Correlation between satellite multispectral imagery and combine harvester CANbus data for corn yield assessment

Adoption of digital tools has led to relevant advancements in the agricultural sector. In particular, precision agriculture techniques have shown effective applications in farming practices. Integrating multisource data is a crucial task for improving agricultural management, and in this context the reliable assessment of crop yield by remote sensed imagery plays a relevant role. In this work, the correlation between satellite multispectral imagery and corn yield was investigated. Sentinel-2 satellite mission was selected as source of multispectral data, with 10-m spatial resolution and 4–6 days revisit time. A corn field in Ferrara was considered as a case study, with the dataset consisting of 17 multispectral images acquired in days with cloud coverage under 7%. The reference yield map was computed using CANbus data from a combine harvester, and its correlation with NDVI and GRVI indices was explored throughout the whole corn life cycle. In addition, the effect of applying a gaussian filter in the raster CY spatial distribution was explored. Results showed an overall good correlation between remotely sensed tiles and in-field farm data. The Pearson correlation coefficients showed a sharp increase during the vegetative stage of the crop (May and June), followed by a slowly decreasing plateau in July and August dates. Imagery from early June provided the highest correlations. The application of a gaussian filter in the CY map revealed an enhance in correlation of more than 25%. These results pave the path to the development of machine learning based methods exploiting multi-band-multi-temporal data for CY estimation.