Estimation of Rice Leaf Area Index Coupling Feature Selection from Multi-Temporal UAV Imagery with Machine Learning

Leaf Area Index (LAI) is a key biophysical parameter characterizing crop canopy structure and growth status. Large-scale, dynamic, and high-precision estimation of LAI using remote sensing is critical for crop growth assessment and yield forecasting. This study investigates rice LAI estimation across multiple growth stages by integrating optimally selected features from multi-temporal unmanned aerial vehicle (UAV) imagery into modeling algorithms. Based on controlled experiments with varying nitrogen fertilizer gradients, rice field LAI data were synchronously collected alongside high-resolution red–green–blue (RGB) and multispectral UAV images. Vegetation spectral indices were then extracted, and principal component analysis (PCA) was performed on these indices to obtain the first principal component (PC1). Subsequently, gray-level co-occurrence matrix (GLCM) calculations were conducted on the PC1 image to extract texture features. Feature selection was performed separately using recursive feature elimination (RFE) guided by Shapley Additive Explanations (SHAP) for random forest (RF), variable importance in projection (VIP) for partial least squares regression (PLSR), and backward elimination for stepwise regression (SR). Each model was developed using its corresponding selected predictors for individual growth stages and the combined dataset. The results demonstrated that the RF model for the entire growth cycle performed the best, with a coefficient of determination (R2) of 0.9197, a root mean square error (RMSE) of 0.4309, and a relative root mean square error (rRMSE) of 16.85%. The estimated LAI maps clearly revealed the spatial distribution patterns and gradient variations in LAI under different growth stages and nitrogen treatments. Ultimately, these findings demonstrate that integrating multi-source, multi-temporal UAV remote sensing features with modeling algorithms enables the high-precision retrieval of rice LAI, effectively capturing within-field growth heterogeneity. This study provides reliable technical insights for dynamic crop monitoring, variable-rate fertilization decision-making, and smart agricultural management.

Authors

Institutions

Publication Details

Journal
Agronomy
Published
2026-11-13
DOI
https://doi.org/10.3390/agronomy16222297
Primary Topic
Remote Sensing in Agriculture
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Estimation of Rice Leaf Area Index Coupling Feature Selection from Multi-Temporal UAV Imagery with Machine Learning

Xiaoyu Song, Zhenhai Li, Guijun Yang, Hanyu Xue et al.
6 links
Agronomy
Remote Sensing in Agriculture
article

Estimation of Rice Leaf Area Index Coupling Feature Selection from Multi-Temporal UAV Imagery with Machine Learning

Xiaoyu Song, Zhenhai Li, Guijun Yang, Hanyu Xue, Xingang Xu, Yuekun Song, Tuo Wang, Xinlei Xu, Yang Meng
article en

Abstract

Leaf Area Index (LAI) is a key biophysical parameter characterizing crop canopy structure and growth status. Large-scale, dynamic, and high-precision estimation of LAI using remote sensing is critical for crop growth assessment and yield forecasting. This study investigates rice LAI estimation across multiple growth stages by integrating optimally selected features from multi-temporal unmanned aerial vehicle (UAV) imagery into modeling algorithms. Based on controlled experiments with varying nitrogen fertilizer gradients, rice field LAI data were synchronously collected alongside high-resolution red–green–blue (RGB) and multispectral UAV images. Vegetation spectral indices were then extracted, and principal component analysis (PCA) was performed on these indices to obtain the first principal component (PC1). Subsequently, gray-level co-occurrence matrix (GLCM) calculations were conducted on the PC1 image to extract texture features. Feature selection was performed separately using recursive feature elimination (RFE) guided by Shapley Additive Explanations (SHAP) for random forest (RF), variable importance in projection (VIP) for partial least squares regression (PLSR), and backward elimination for stepwise regression (SR). Each model was developed using its corresponding selected predictors for individual growth stages and the combined dataset. The results demonstrated that the RF model for the entire growth cycle performed the best, with a coefficient of determination (R2) of 0.9197, a root mean square error (RMSE) of 0.4309, and a relative root mean square error (rRMSE) of 16.85%. The estimated LAI maps clearly revealed the spatial distribution patterns and gradient variations in LAI under different growth stages and nitrogen treatments. Ultimately, these findings demonstrate that integrating multi-source, multi-temporal UAV remote sensing features with modeling algorithms enables the high-precision retrieval of rice LAI, effectively capturing within-field growth heterogeneity. This study provides reliable technical insights for dynamic crop monitoring, variable-rate fertilization decision-making, and smart agricultural management.

AgronomyVol. 16(22)
Beijing Academy of Agricultural and Forestry Sciences (CN), Ministry of Agriculture and Rural Affairs (CN), Shandong University of Science and Technology (CN)
National Natural Science Foundation of China, National Key Research and Development Program of China
Zero hunger
Remote Sensing in Agriculture
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon