Integration of in situ and satellite data for top-down mapping of Ambrosia infection level.
A new approach of integration of remote sensing data with in situ pollen measurements is developed to explore the effect of changing land use on the local pollen records variability. It was conducted in a predominantly agricultural region of Serbia, with the focus on Ambrosia, an invasive weed that is the source of highly allergenic pollen. The land use characteristics were extracted from the Corine Land Cover and more precise crop classification maps for years 2013-2017 created using machine learning from the available satellite (RapidEye, Landsat, Sentinel) and ground truth data. Airborne pollen was collected at five locations for the same period. To integrate in situ and products derived from satellite data we defined catchment areas surrounding pollen measurements stations. Different shapes of assumed catchment areas equivalent to 5 km, 10 km and 30 km radius were tested: circular, wind rose and footprint from SILAM (System for Integrated modeLling of Atmospheric coMposition). The simple fixed circle, used as the rule of thumb in the literature, is a reasonable approximation of the representativeness of the aerobiological data. A gridded Ambrosia distribution and abundance map over Vojvodina was produced by using top-down approach that combines distribution of suitable habitats and airborne pollen concentrations. The results confirmed that variation in the agriculture-associated land use area explains notable amount of variability in the amount of Ambrosia airborne pollen. Detailed crop classification inferred from satellite data revealed the strongest relationship between pollen and variation in areas under soya bean and sugar beet. Maps of Ambrosia infection levels, based on distribution of sugar beet and soya bean fields, instead of total arable land, reveal its additional spatial variability in Vojvodina region.