Estimation of grain protein content in commercial bread and durum wheat fields via traits inverted by radiative transfer modelling from Sentinel-2 timeseries.

Wheat (Triticum spp.) is crucial to food security. Grain protein content (GPC) is key to its nutritional and economic value and is controlled by genetic and agronomic factors, soil properties and weather. GPC prediction from remote sensing could reduce nitrogen (N) losses, help management decisions, and improve profit. However, GPC prediction is complex because multiple plant traits influence GPC and their effects change through the growing season. Traits with known physiological links to GPC, which can be retrieved from imaging spectroscopy, include leaf area index (LAI), chlorophyll (C a+b), and stress indicators. Further inspection of these and other traits retrieved from satellite data can advance research relevant to precision agriculture. Sentinel-2 (S2) timeseries (TS) were acquired for 6,355 ha of commercial dryland bread (T. aestivum) and durum (T. durum) wheat fields in south-east Australia through two consecutive years with dissimilar rainfall. Wheat growers provided ∼ 92,000 GPC data points from harvester-mounted protein monitors. For each, C a+b , leaf dry matter, leaf water content (C w) and LAI were retrieved from the S2 images by radiative transfer model inversion. A gradient boosted machine learning algorithm was applied to analyse these traits' importance to GPC and to predict GPC in 30% of samples unseen by the algorithm in training. The strongest relationships between predicted and observed GPC (R 2 = 0.86, RMSE = 0.56 %), in a model built from five S2 images across a season, were better than those from single-date hyperspectral (HS). In severe water stress, LAI was the main predictor of GPC early in the season, but this switched to C w later. Trait importance was more evenly distributed in milder conditions. S2 TS had a clear accuracy advantage over single-date S2 and HS, especially in benign conditions, emphasising the potential of S2 TS for large-scale GPC monitoring. [ABSTRACT FROM AUTHOR]