Effect of three pillars on hydrological model calibration: data length,
spin-up period and spatial model resolution
Abstract
In general, calibration of a hydrologic model is essential to better
simulate the basin processes and behaviour by fitting the model
simulated fluxes to observed fluxes. A major challenge in the
calibration process is to choose the appropriate length of the observed
data series and spatio-temporal resolution of the model schematization.
We present a multi-case calibration approach for determining three
pillars of an optimum hydrological model configuration: calibration data
length, spin-up period and spatial resolution of the hydrological model.
The approach is evaluated for the Moselle River basin using calibration
and validation results from the spatially distributed meso-scale
Hydrological Model (mHM) for 105 different cases representing the
combinations of three calibration data lengths, seven spin-up periods
and five spatial model resolutions. A metaheuristic global optimization
method, i.e. Dynamically Dimensioned Search (DDS) algorithm, and a
well-known hydrological performance metric, i.e. Nash Sutcliffe
Efficiency (NSE), are utilized for each of the 105 calibration cases.
The results show that a 10-year calibration data length, 2-year spin-up
period and a 4 km model resolution are appropriate for the Moselle basin
to reduce the computational burden. Analyzing the combined effects
further allowed us to understand the interactions of these three usually
overlooked pillars in hydrological modeling.