Comparison of land surface scheme simulations with field observations versus atmospheric model output as forcing

Abstract

The low density of meteorological stations in parts of Canada necessitates using numerical weather prediction (NWP)/assimilation output for hydrological modelling. In this study, comparisons are made of simulated land surface variables when using field observations versus NWP output as forcing for two well-instrumented sites: the mountainous and forested Marmot Creek Basin (MCRB) in the Canadian Rocky Mountains, and a prairie cropland/grassland site (Kenaston). The Canadian Land Surface Scheme 3.6 (CLASS) was used for modelling. The Global Environmental Multiscale (GEM) model with Canadian Precipitation Analysis (CaPA) was also used as forcing. There was good agreement between observed meteorology and GEM/CaPA, though some deficiencies in GEM/CaPA were identified: the effects of subgrid topography on incoming radiation and wind speed were not accounted for at MCRB, and CaPA did not capture some convective rainfall events at Kenaston. CLASS simulations using both sets of forcing showed difficulties in simulating snow depth, soil moisture and evapotranspiration; certain difficulties were linked to GEM/CaPA deficiencies and/or CLASS. Both sets of forcing tended to overestimate the duration of snow cover at MCRB, but during different years. With GEM/CaPA as forcing, CLASS overestimated the duration of frozen soils. The GEM/CaPA precipitation difficulties at Kenaston degraded soil moisture simulations.

A strategy to represent impacts of subgrid-scale topography on snow evolution in the Canadian Land Surface Scheme

“…For example, the QRRC observed maximum daily snow depth of 94 cm occurred on 15 March 2009, while CLASS simulated peak daily snow depths of 86, 91 and 92 cm for T rs = 0, 2 and 6 °C, respectively, on 2 February 2009. Thus, as recommended by Brown and others (2006), the original (unadjusted) precipitation data with T rs = 0 °C are used henceforth to run CLASS, and it is assumed that the model underestimates the snow depth and simulates an early melt at the QRRC (as reported for other locations; see Pomeroy and others, 1998; MacDonald and others, 2016).…”