Estimating Soybean Yield Spatial Variability Within-field Scale through Google Earth Engine in Northeast Italy

The study of spatial variability within agricultural fields is essential for all farmers who want to apply modern precision agriculture. This study investigated the possibility of estimating soybean yield spatial variability at field scale through different vegetation indices (VIs) derived from Sentinel-2 satellite images at different crop growth stages. The study considered yield records from seven fields located in North-East of Italy, with areas ranging between 10 to 19 ha and cultivated by soybean from 2016 to 2018. Sentinel-2 satellite images were used to calculate eight VIs through Google Earth Engine (GEE) between June to October. One-way ANOVA tested the linear correlation between yield and VIs measured at different soybean phenological stages corresponding to the available cloud-free Sentinel-2 images. Results showed that Green Chlorophyll Vegetation Index (GCVI), Green Normalized Difference Vegetation Index (GNDVI) and Wide Dynamic Range Vegetation Index (WDRVI) were the best correlated VIs with soybean yield variability. The highest correlation was observed between 85 and 105 days after sowing corresponding to grains forming and filling (phenological stage R4-R6). R2 values ranged between 0.21 and 0.68 across whole fields and growth stages. The study proved the effectiveness of a linear model exploiting the equation of the regression line between the VIs and soybean yield from the field with the highest correlation. The model showed high yield estimation accuracy results in 2018 and 2017 seasons with root mean square error (RMSE) of 0.47 and 0.49 Mg/ha respectively compared to less accuracy in 2016 where RMSE was 1.02 Mg/ha. This study approach proved the ability to estimate the within-field variability of soybean yield which could be applied to other Sentinel-2 archived images starting from 2015, while a new model should be calculated each year for each geographic region to ensure the estimation accuracy.