Development of a color-based, non-destructive method to determine leaf N levels of Hass avocado under field conditions

Abstract

Excessive fertilization in avocado trees might be avoided by providing producers with affordable supporting tools for constant monitoring of nutrient levels. Leaf color guides have been produced for cereals and might be useful, but they are so far rare for trees because of low variation in color. We investigated the potential of leaf color to indicate N and P levels in avocado leaves to develop a monitoring tool not requiring expensive chemical analyses. We carried out three experimental phases towards the development of a solid, reproducible monitoring tool. In the first phase, we found a good relation between color and chemically-measured N levels, but not P levels. That allowed us to develop a leaf color chart only for N levels. In the second phase, this visual guide was tested using print and mobile app versions. We found that visual identification of N levels by the users was highly variable, subjective, and prone to error regardless of the materials used for detection. The third phase aimed to develop a user-independent evaluation of leaf color to define the leaf N level using leaf pictures. Machine and deep learning algorithms were used to generate, calibrate, and validate models for estimating the N concentration of avocado leaves using digital images captured in field conditions. Applying the models generated, we can now develop an automated color detection and N-level identification tool for mobile applications that will assist avocado producers in adequate application of nitrogen fertilizers, saving money and reducing N pollution from leaching in orchards.