Image analysis and model development for eggplant fruit growth in a greenhouse

In greenhouse eggplant cultivation, fruit growth is a critical determinant of crop yields and depends on the surrounding environment. In this study, to evaluate fruit growth responses to the surrounding environment, we developed a mathematical model for eggplant fruit growth in a greenhouse. We first measured eggplant fruit growth using digital photographs. The growth of eggplant fruits from flowering until harvest was represented accurately by an exponential growth function rather than a commonly used single sigmoid growth function, as eggplant fruits were harvested at a market-specified size during an exponential growth period before maturation. We attempted to construct a statistical model that estimates the relative growth rate (μ) in the exponential function from environmental and plant information (i.e., daily accumulated solar radiation (I), daily average temperature (Ta), daytime average temperature (Ta,day), nighttime average temperature (Ta,night), daytime average CO2 concentration (Ca), and daily accumulated photosynthesis (P)). This model construction was based on two datasets that recorded the flowering dates, harvest dates, and harvested fruit weights of six plants throughout an entire year. A simple correlation analysis between μ and each of the explanatory variables revealed that correlation with μ was highest with Ta,day (correlation coefficient R = 0.75), while that with Ca showed a negative R value (-0.41). Further, we used principal component regression (PCR) to construct a statistical model for predicting μ while avoiding multicollinearity among the explanatory variables. Different combinations of the explanatory variables were tested to find the most generalizable model. It was confirmed that overfitting occurred to the training data depending on the combination of explanatory variables used in the model, leading to a decrease in the generalization ability of the model. The best combination of explanatory variables for predicting μ was P, Ta,day and Ta,night.