Subtropical region tea tree LAI estimation integrating vegetation indices and texture features derived from UAV multispectral images

Abstract

Monitoring perennial tea tree leaf area index (LAI) provides essential insights into crop growth, and unmanned aerial vehicle (UAV) -based LAI estimation has recently proven effective across various crops. In subtropical regions, the conventional farming method (CFM), involves intensive irrigation and fertilization, and the agroecological farming method (AFM) emphasizes environmentally sustainable practices. This research employs machine learning (ML) to analyze thirty-four vegetation indices (VIs) and eight texture features (TFs) derived from UAV imagery, alongside in situ LAI measurements, to estimate tea tree LAI under CFM and AFM. Additionally, minimum redundancy-maximum relevance analysis was utilized to assess the mutual information of features for regression analysis using Polynomial Regression (PR), Ridge Regression, Decision Tree Regression, and Random Forest Regression (RFR) models. The results showed that: (1) AFM had higher in situ LAI values (mean=4.323, SD=1.594) compared to CFM (mean=3.901, SD=1.816), with less seasonal variation, mainly attributed to agronomic practices like harvesting and winter pruning. (2) Optimal image features for LAI estimation were identified by extracting pixel-based features from tea tree regions to enhance the correlation between LAI and imagery. (3) Combining VIs and TFs improved LAI estimation accuracy. The RFR model, using two VIs and one TF, achieved R²=0.710 and RMSE=1.357 for CFM, while the PR model, incorporating nine VIs and 10 TFs, had R²=0.540 and RMSE=1.283 for AFM. This study demonstrates that ML techniques are promising for estimating tea tree LAI using UAV-based multispectral imagery, offering practical guidelines for efficient field management.