Comparative Assessment of Remote and Proximal NDVI Sensing for Predicting Wheat Agronomic Traits

Abstract

Highlights: What are the main findings? The UAV-derived NDVI values exhibited greater stability, higher normality, and superior predictive performance for wheat traits compared to proximal sensing. The observed differences between the platforms underline the importance of considering sensor characteristics and the spatial sampling strategy when designing workflows for agronomic prediction. What are the implications of the main findings? UAV-based sensing is better suited for operational field phenotyping and large-scale agronomic monitoring due to its higher spatial integration and reduced background noise. Sensor selection significantly affects NDVI reliability and predictive modeling, emphasizing the need for platform-specific calibration in precision agriculture. Monitoring wheat traits across diverse environments requires reliable sensing tools that balance accuracy, cost, and scalability. This study compares the performance of proximal and UAV-derived NDVI sensing for predicting the key agronomic traits in winter wheat. The research was conducted at a long-term NPK field experiment on Haplic Chernozem soils in Rimski Šančevi, Serbia, using UAV multispectral imagery and a handheld proximal sensor to collect NDVI data across 400 micro-plots and six phenological stages. The UAV-derived NDVI achieved a higher mean value (0.71 vs. 0.60), lower coefficient of variation (29.2% vs. 33.0%), and stronger correlation with the POM readings (R2 = 0.92). For trait prediction, the UAV-based NDVI reached R2 values up to 0.95 for grain yield and 0.84 for plant height, outperforming the POM (maximum R2 = 0.94 and 0.83, respectively), and it showed superior temporal consistency (average R2 = 0.74 vs. 0.64). Although the POM performed comparably during mid-season under controlled conditions, its sensitivity to operator handling and limited spatial resolution reduced robustness in more variable field scenarios. A cost–benefit analysis revealed that the POM offers advantages in affordability, ease of use, and deployment in small-scale settings, while UAV systems are better suited for large-scale monitoring due to their higher spatial resolution and data richness. The findings highlight the importance of selecting sensing technologies based on biological context, operational goals, and resource constraints, and suggest that combining methods through stratified sampling may improve the efficiency and accuracy of crop monitoring in precision agriculture.