Can the satellite image resolution be improved to support precision agriculture in the vineyard through vegetation indices?

Aim: This study aims to show the application of a new methodological approach to improve the resolution of Sentinel-2A images and derived vegetation indices through the results from different vineyards. 

Methods and Results: A multiscale fully-connected Convolutional Neural Network (CNN) was constructed and applied for the pan-sharpening of Sentinel-2A images by high resolution UAS-based orthophoto. The reconstructed data was validated by independent high resolution multispectral UAS-based imagery and in-situ spectral measurements. The reconstructed Sentinel-2A images provided a temporal evaluation of plant responses to environmental factors using selected vegetation indices. The proposed methodology has been applied on different vineyards in southern Italy. Here, the outputs of CNN were compared with morpho-physiological data, both collected in-vivo and reconstructed through the retrospective analysis of vine trunk wood (tree-rings). The functional anatomical traits and isotopic signals were measured and used to derive indices such as water use efficiency. The obtained results showed a valuable agreement between the vegetation indices derived from reconstructed Sentinel-2A data and plant hydraulic traits obtained from tree-ring based reconstruction of vine eco-physiological behavior.

Conclusions: 

The multiscale CNN architecture for remote sensing imagery pan-sharpening and reconstruction can overcome the constraints in use of satellite images in precision agriculture, by creating new fused data valid for applications that could not be supported by the original Sentinel multispectral or UVS data. The validation of such an approach on different and real vineyard systems, with data collected in-vivo and through retrospective analyses on tree-ring chronologies has shown great potential to extend the approach to other woody crop systems. 

Significance and Impact of the Study: The integration between knowledge from different scientific domains represents a powerful approach to support the farmer in the field management and, at the same time, a valuable opportunity to study the plant adaptation to variable pedo-climatic conditions. This represents the base for understanding the vine adaptive capability and planning the actions for vineyard management under different climatic scenarios. Finally, emerging CNN methodologies can be implemented in DSS to support real-time monitoring of several parameters related to plant health status, to better follow plant growth in the field and evaluate its performance under changing environmental conditions.