Abstract:

【Objective】 This study aims to establish an accurate and non-destructive model for apple leaf area estimation.【Methods】 Leaves were collect from vegetative and bourse shoots considering their length for Fuji and Gala apple trees.Leaf morphological parameters, such as leaf length (LL), leaf width (LW) and leaf area (LA), were obtained by digital scanner.17 candidate models with or without constant were evaluated by determination coefficient (R²), root mean square error (RMSE) and Akakchi Information Criterion (AIC).【Results】 Leaf morphological characteristics of 5,207 leaves were obtained, and among those leaf are had the highest coefficient of variation by 51.59%.Leaf morphologies were significantly affected by shoot types and cultivars The LL, LW and LA of long shoots were significantly larger than those of short shoot, and those were significantly higher in vegetative shoot than bourse shoot when the shoots belong to same category determined by shoot length.Leaves in Gala were more slender than those in Fuji.The accuracies of following selected models where LL and LW were used as independent variables can meet the requirements for leaf area estimation in apple trees, including model 5: LA = a(LL×LW), model 6: LA = a(LL+LW)², model 9: LA = aLL²+bLW², model 16: LA = a(LL×LW)^b, and model 17: LA = (LL×LW)^b., but model need to be established for each shoot type and cultivar.Those models had both LL and LW variables were more reliable than those models only had LL or LW solely.Among shoot types, regarding the accuracy of selected models, bourse was the highest, short shoot was higher than that of long shoot, and vegetative long shoot was the lowest.The all-data model 5 (R²= 97.72%, RMSE = 1.756,6 cm²) and 17 (R²= 97.37%, RMSE =1.831,9 cm²) were available to estimate the leaf area no matter of cultivars or shoot types with same model coefficients.The accuracies of the models 5 and 17 were not improved by adding constant.The model coefficients for model 5 and 17 were 0.712,3 and 0.910,8, respectively.The minimum number of samples was 241 and 3,939 for model 7 and 17, respectively; to have relative errors lower than 5%.【Conclusion】 The selected models in current study can be used for leaf area estimation irrespective of cultivars and shoot types, without the need of establishing model for each type of shoot and cultivar.The addition constant had no effect on selected model accuracy, and there is a minimum number of samples for model were required to have lower relative error.

Key words: apple; leaf area; model; akachi information criteria

摘要：

【目的】建立准确、无损的适宜于苹果不同品种和枝梢类型的叶面积估算模型。【方法】以富士及嘎啦不同长度的营养枝梢和果台枝梢叶片为试材,采用数字扫描仪获取叶片长度(LL)、宽度(LW)和叶面积(LA)等叶片形态参数,并采用决定系数(R²)、均方根误差(RMSE)、赤池信息准则(AIC)对建立的17个有常数项和无常数项叶面积模型精度进行筛选和适宜性评价。【结果】共获得5 207枚叶片形态参数,其中叶面积变异系数最大,达51.59%。叶片形态受品种及枝梢类型的显著影响,其中长枝梢叶片长、宽和叶面积显著大于同类型短枝梢叶片,而营养枝梢叶片长、宽和叶面积显著大于同长度果台枝梢,嘎啦叶片相比富士更为细长。以LL和LW复合变量为自变量的模型5:LA = a(LL×LW)、模型6:LA = a(LL+LW)²、模型9:LA = aLL²+bLW²、模型16:LA = a(LL×LW)^b、模型17:LA = (LL×LW)^b的精度可满足富士和嘎啦各类枝梢叶面积的估算,但需针对各品种和枝梢类型单独建模。同时包含LL和LW双变量的叶面积模型精度高于单一LL或LW变量的模型。无论富士还是嘎啦,对于不同类型枝梢,果台叶面积模型精度最高,短枝梢均比长枝梢精度高,营养长枝梢拟合精度最低。模型5和17可适用于富士和嘎啦各类枝梢叶面积的估算。模型5和17的R²分别为97.72%和97.37%,RMSE分别为1.756 6和1.831 9 cm²,模型系数分别为0.712 3和0.910 8。满足模型5和17精度所需的最低样品数量为241与3 939枚叶片。【结论】筛选的叶面积模型可用于各品种和各个类型枝梢叶片面积估算,无需针对单个品种及枝梢独立建模,且模型无需添加常数项。叶片数量影响叶面积模型精度,需满足最低临界值。

关键词: 苹果, 叶面积, 模型, 赤池信息准则