High resolution wheat yield mapping using Sentinel-2.

Accurate crop yield estimates are important for governments, farmers, scientists and agribusiness. This paper provides a novel demonstration of the use of freely available Sentinel-2 data to estimate within-field wheat yield variability in a single year. The impact of data resolution and availability on yield estimation is explored using different combinations of input data. This was achieved by combining Sentinel-2 with environmental data (e.g. meteorological, topographical, soil moisture) for different periods throughout the growing season. Yield was estimated using Random Forest (RF) regression models. They were trained and validated using a dataset containing over 8000 points collected by combine harvester yield monitors from 39 wheat fields in the UK. The results demonstrate that it is possible to produce accurate maps of within-field yield variation at 10 m resolution using Sentinel-2 data (RMSE 0.66 t/ha). When combined with environmental data further improvements in accuracy can be obtained (RMSE 0.61 t/ha). We demonstrate that with knowledge of crop-type distribution it is possible to use these models, trained with data from a few fields, to estimate within-field yield variability on a landscape scale. Applying this method gives us a range of crop yield across the landscape of 4.09 to 12.22 t/ha, with a total crop production of approx. 289,000 t. • Within-field yield variability is mapped at landscape-scale using Sentinel-2 data. • Yield is estimated using Random Forest models trained with combine harvester data. • Impact of data resolution, estimator variables and data availability is quantified. • Sentinel-2 data produces accurate yield estimates (RMSE 0.66 t/ha). • Accuracy improves when environmental data is added (RMSE 0.61 t/ha). [ABSTRACT FROM AUTHOR]