[Proposed by Gerard Heuvelink]
We are still struggling with the concept of scale. We too often make it vague and obscure because we use poor definitions. In the meantime, there are burning issues. For instance, soil physicists model water infiltration using models based on the Richards equation. Such models are meant for the pedon scale but are also applied (by ‘scaling’ the model parameters) at the scale of regional Land Surface Models. Should not the model structure change as well when upscaling? If yes, how? How does a non-linear partial differential equation interact with spatial variation? Similar issues arise when modelling soil-landscape evolution. We also haven’t really solved the problem of how to statistically validate a model that makes predictions at a support much greater than the support of validation measurements.
Another very relevant challenge. I let the soil physicists react about the right model at the right scale. Our job, as Digital Soil Mappers, is to provide estimations of soil properties and of the associated uncertainties for any required geographical support whereas we stick at working at the horizon or pedon level. My experience is that most of the users do not ask for such a fine support for making decisions. Delivering soil prediction at larger support is not only relevant for users but also would avoid discouraging the users with the large uncertainties that are often obtained when working at the pedon level. Spatial aggregation techniques exist, including those provided by geostatistics. As often, the main limiting factor for applying these techniques is the input data, in particular a minimum knowledge of the short-range soil spatial variabilities. Ad-hoc spatial sampling, collection of legacy proximal and high-resolution remote sensing data, or meta-analysis of the literature should be explored as possible solutions for obtaining this information.