Evaluating Sentinel-2 based phenology and productivity data for crop monitoring in Swedish agricultural fields

Sammanfattning: Sentinel-2 based data provides interesting opportunities for land monitoring and applications in agriculture. High-Resolution Vegetation Phenology and Productivity (HR-VPP) is derived from Sentinel-2 data and offers up to 13 metrics. These metrics have a high spatial resolution (10x10m) and enables the possibility for in-field agricultural monitoring. The aim of this study is to evaluate four of these metrics namely, start of season, end of season, green-up and seasonal productivity in regard to variation between fields and within. Additionally, the parameters correlation with monthly weather data on radiation, temperature and precipitation is computed. The parameters correlation with the resulting yield is also computed on both a county level and a field level. Two counties are studied, Skåne county and Västra Götaland county. Several crops with different characteristics are included in the analysis. These include winter wheat, winter rape, oats, spring barley, field bean and potato. A large variation in start of season is seen for the winter crops while the spring crops are quite stable from year to year. The highest variability between and within-fields is found in the parameters green-up and seasonal productivity. Monthly weather variables have a weak but significant correlation for most parameters and months for both spring barley and winter wheat. The correlation between yield and HR-VPP parameters indicate that seasonal productivity has a strong correlation with the resulting yield. Green-up has a stronger correlation with the yield of spring crops than winter crops. There are no large regional differences between the two counties. The higher variability in winter crops is partly accredited to less influence of agricultural practices but also to difficulties in determining the season start. Green-up’s correlation with the yield for spring crops is stronger and a possible reason could be green-up’s dependence on a correct identification of the season start. Weak correlations with weather data are expected as not one single factor can explain the spatial variability of the parameters. It is likely a combination of several weather factors and the practices of the farmer. The resulting yield has a strong correlation with seasonal productivity for all the crops in the study, which could mean that the seasonal productivity could be used in agricultural monitoring. However, the usefulness of the HR-VPP would probably be increased if finetuned for agricultural fields performed instead of being calibrated for all vegetation types.