 multispectral and thermal vegetation indexes were calculated to predict tree water status (CWP). CWP was calculated as the product of LWP and PAI, based on a photogrammetric point cloud. CWSI was the only thermal-based indicator for water status prediction, the other 9 vegetation indexes were calculated from multispectral data. ML models are widely used in remote sensing, Random Forest (RF) is a simple and reliable ML model, hence it was used in this research.
The 'Nova' 2018 treatments did not show differences in splitting incidence and yield, compared to the control. The control treatment in the 'Nova' 2019 experiment showed the lowest splitting incidence (35.2%), and correspondingly, the high yield with 86.6 kg/tree. Conversely, the late treatment showed the highest splitting incidence (51.1%) and the lowest yield with 34.1 kg/tree, among all four treatments. It is assumed that the single block assigned to the control and late treatments did not sufficiently represent the variability of the trees in the plot. The RDI in the 'Ori' 2019 experiment lowered the control's already low splitting incidence (6.8%) to 5.6%, on account of a reduction in yield (121.7 kg/tree in the control, 103 kg/tree under an RDI). The irrigation levels in the 'Ori' 2020 experiment were not similar across treatments, regardless of the initial planning, due to technical difficulties in the irrigation system. Still, results showed a clear trend from the onset of the splitting period, according to which differences in fruit splitting based on irrigation levels were more significant compared to differences based on irrigation frequencies. Irrigation levels of 48% and 61% resulted in a similar splitting incidence of 47%. While lower irrigation levels of 13% and 21% resulted in lower splitting incidence of 32.8% and 38.8%, correspondingly. The treatment with the lowest splitting incidence displayed the highest yield and number of fruits. Yet, the high number of fruits was complimented by a lower average fruit weight. Most trees where MDS was correlated to splitting, showed an inverse relationship where higher MDS was correlated with lower splitting incidence (R²>0.4). Also, most trees where FDV was correlated to splitting, showed a positive relationship where higher FDV was correlated with higher splitting incidence (R²>0.4). The relation between trunk shrinkage, diurnal fluctuation of fruit diameter and splitting, brought forward the assumption that RDI (and high MDS), reduces daily fluctuation in fruit diameter (FDV), thus reducing the likelihood of a fruit to split.
The method that was chosen for the prediction model is a canopy-based approach, this approach requires moving from the single leaf (LWP) to the canopy level (CWP). Scaling-up from a simple LWP linear regression model to a RF-based ML prediction model has several advantages: (1) it accounts for the spatial variability of the canopy; (2) simplifies the process of predicting canopy level water potential (CWP); (3) reflects the tree water status across a large area; (4) mitigates variations in canopy size and structure by incorporating the LAI element. Still, future research should require the calibration of LAI using ground measurements. Results of RF model showed a correlation of R²=0.9; RMSE=0.67. The element with the highest contribution to model performance was the OSAVI. Conversely, CWSI displayed the lowest contribution: the spatial variability of thermal values within the canopy was very high across all flight missions. Another explanation relates to the isohydric behavior of citrus trees. Finally, the relatively low resolution of thermal imagery may have incorporated soil reflectance from pixels classified as vegetation. The wide sample size from the predicted CWP values allowed us to identify differences in tree water status between treatments, during the September and January missions. During the month of June, an after a prolonged heatwave, RDI trees displayed lower CWP values compared to the non-RDI trees.
Drone-based remote sensing captures a single moment in time and thus cannot serve as a continuous indicator of splitting-related factors. Its superiority lies in its capacity to detect spatial and seasonal trends in tree water status over large areas. On the other hand, trunk and fruit dendrometers can continually monitor the water status of the tree and the fruit. Since the number of installed dendrometers is limited, they are unable to account for the water status of every tree across the plot. The marriage of both methods, remote sensing and fruit and trunk dendrometers, relies heavily on understanding the mechanism behind fruit splitting in the vascular system. Once this relationship is established, a spatially explicit model of fruit splitting could potentially be formulated.
|Date of Award
|1 Dec 2021
|Avi Sadka (Supervisor) & Tarin Paz-Kagan (Supervisor)