Comparison between artificial neural networks and some mathematical models in leaf area estimation of Red Chief apple variety

Abstract

Leaf area index is an important variable in ecological and physiological studies. This studywas aimed to determine the most suitable model explaining the leaf area estimation andweekly growth of leaf parameters in Red Chief apple variety. In the first part of the study, theleaf area was modeled through two different models (Model-1 and Model-2) developed basedon ANN and power function (LA= AxB). In the second part, the weekly growth of each of theleaf width, length and area parameters were analyzed according to the Gompertz and Logisticsfunction. The results of analysis revealed that leaf area estimations performed by ANN(Training: R2= 0.98, RMSE= 0.922, MAD= 0.614, MAPE= 4.22; Testing: R2= 0.94,RMSE= 3.346 MAD= 1.889 MAPE= 4.88) were more successful than Model-1 and Model-2.In addition, Gompertz has come to the fore as the model that best describes the weekly growthin all leaf parameters (Width: R2= 0.98, RMSE= 0.154, MAD= 0.134, MAPE= 3.65, Length:R2= 0.98, RMSE= 0.180, MAD= 0.145, MAPE= 2.26 and Leaf area: R2= 0.99, RMSE= 0.73,MAD= 0.654, MAPE= 4.60).