RETRACTED: Enhanced Crop LAI Estimation via Random Forest Regression: Bayesian Optimization and Feature Selection Approach

Abstract

LAI (Leaf area index) is an important structural parameter closely linked to the photosynthetic capacity and biomass changes of crops. The combination of machine learning algorithms and spectral variables has demonstrated superior performance in LAI estimation compared to traditional methods. However, too many input parameters may lead to data redundancy and reduced computational efficiency. Reasonable hyperparameters combination are beneficial to the performance of LAI estimation models, yet existing studies have paid less attention to this aspect. In this paper, a model framework based on Bayesian optimized random forest regression (bayes-RFR) is constructed. The framework adequately extracts important features for estimating crop LAI using a tree-model feature selection method. It uses a Gaussian process as an a priori model to determine the sampling strategy and construct the optimal hyperparameter combination. The robustness of the proposed model was tested by conducting field planting experiments of maize and wheat, simultaneously acquiring LAI and canopy spectra during 2021 and 2022. The results demonstrate that the tree model-based feature selection method adequately extracted important features for estimating crop LAI, surpassing correlation analysis. The bayes-RFR approach significantly improved the accuracy of the LAI model estimation compared to the traditional RFR method. This indicates that the LAI estimation model, optimized with Bayesian algorithms for hyperparameters, offers enhanced stability and predictive ability.