How might we use erosion models to meet future policy challenges?

Notable recent growth in the number of publications describing RUSLE/GIS-based soil erosion model studies indicates a previously-unmet policy-driven need for maps of water erosion rate covering nationally-sized (or larger) areas, at a field-scale (or smaller) resolution. Other current erosion modelling approaches have not been able to fulfil this need. What is necessary for future soil erosion models to do so?

Assume a hypothetical future farmer-focused policy-driven study, requiring a modelled estimate of erosion rate for ‘tiles’, each 10 m², to cover 10⁵ km² (1/3 of Germany). Thus needing 10¹⁰ tiles.

RUSLE calculates long-term average soil loss. Running once for each tile would give the required spatial coverage, but no information regarding temporal change. Future anthropogenically-driven climate change, and future climate-change-driven land-use change, will have a major global impact on soil loss. The hypothetical study must consider these impacts. Additionally, the RUSLE model is empirical. Confidence in statistical relationships decreases when the relationship is extrapolated i.e. used with values outside the range of values employed to derive it. Future global change will produce surprises, needing RUSLE input values outside the values used to derive the model. A temporally dynamic, and not wholly empirical, erosion model will therefore be necessary for the hypothesised study.

Other anthropogenic choices are important. Especially in agricultural areas, there will be ‘pinch points’: locations where a minor change in some attribute has a major downslope impact. Where sediment and flow encounters a barrier (e.g. a hedge), the degree of barrier permeability will strongly influence subsequent flow. Further, the permeability of the barrier may depend on past events e.g. previous accumulation of of waterborne trash.

Our study will need to consider combinations of climate and land-use change scenarios: parsimoniously, assume 10 model runs per tile. It is less obvious how we would cope with ‘pinch point’ problems, but assume (again, parsimoniously) the need for 10 more scenarios. These two would bring the number of model runs up to 10¹²: an impressively large number.

For this daunting challenge, what might be the next steps?

So far, we assume data availability for every tile. Whilst not a problem for RUSLE, this would be a major headache for physically-based models. Thus there is a need to (a) develop erosion models of reduced complexity (but not wholly empirical), with correspondingly reduced data needs; and possibly (b) an approach to determining the ‘importance’ of each tile.

A potentially fruitful approach would adopt a systems-based approach, by explicitly considering the scaling relationships (often fractal, and with power-law frequency-magnitude) which result from multiple process feedback loops operating within the erosional system. If such a scaling relationship were derived between a plot-scale process-focused model (e.g. RillGrow) and another erosion model operating on a larger area with coarser resolution and reduced data requirement; then this relationship could be used to interpolate between scales.

Our study might then involve running multiple scenarios of the coarser-scale model for the whole area, using the scaling relationship to identify potentially problematic tiles, and running the plot-scale model only for these locations/situations.