Multi-annual crop maps reveal cropping patterns in the Vojvodina region (Serbia)

Cropping patterns give useful information in agricultural production from the perspective of logistics and pricing. In this study, a machine learning classification model based on Random Forest algorithm and Sentinel-2 imagery was used to provide spatial maps of the crops in the Vojvodina region (Serbia), a predominantly agricultural area with $77 \%$ cropland and intensive production of arable crops. Ground truth data on geolocation of the fields was collected during the seven-year period $(2016-2022)$ to train the classification algorithm. The most common crops identified are wheat (together with barley), maize, soybean, sugar beet, sunflower and rapeseed. Crop maps resulting from the classification were analyzed. By multiplying the pixel count by the size of each pixel the percentage of area covered by most common crops within the region was calculated annually and averaged across all years. Multi-annual trends were also examined. For the sake of clarity, the most frequent rotations were analyzed at both the county and municipality level. The spatial distribution analysis highlighted variability in total area coverage by certain crops in the region. While some crops maintained a significant area coverage across all counties, others exhibited localized cultivation. Maize occupies the most areas in all counties $(29-39 \%)$, however there are cases where wheat is equally present as maize $(26-33 \%)$ and others where wheat is less cultivated in favor of soybean and sunflower. Counties with the most arable area covered by sunflower $(18-21 \%)$ are in the eastern part of the Vojvodina region, experiencing slightly drier conditions than counties in the western part of the region where soybean takes up $22-30 \%$ of arable land. Lastly, sugar beet and rapeseed together cover $2-9 \%$ of arable area in all counties and are consequently less present in rotation patterns. Notable variations in crop transition frequencies and spatial distribution patterns were revealed among all counties in the study region. Certain crop transitions demonstrated consistency across all counties, indicating widespread adoption of specific rotation schemes, others showed variability, reflecting local agronomic practices. The transition between maize and wheat emerged as the most frequent across all counties throughout the observed years, however at varying scales. It is observed that the counties in northern and southern parts of the region practice maizewheat transition most frequently ($18-32 \%$ of all transitions). Sunflower in transition with maize and wheat is more present in the eastern part of the region ($\gt20 \%$), while soybean and maize transition is the most prevalent in counties in central and western Vojvodina ($\gt15 \%$). Another notable observation is that some municipalities within one county practice different crop rotation patterns. A typical example is Južnobački county, where western municipalities show more arable areas covered by soybean, hence the more frequent transitions with maize. On the other side, it is observed that the eastern parts of the same county incorporate more wheat in rotating crops. The findings of this study contribute to our understanding of crop rotation practices