Improved estimation of nitrogen use efficiency in maize from the fusion of UAV multispectral imagery and LiDAR point cloud

Abstract

Nitrogen use efficiency (NUE) is a key indicator for selecting nitrogen-efficient crop cultivars and optimizing fertilization strategies. However, NUE is typically assessed using destructive and laborious sampling methods, hindering the advancement of sustainable agriculture. The objective is to test whether the fusion of phenotyping data simultaneously acquired by multi-source sensors that reflect more functional and structural traits can improve the estimation accuracy of the highly integrated trait NUE in maize. Multispectral (MS) and light detection and ranging (LiDAR) data were simultaneously acquired during critical growth stages across two years of maize cultivar and nitrogen fertilizer field experiments using a multi-sensor UAV platform. Three machine learning algorithms, Partial Least Squares Regression (PLSR), Random Forest Regression (RFR) and Support Vector Machine Regression (SVR) were selected to construct NUE estimation models based on three data sources：MS, LiDAR, and MS+LiDAR. The results demonstrated distinct differences in nitrogen utilization efficiency (NU_tE) and nitrogen agronomy efficiency (NAE) among maize cultivars at critical growth stages. These differences were efficiently and accurately identified using multi-source data combined with the machine learning algorithms. The RFR method obtained the highest model validation accuracy with an average R_test² = 0.68 and RMSE_test = 6.66 kg kg⁻¹. The average accuracy of multi-source data fusion was improved by 20.21 % compared to a single data source, and the RFR+MS+LiDAR method for NU_tE estimation obtained the highest model accuracy in the two-year validation dataset with R_test² = 0. 86 and RMSE_test = 8.5 kg kg⁻¹. The method proposed in this study mitigates the impact of canopy spectral saturation during the late growth stages of maize, enhancing the accuracy of NUE estimation by improving the convergence between predicted and observed values. This multi-source data fusion approach, based on a UAV platform, enables effective monitoring of NUE at critical growth stages. Consequently, it advances rapid, non-destructive NUE assessment in maize, supporting efficient breeding and precision nitrogen management strategies.