Tillage direction analysis in agricultural fields from Digital Orthophotos and Sentinel-2 imagery

Abstract

For questions of soil and water protection, knowledge about agricultural management is relevant, especially in hilly and mountainous areas. In sloping areas, an area-wide knowledge of whether farming is done with or across the contour line would be very valuable for use in regional soil conservation management. In order to ascertain the prevalence of farming practices conducted with or against the slope in a given region, it is necessary to obtain data on the direction in which fields are cultivated. This information can be derived from remote sensing data through the application of geographic information system (GIS) methods. While previous studies have attempted to provide knowledge primarily through the use of small-scale but high-resolution Unmanned Aerial Vehicle (UAV) imagery, this study used medium-resolution imagery from satellite imagery (Sentinel-2 at 10 m × 10 m) and high resolution imagery (0.2 m × 0.2 m) Digital Orthophotos (DOP) from aircraft flights.

The use of medium-resolution satellite images (such as Sentinel-2) has yet to be explored in the context of addressing this research question, and this study represents their preliminary application in this domain. For this purpose, two GIS-based methods of analysis were proposed, which mainly made use of high-pass filtering, reclassification, vectorization, and compass orientation calculation. The results are promising, as in the best cases the correlation, between processing and ground truth orientation of the field tillage direction, for the DOP is R² of 0.867 for 170 fields and 2.687 ha. For the Sentinel-2 evaluation, an R² of 0.833 was obtained for 141 fields with 2.611 ha. Despite the different spatial resolution of both systems, the results are very comparable in terms of spatial coverage and accuracy of validation. However, for these two cases, this also meant that less than 50% of the total agricultural area and less than 20% of all fields in the study area could be covered. The data obtained from the DOP and Sentinel-2 sensors were collected at different times, resulting in the identification of distinct preferences for specific crop types. These preferences were observed to yield both accurate and less accurate evaluations, respectively. For instance, wheat exhibited favorable outcomes. Overall, the proposed approach demonstrated the capacity to derive area-wide information on farming direction with satisfactory results. Especially the temporarily high data availability of Sentinel-2 should be used to generate an overall picture using crop rotation and different phenological stages of arable crops in the long term.