Inference of parameters for a global hydrological model by applying
Approximate Bayesian Computation: Identifiability of climate-based
parameters
Abstract
The calibration of global hydrological models has been attempted for
over two decades, but an effective and generic calibration method has
not been proposed. In this study, we investigated the application of
Approximate Bayesian Computation (ABC) to calibrate the H08 global
hydrological model by running global simulations with 5000 randomly
generated sets of four sensitive parameters. This yielded satisfactory
results for 777 gauged watersheds, indicating that ABC can be used to
optimize H08 parameters to calibrate individual watersheds. We tested
the identifiability of the parameters to yield satisfactory
representations of hydrological functions based on Köppen’s climate
classification (“climate-based” calibrations hereafter) We aggregated
5000 simulation results per catchment based on the 11 Köppen climate
classes, then selected the parameters that exceeded the Nash–Sutcliffe
efficiency (NSE) scores predefined by the acceptance ratio for each
climate class. Our results indicate that the number of stations showing
satisfactory (NSE > 0.0) and good (NSE>0.5)
performances were 480 and 234 (61.7% and 30.1% of total stations,
respectively), demonstrating the effectiveness of climate-based
calibration. We also showed that the climate-based parameters
outperformed the default and global parameters in terms of
representativeness (global-scale differences of hydrological properties
among climate classes) and robustness (consistency in yielding
satisfactory results for watersheds in the same climate class). The
identified parameters for 11 Köppen climate classes showed consistency
with the physical interpretation of soil formation and efficiencies in
vapor transfer with a wide variety of vegetation types, corroborating
the strong influence of climate on hydrological properties.