Which Forcing Data Errors Matter Most When Modeling Seasonal Snowpack?
Raleigh, Mark S. 1 ; Lundquist, Jessica D. 2 ; Clark, Martyn P. 3
1 NCAR
2 University of Washington
3 NCAR
High quality forcing data are critical when modeling seasonal snowpack and snowmelt, but their quality is often compromised due to measurement errors or deficiencies in gridded data products (e.g., spatio-temporal interpolation, empirical parameterizations, or numerical weather model outputs). To assess the relative impact of errors in different meteorological forcings, many studies have conducted sensitivity analyses where errors (e.g., bias) are imposed on one forcing at a time and changes in model output are compared. Although straightforward, this approach only considers simplistic error structures and cannot quantify interactions in different meteorological forcing errors (i.e., it assumes a linear system). Here we employ the Sobol' method of global sensitivity analysis, which allows us to test how co-existing errors in six meteorological forcings (i.e., air temperature, precipitation, wind speed, humidity, incoming shortwave and longwave radiation) impact specific modeled snow variables (i.e., peak snow water equivalent, snowmelt rates, and snow disappearance timing). Using the Sobol' framework across a large number of realizations, we test how (1) the type (e.g., bias vs. random errors), (2) distribution (e.g., uniform vs. normal), and (3) magnitude (e.g., instrument uncertainty vs. field uncertainty) of forcing errors impact key outputs from a physically based snow model (the Utah Energy Balance). We also assess the role of climate by conducting the analysis at sites in maritime, intermountain, continental, and tundra snow zones. For all outputs considered, results show that (1) biases in forcing data are more important than random errors, (2) model outputs are minimally sensitive to the assumed error distribution, and (3) the level of uncertainty is a key factor in the relative importance of forcings. While the relative importance of forcings varied with snow variable and climate, the results broadly identify the need to reduce uncertainties in precipitation and longwave radiation.
Raleigh, M. S., J. D. Lundquist, and M. P. Clark, 2014, Exploring the impact of forcing error characteristics on physically based snow simulations within a global sensitivity analysis framework: Hydrol. Earth Syst. Sci. Discuss., 11(12), 13745-13795, doi:10.5194/hessd-11-13745-2014.