Estimation of sunflower yields at a decametric spatial scale - A statistical approach based on multi-temporal satellite images

Earth observation capabilities provided by satellite missions constitute useful tools in agricultural management, particularly in the context of forecasting yields. Recent advances in sensors onboard harvesting machines allow accessing the intra-plot variability of yields, spatial scale fully compatible with numerous on-going satellite missions. In this context, the aim of this study is to estimate the sunflower yield at the intra-plot spatial scale using the multi-temporal satellite images provided by the Landsat-8 and Sentinel-2 missions. The proposed approach is based on random forest or artificial neural networks, testing different sampling strategies to partition the dataset into independent training and testing sets: a random selection by testing different ratio of data, a systematic selection by focusing on different plots, and a forecast procedure by using an increasing number of satellite images. Emphasis is put on the use of high spatial and temporal resolution satellite data acquired throughout the agricultural seasons 2016 and 2017, on a study site located in southwestern France, near Toulouse. Ground measurements consist in intra-plot yields collected over ~250 hectares by a surveying harvesting machine with GPS system on track mode. Interesting accurate statistical performances are obtained regardless the considered sampling strategy, providing complementary information useful for the yield retrieval. The results based on the random selection are satisfactory for a large range of tested ratio, with for instance R² upper than 0.64 and RMSE lower than 0.45 tons per hectare (t ha^-1), on the 50% of independent data used to validate the approach. While the systematic selection allows analyzing the plot representativeness, the forecast of yield throughout the agricultural season provides early accurate estimation during the crop flowering (two months before the harvest), with R² equal to 0.59 or 0.66 and RMSE of 0.47 or 0.34 t ha^-1, for the agricultural seasons 2016 and 2017 respectively.