Optimized resolution of gridded data from European agricultural census

The European agricultural census  in 2020 collected a large number of variables from the major share of all the farms within the European Union. There are many potential applications of such a data set, from direct estimates of agricultural indicators to use as input in more complex analytical models. However, the individual responses in the data set cannot be shared directly, as they are regarded as confidential information. Instead, the data must be aggregated to a level where individual responses cannot be identified, typically NUTS regions or grid cells. As a minimum, each aggregated value must be estimated from at least 10 farms (frequency rule). Additionally, a dominance rule requires an aggregated value to be treated as confidential if the 2 largest farms are responsible for more than 85% of the value within a grid cell.

Whereas such requirements are clear, there are many methods for creating grids that respect them. The distribution of data is usually not homogeneous, and different methods have varying effects on the result.  We will outline the advantages and drawbacks of certain methods and present the most promising one that involves grid cells of varying sizes. Whereas there are some examples of this method in the past, it will be the first time it is applied on a continental scale and high-resolution data set such as the European agricultural census data.