Leaf area modeling of bell pepper (Capsicum annuum L.) grown under different stress conditions by soft computing approaches

Abstract Some leaf area (LA) estimation models have been developed for different plants under optimum conditions, but to date, none has been developed to model for those grown under stress conditions. In this study, LA of bell pepper grown under different levels of irrigation water salinity (IWS) and irrigation regimes (IR) were estimated by means of comparing different procedures including a simple model derived from ellipse area (EM), parabolic model (PM), geometric model (GM), multiple linear regression analysis (MLR), and artificial neural networks (ANN). To this end, two experiments were carried out under greenhouse conditions. First, the LA of bell peppers grown under five IWS levels were identified. In the second experiment, LA was determined under four different IR. Besides the general models elicited from EM, PM, GM, MLR, and ANN for each stress condition, prediction models of the bell peppers for each treatment under both stress conditions also were validated. Performance of the models also were evaluated using root mean square errors (RMSE), mean absolute errors (MAE), coefficient of determination (R2) and a Taylor diagram, which illustrates the accuracy of the models in a concise statistical analysis of how well the correlation (r) and standard deviation (SD) patterns match. Based on these results, the ANN model produced more reliable LA estimations compared to MLR, EM, PM, and GM. The R2, RMSE and MAE values were ranged 0.96–0.99, 1.05–2.99 cm2, and 0.78–1.12 cm2 in all ANN models. Overall, the ANN models are a valuable tool to investigate and understand the estimation of the LA of the bell peppers grown under different levels of IWS and IR.