[Proposed by Gerard Heuvelink]
The GlobalSoilMap project (which, actually, is a perfect example of setting a long-term PM challenge that we jointly work on, and guess what: it worked!) had as its aim to map the soil on a global scale at 90 m resolution. We are very close to reaching this goal. But resolution is easier reached than accuracy, and we now need to set a new aim of making global soil maps that not only satisfy the resolution requirements but that in addition meet pre-defined accuracy standards. Part of the solution may be to develop optimal sampling schemes that meet the requirement (like OSSFIM, but then for the modern DSM world). And when we get down to this, maybe at the same time we should also solve the problem of how to ensure that country borders do not show in a global soil map that is a stitch of bottom-up country-based maps.
The main driver of the map accuracy is the density of locations with known soil properties. For anyone that went once in the field for studying the soil variations, it is easy to understand that these soil variations cannot be captured with a density of one observation for several tens or hundreds of square kilometers, however good the covariates are. We should therefore densify our soil observations. This means first of all fighting for conserving, sometimes restoring and developing across the world teams of experienced soil surveyors able to collect good soil data. This means also developing optimal sampling schemes and feeding our models with less expensive soil data such as soil sensing data or qualitative soil characterizations, both legacy or participative ones. However, could these data be collected for the global level? Even, must we make a significantly more accurate Global Soil Map? (see Challenge 12)