Using SAR data as an effective surrogate for optical data in nitrogen variable rate applications: A winter wheat case study

This study highlights the feasibility of using SAR data as a surrogate for optical acquisitions in the generation of nitrogen prescription maps in wheat cultivation. Unlike the optical-based approaches which are negatively affected by adverse meteorological conditions, the proposed strategy provides the possibility to compute the fertilization maps at any date by exploiting the all-weather, day-and-night SAR capabilities. We train a U-Net-like CNN architecture on the Sentinel-2 optical and Sentinel-1 SAR datasets after a properly alignment in time. The trained model returns a surrogate NDVI distribution starting from SAR acquisitions, when optical data are not available. The recovered NDVI information is converted into LAI and GAI distributions, by resorting to an exponential and a linear law, respectively, according to the literature. Finally, the nitrogen prescription map is obtained out of the recovered GAI values. A qualitative and quantitative analysis of the error between the optical and SAR-derived prescription maps shows that the procedure is accurate, especially during the tillering and the stem elongation growth phases.