Multiscale Parameter Regionalization
The reliable application of distributed land surface models at regional scale require, among other things, a robust parameterization scheme that avoids scale dependent issues, reduces the over-parameterization problem, and allows the transferability of model parameters across calibrated scales and locations (e.g. to ungauged basins). To effectively address these issues, we develop a general framework consisting of a multiscale parameterization technique which, in contrast to other available techniques (e.g., the Hydrologic Response Units (HRU) scheme), explicitly takes into account the sub-grid variability of basin physical characteristics (e.g., soil textural, vegetation, and geological properties), and facilitates the model runs at several spatial resolutions and locations other than those used during calibration, while keeping the model performance within reasonable ranges.
Within the MPR technique, model parameters (β0) are first mapped on to the basin physical characteristics (u0) at a much finer scale (supported by the available input datasets; 100-500 m) through a set of global calibration parameters (ϒ) and transfer functions which are based on either hydrological reasoning or empirical relationships such as pedo-transfer functions. In a subsequent step, these fine scale parameter fields (β0) are up-scaled to generate the effective parameter fields (β1) at any required coarser scale (ranges from 1 to 100 km) to support the model runs at that scale.