Modelling of the leaf area for various pear cultivars using neuro computing approaches

Abstract

Aim of study: Leaf area (LA) is an important variable for many stages of plant growth and development such as light interception, water and nutrient use, photosynthetic efficiency, respiration, and yield potential. This study aimed to determine the easiest, most accurate and most reliable LA estimation model for the pear using linear measurements of leaf geometry and comparing their performance with artificial neural networks (ANN). Area of study: Samsun, Turkey. Material and methods: Different numbers of leaves were collected from 12 pear cultivars to measure leaf length (L), and width (W) as well as LA. The multiple linear regression (MLR) was used to predict the LA by using L and W. Different ANN models comprising different number of neuron were trained and used to predict LA. Main results: The general linear regression LA estimation model was found to be LA = -0.433 + 0.715LW (R-2 = 0.987). In each pear cultivar, ANN models were found to be more accurate in terms of both the training and testing phase than MLR models. Research highlights: In the prediction of LA for different pear cultivars, ANN can thus be used in addition to MLR, as effective tools to circumvent difficulties met in the direct measurement of LA in the laboratory.