Crop Yield Mapping with ARC using only Optical Remote Sensing

Abstract

Abstract. ARC is a new method to generates time series of a full set of biophysical parameters derived from optical EO. Here, we examine relationships between this ‘full’ set and maize yield. 15 Parameters per pixel are estimated over the US corn belt using ARC, to fully describe the phenology, soil, and crop status over time for typical behaviour. ARC is tested for a new model over an area of irrigated and rain-fed winter crop in South Africa. We find that care must be taken for episodic events, and robust filtering methods should be developed for ARC, but average magnitude and timing is well-expressed. We find that a robust yield model (over time and space) can be created at the county-level for maize using only EO parameters with RMSE of 704-938 kg/ha using a non-linear model, but the results are only slightly poorer if a linear model is used. It compares well to a model that also includes weather data, showing that a model can be driven by optical EO data alone.