Construction of empirical models for leaf area estimation in six Passiflora species

Abstract:【Objective】Passiflora belongs to the Passifloraceae, most species of which belong to peren- nial vines. Edible passionflower is cultivated widely, and is mainly produced abroad in North America, South America, Australia, Malaysia and southern Africa and other countries and regions. Passionflower and the dominant companion species can be mutually beneficial and symbiotic, adapt to the environ- ment, growing rapidly, and have high yield. Cultivating excellent varieties of passionflower as an eco- nomic fruit tree in rocky desertification areas can promote the development of local industries. At pres- ent, there are few studies on the construction of empirical models for estimation of leaf area of Passiflo- ra plants. The size and distribution of plant leaf area (LA) directly affect the ability of leaves to inter- cept sunlight, carry out carbon utilization and gain the productivity of plant communities. LA is often used as a basic parameter to measure plant photosynthesis capacity, and is used in plant physiology, pa- thology, and crop breeding. agriculture and forestry production and management, model estimation, etc. Simple, fast and accurate detection of leaf area is of great significance for guiding the reasonable high-density planting, the adjustment of population structure, variable fertilization and precise sprin- kling irrigation to obtain high yield. This paper aimed to construct the optimal leaf area eatimation mod- el for 6 species in Passiflora. The empirical model for leaf area estimation can provide basic parameters for studying the physiological ecology and cultivation techniques for Passifloraceae Passiflora【. Meth- ods】In this study, six species of Passifloraceae Passiflora, i.e., P. edulis, P. yucatanensis, Lady Marga- ret, P. foetida, P. caerulea and P. laurifolia served as the research samples and these plants were planted in Kedu Town Passion Fruit Production and Academic Research Base, Guizhou Province Mountain Re- sources Research Institute in March 2020. In August 2020, sample plants were randomly selected with the same growth vigor from each species, good-growing branches were randomly selected from each tree, and about 200 sample mature leaves with typical characteristics and without pests and diseases were picked up from each tree species on 2th, 4th, 6th, 8th and 10th internodes from the end of the branches. After the sample leaves were collected, they were marked, packaged and brought back to the laboratory. Place the leaf flat on an Epson850 scanner and scan it to obtain the leaf image. Use the Digimizer digital image method to measure the leaf length, width and area. Take the maximum length of the leaf as L, the approximate maximum width of the leaf as W, and the area as LA. Select 12 target models, including linear and non-linear equations, respectively, to construct LA empirical models for each species. Select two species with similar leaf shapes to merge modeling, and screen and evaluate these models, so as to select the basis for the optimal empirical model. The optimal empirical model was selected according to the R2, AIC and RMSE values of each empirical model. The optimal empiri- cal model was selected as the model with the R2 closest to 1, minimum AIC and RMSE values. When the best AIC difference between the first two empirical models is less than 2, the model with the smaller RMSE value is selected as the optimal empirical model. In order to further evaluate the reliability of the empirical model, the predicted value of LA is calculated based on the optimal empirical model, and the distribution of its residuals is analyzed. When the distribution is approximately normal and most residu- al points fall within ±3 times of the standard deviation of the residual average if the empirical model is within the range of the measured value of LA as y and the predicted value as x to test the fitting effect of the regression equation. According to the closeness of the slope of the regression line to 1, the size of the intercept and the coefficient of determination R2, the degree of agreement between the predicted val- ue and the actual value is determined. At the same time, based on the optimal experience model, the pre- diction accuracy of each species LA is calculated【. Results】The results showed that the R2 of leaf area modeling for each species was higher, and the optimal LA empirical model for each species were: P. edulis: LA=0.529 L1.141W0.868; P. laurifolia : LA=0.828L0.933W1.054; P. yucatanensis: LA=0.665L0.665W1.367; P. coerulea: LA=0.559L1.154W0.797; P. foetida: LA=0.763L0.64W1.327 and Lady Margaret: LA=0.6 L0.213. The op- timal LA empirical model of each species is the optimal independent variable to construct the LA empir- ical model. Each model meets various evaluation criteria, such as P. edulis, P. laurifolia, P. yucatanen- sis, P. caerulea, P. foetida and Lady Margare. The prediction accuracy of the optimal empirical model is 91.28%±0.58%, 94.35%±0.63%, 95.54%±0.27%, 92.59%±1.62%, 96.88%±1.34% and 94.25%±0.78%. The leaf surface parameters of two species with similar aspect ratios were merged and modeled, and the two species with similar leaf length and width ratios were P. edulis and Lady Margaret. The leaf surface parameters of two species with similar leaf length and width ratios were merged and modeled. The opti- mal empirical model predicted LA with accuracy of 92.63%±0.81%, and the optimal model for com-1.195 0.822 bined modeling is LA=0.52 L W 【. Conclusion】The LA empirical model was constructed to predict the LA of 6 species in Passifloraceae Passiflora and the LA prediction accuracy reached more than 90%. LA is reliable, ,two , species with similar leaf shapes and leaf length-to-width ratios can be com- bined and modeled, and the model accuracy is(92.63±0.81)%. The construction of the LA empirical model can classify the leaves according to the leaf shape and leaf length-to-width ratio, and build the LA model with certain reliability. Using 6 species of Passiflora plants in the Passifloraceae as research materials, the computer image method is used to measure the leaf length (L), leaf width (W), and leaf ar- ea (LA) of each species, with LA as the dependent variable and L, W and LL, WW, LW as the indepen- dent variables. Twelve target models are selected, the LA estimation empirical model for each species is constructed respectively, two species with similar leaf shapes are selected and combined for modeling, and these models are screened and evaluated. With a view to constructing an empirical model for esti- mating the optimal LA of 6 species of Passiflora plants in the Passifloraceae, it will provide a basic the- ory for the study on Passiflora plants in the Passifloraceae, and will preliminary discuss the general rules of LA model construction. The results show that the optimal empirical models of the six species: P. edulis:LA=0.529 L1.141W0.868;P. yucatanensis:LA=0.828 L0.933W1.054;Lady Margaret:LA=0.665 L0.665W1.367;P. foetida:LA=0.559 L1.154W0.797;P. caerulea:LA=0.763L0.64W1.327 and P. laurifolia :LA=0.6 L0.213. The optimal LA empirical model of each species corresponds to the optimal independent variable of the LA empirical model. Studies have shown that the LA empirical model is constructed to predict the LA of these 6 species of Passiflora plants in the Passifloraceae, and the LA prediction accuracy is above 90%, indicating that the LA empirical model is reliable. The combined modeling of two species with similar leaf shape and leaf length-to-width ratio yields a model prediction accuracy of 92.63%± 0.81%, which preliminarily shows that the combined modeling of tree species with similar leaf shape and leaf length-to-width ratio has certain reliability to predict LA.