WHEAT TRAITS RETRIEVAL THROUGH MACHINE LEARNING AND HYPERSPECTRAL DATA: MODELLING PERFORMANCE AND INTERPRETATION

Crop traits monitoring is a fundamental step for controlling crop productivity in the context of precision agriculture and field phenotyping. Currently, the use of hyperspectral data in machine learning regression algorithms (MLRAs) has attracted increasing attention to alleviate the challenges associated with traditional crop trait measurements. In this framework, an experiment was set up to assess the performance of partial least squares regression (PLSR) and random forest (RF) models to estimate several wheat crop traits (leaf area index: LAI, canopy water content: CWC, canopy chlorophyll content: CCC, and canopy nitrogen content: CNC) at the canopy level, using full-range hyperspectral data (350 – 2500 nm) as inputs. The study compared the performance of the two MLRA focusing on the physical interpretation of the results for each particular crop trait. Overall, PLSR provided remarkably higher accuracy, tested with a cross-validation strategy, as compared to RF for all the crop traits. In particular, PLSR denoted R2 (nRMSE%) values of 0.72 (11.97%), 0.77 (10.89%), 0.70 (14.61%), and 0.74 (14.38%) for LAI, CWC, CCC, and CNC, respectively. All PLSR models indicated robust prediction capability (RPD > 1.4). In general, analysis of band importance revealed physically-meaningful and consistent patterns for each specific crop trait.