Nondestructive Leaf Area Estimation for Chia

Abstract

Core Ideas Leaf area in chia cannot be accurately predicted by the product of leaf width and length. Regressing leaf area log linearly on width and length accounts for change of shape with size. We provide accurate prediction models valid across experiments, populations, and N levels. Mixed‐model meta‐regression allows integrating leaf area data across experiments. Leaf area (LA) is an important agronomic trait but is difficult to measure directly. It is therefore of interest to estimate LA indirectly using easily measured correlated traits. The most commonly used approach to predict LA uses the product of leaf width (LW) and leaf length (LL) as single predictor variable. However, this approach is insufficient to estimate LA of chia (Salvia hispanica L.) because the leaves are differently shaped depending on leaf size. The objectives of this study were to develop a nondestructive LA estimation model for chia using LW and LL measurements that can take differences in leaf shape into account and to determine whether population and nitrogen fertilizer level have an effect on the accuracy of a LA estimation model. A total of 840 leaves were collected from five different field experiments in 2015 and 2016 conducted in southwestern Germany. The experiments comprised eight populations of black‐ and white‐seeded chia (07015 ARG, 06815 BOL, 06915 ARG, G8, G7, G3, W13.1, and Sahi Alba 914) and three nitrogen fertilizer levels (0, 20, and 40 kg N ha−1). We used meta‐regression to integrate the data accounting for heterogeneity between experiments, populations, and nitrogen levels. The proposed LA estimation model with two measured predictor variables (LL and LW) was LA = 0.740 × LL0.866 × LW1.075 and provided the highest accuracy across populations and nitrogen levels [r = 0.989; mean squared deviation (MSD) = 2.944 cm4]. The model LA = 1.396 × LW1.806 with only LW was almost as accurate (r = 0.977; MSD = 5.831 cm4).