Numerical ragweed pollen forecasts using different source maps: a comparison for France

Abstract

One of the key input parameters for numerical pollen forecasts is the distribution of pollen sources. Generally, three different methodologies exist to assemble such distribution maps: (1) plant inventories, (2) land use data in combination with annual pollen counts, and (3) ecological modeling. We have used six exemplary maps for all of these methodologies to study their applicability and usefulness in numerical pollen forecasts. The ragweed pollen season of 2012 in France has been simulated with the numerical weather prediction model COSMO-ART using each of the distribution maps in turn. The simulated pollen concentrations were statistically compared to measured values to derive a ranking of the maps with respect to their performance. Overall, approach (2) resulted in the best correspondence between observed and simulated pollen concentrations for the year 2012. It is shown that maps resulting from ecological modeling that does not include a sophisticated estimation of the plant density have a very low predictive skill. For inventory maps and the maps based on land use data and pollen counts, the results depend very much on the observational site. The use of pollen counts to calibrate the map enhances the performance of the model considerably.

Keywords: Distribution map; Land use; Numerical simulation; Pollen; Ragweed; Ragweed inventory.

Fig. 1

The six different ragweed distributions for France. The colors denote the quantity: for map INV-# the number of plants per grid cell, for map INV-yn the numbers of populations per grid cell, for map LUPC the plant density in %, for the potential maps (POT1, POT2, and POT3) the suitability of the grid cell transformed into a plant density in %. The images show the original maps before they were calibrated using part of the NWP simulations

Fig. 2

Sites where pollen concentrations are recorded. The numbers refer to the numbering in Table S.1. The coloring denotes the regions that are introduced during the analysis of the results: sites in region A are colored in red, sites in region B are colored in blue, sites in region C are colored in green, sites in region D are colored in orange, sites in region E are colored in pink

Fig. 3

Time series of observed and simulated pollen concentrations at exemplary observational sites based on simulations using the calibrated maps. The concentrations are given as daily mean values in pollen per cubic meter of air

Fig. 4

Sums of points representing the goodness of each map (see explanations in Chapter “Statistical analysis of simulated pollen concentrations”). Good results are displayed in green colors, bad results in red colors. The scale gives the number of points that each map scores at each observational site. The background shading in grey represents the distribution map