Improved estimation of cotton (Gossypium hirsutum L.) LAI from multispectral data using UAV point cloud data.

Real-time monitoring of leaf area index (LAI) in cotton (Gossypium hirsutum L.) plays a vital role in guiding field fertilization, water management, growth observation and yield prediction. The use of unmanned aerial vehicles (UAVs) equipped with diverse sensors enables flexible and rapid LAI measurement across extensive areas. This study evaluates the efficacy of UAV-mounted LiDAR and high-resolution camera-derived point cloud data in LAI prediction. We assessed the integration of canopy spectral-textural characteristics from multispectral data with structural features from point cloud data for LAI forecasting. Furthermore, we compared various machine learning and deep learning models, selected the optimal one, and applied the SHAP (Shapley Additive Explanations) method to identify key features and their influence patterns in this model. The findings are distilled into four key points: (1) The performance of canopy structure metrics based on the two sensors varied across fertility periods due to differences in canopy closure; (2) The DNN-based (Deep Neural Network) LAI prediction model excelled with a single-period dataset, achieving an R2 of 0.81 and an RMSE% of 11.36 %. Similarly, in full-period multimodal data fusion, it demonstrated superior performance, evidenced by an R2 of 0.84 and an RMSE% of 9.94 %. (3) Compared to the unimodal data model, the multimodal data model yielded superior results and exhibited greater robustness. (4) In the DNN-based LAI prediction model utilizing multimodal data, texture features contributed most significantly. The results suggest that the DNN model, when employing multimodal data fusion, offers not only relatively precise and robust estimates of crop LAI but also contributes valuable insights for crop phenotyping and enhanced field management. This approach subsequently improves spatial prediction accuracy and the quality of decision-making in crop production. • Canopy structure metrics improve cotton LAI estimation accuracy. • Point cloud data and multispectral fusion improve LAI prediction accuracy. • Compared two sensors' structural indicators for LAI prediction. • DNN performs best in LAI prediction. [ABSTRACT FROM AUTHOR]