Linear mixed model analysis of NDVI-based canopy coverage, extracted from sequential UAV multispectral imagery of an open field tomato irrigation experiment

Abstract

This study presents the use of a linear mixed model to accurately estimate the dynamics of canopy cover (CC) during an irrigation experiment on a tomato crop, performed in the Peruvian coastal desert. Initially, the normalized difference vegetation index (NDVI)-based CC was computed using multispectral sensors over consecutive weekly UAV flights, during the growing season of the crop. The drone high-resolution sequential imagery, combined with image segmentation on plant level, results in repeated measures growth curves of the coverage per plant. To analyze these repeated measures data per plant, cubic polynomials without intercept showed the best goodness-of-fit of the CC dynamics per plant. As a consequence, this polynomial was incorporated in a linear mixed model (LMM) as a random coefficient model to fit the plant-specific time evolution of CC as deviations from the mean time effects per irrigation treatment. The mixed model approach is capable to estimate the coverage curves per plant with high accuracy and a very limited number of model parameters. The results and statistical analysis demonstrate the potential benefits of the linear mixed model for incorporating plant-specific random components in addition to classical fixed effects models. Thus, the proposed linear mixed canopy model is a very efficient way to model plant-specific growth curves together with the treatment and design structure of the experiment.