**Downscaling Land Surface Parameters for Global Soil Erosion Estimation Using no Ancillary Data**

The assessment of soil erosion at the hillslope or cathment scale has been successful using models based on detailed data measured at such scales. One such model is that of Thornes (1985) which has been parameterised at the plot and field scale and employs soil erodibility, overland flow, slope and vegetation cover in order to estimate rainsplash erosion. There is, however, a need to estimate soil erosion at larger scales but at the continental and global scale only coarse resolution data are available. Investigations into the sensitivity of this soil erosion model to the scale of some of its parameters show increased scale leads to decreased predicted erosion. Investigating the effect of vegetation cover pattern and spatial resolution on predicted soil erosion showed that when cover is uniformly distributed the minimum possible erosion rate is produced and predicted erosion is stable at all the scales, however, when cover is heterogeneous erosion is reduced with an increase in pixel size. Furthermore, erosion is reduced exponentially by two orders of magnitude (from 0.03 to 0.0007 mm.month-1) as the DEM from which slope is calculated is reduced in spatial resolution from 30 arc seconds to 10 degrees.

As vegetation cover and slope scale non-linearly in the soil erosion model methods that overcome this need to be developed. To do this we have taken the approach of downscaling these parameters to the plot scale (30m) based only on coarse resolution data (as this is the only data available in many areas of the globe). To overcome the reduction in slope as resolution decreases we have developed a method that uses the fractal properties of topography to estimate slope at a specified scale using a global scale DEM. To overcome the problem of a reduction in the spatial variability of the vegetation cover with decreasing resolution we use the block variance (in an 8x8 pixel window) and a Polya probability mixture mass function to estimate the subpixel frequency distribution of vegetation cover at a specified finer resolution. These methods have been validated using multiple scale datasets and produced low RMS errors at all scales investigated. Finally, monthly soil erosion was calculated at the global scale using these techniques and validated using available data.