Airborne hyperspectral and Sentinel imagery to quantify winter wheat traits through ensemble modeling approaches

Abstract Early prediction of crop production by remote sensing (RS) may help to plan the harvest and ensure food security. This study aims improve quantification yield, grain protein concentration (GPC), nitrogen (N) output in winter wheat with RS imagery. Ground-truth traits were measured at flowering a field experiment combining four N two water levels central Spain over 2 years. Hyperspectral thermal airborne images coincident Sentinel-1 Sentinel-2 acquired flowering. A parametric linear model using all hyperspectral normalized difference spectral indices (NDSI) non-parametric models (artificial neural network random forest) used assess their estimation ability NDSIs other indicators. The feasibility freely available multispectral satellite was tested applying same methodology but bands. Yield obtained highest R value, showing that visible short-wave infrared region (VSWIR) had similar accuracy imagery ( ≈ 0.84). SWIR bands important GPC both sensors, whereas better estimated red-edge-based NDSIs, obtaining satisfactory results sensor = 0.74) 0.62). When including index, NDSI (B11, B3) improved 0.71). Ensemble based on Sentinel found be as reliable those imagery, information N-related traits.