A data fusion and spatial data analysis approach for the estimation of wheat grain nitrogen uptake from satellite data

The selection of the optimal band combination for the estimation of specific crop variables is a key aspect in order to obtain reliable estimation of in-field variability from multi- and hyperspectral remote-sensing data. The selection of the bands is strongly influenced by the phenological stage of the crop at the acquisition time. In this work, the influence of the growing stage on the combination of spectral bands related to grain nitrogen (N) uptake in wheat was evaluated using multispectral (Satellite Pour l’Observation de la Terre – SPOT) and hyperspectral (Compact High Resolution Imaging Spectrometer – CHRIS-PROBA) satellite images at different growth stages over two wheat growth seasons in central Italy. In order to identify the more appropriate covariates (spectral bands) for each phenological stage, stepwise regression with backward selection was combined with stepwise variance inflation factors (VIFs) analysis and linear mixed effect model (LMEM). The results obtained in this study suggest that the spectral region most related to N uptake varies over the growing season of the wheat crop. For SPOT data, near-infrared (NIR) region was selected at all the phenological stages in both growing seasons, except for the latest stage, with low chlorophyll content due to the onset of senescence, in which the red band was selected. At stem elongation, the shortwave infrared (SWIR) band of SPOT data was also selected. At this stage, the best N estimation accuracy was obtained using an LMEM (root mean square error, RMSE = 0.012 t ha^?1). The inclusion of a spatial component in the estimation model by means of LMEMs provided a more accurate estimation than ordinary least square (OLS) models at all growth stages. The test carried out with CHRIS-PROBA data at the fourth stage confirmed the importance of NIR and in particular of the red-edge region for N uptake prediction. A novel methodology is proposed, which involves two crucial aspects in the context of the use of remote-sensing data in precision agriculture: i) the standardization of the spatial resolution for in-field and satellite data by a geostatistical data technique (data fusion); and ii) the selection of the most appropriate spectral bands for each phenological stage, taking into account both correlation with the target variable and collinearity.