Calibration of food and feed crop models for sweet peppers with Bayesian optimization
2023
Moon, T.W. | Sim, S.E. | Son, J.E.
Crop models are tools used to analyze the interaction of crops and the environment. Since crop models can be applied to diverse research scales and purposes, models and their modifications vary. The parameters of a crop model could be biased for unseen data; thus, crop models should be calibrated for the adequate simulation of the given data. In this study, we aimed to calibrate food and feed crop models for sweet peppers (Capsicum annuum var. annuum) using Bayesian optimization. The algorithm does not require domain knowledge because it only considers input and output distributions based on Bayesian probability. For the implementation of Bayesian optimization, HyperOpt, an algorithm for optimizing high-dimensional hyperparameters, was used. The target growth factors were fruit yield and leaf area index, and the loss function was mean squared error (MSE). As a result, the calibrated crop model showed the highest modeling efficiency (EF) of 0.53, compared to − 1.91 and 0.62 from NLopt, a nonlinear optimization methodology, and random walk, respectively. The methodology showed adequate performance with reasonable ranges of convergence. The optimization method can be used for unknown distribution spaces of parameters because it does not require an initial status. Among the selected food crops, the groundnut model was suitable for sweet pepper. Since the optimized crop models yielded reasonable simulations, Bayesian optimization could be introduced for horticultural purposes. However, more data could be required to ensure convergence of the parameters and construct a robust crop model.
显示更多 [+] 显示较少 [-]