Factors Influencing Snow Model Performance in Boreal Forests - Results from the ESM-SnowMIP Forest Site Simulations

<p>The site level component of the Earth System Model – Snow Model Intercomparison Project has 28 participating model variants. We summarize model performance at the Boreal Ecosystem Research and Monitoring Sites (BERMS) Old Aspen (OAS), Old Black Spruce (OBS) and Old Jack Pine (OJP) forests in Saskatchewan.</p><p>Many CMIP5 models have been previously shown to overestimate the winter albedo in the boreal forest due to errors in plant functional type (PFT) and leaf area index (LAI). In this project provided values for PFT and LAI were not implemented in a few models, but many models show a positive albedo bias in excess of 0.1 and some show a much larger positive bias. A larger positive albedo bias at OAS by some models suggests that snow masking by leafless trees requires attention. Average albedo bias from these off-line simulations, which lack atmospheric feedbacks, is not strongly related to bias in snowpack properties or the treatment or lack thereof of intercepted snow.</p><p>About half the models simulated snow water equivalent (SWE) with a RMSE smaller than the standard deviation of the observations. Snow depth was simulated slightly worse and only three models met this standard with respect to snowpack density. SWE was underestimated by just over half the models but the density of these sheltered snowpacks was overestimated by most models, resulting in snowpack depth being underestimated by an average 0.1 m. Models with multiple simplified surface parameterizations tend to show the greatest underestimation of SWE and depth and overestimation of density.</p><p>Biases in above-canopy radiative, snow surface and bulk snowpack temperatures are not consistent with respect to size and sign; many models show a combination of positive and negative biases. Radiative and snowpack surface temperatures are associated with trends in turbulent heat fluxes. Models with multiple simplified surface parameterizations (e.g. large or fixed density or thermal conductivity values, a composite snowpack, no organic soil) show more negative soil temperature biases and appear to be associated with a colder snowpack, but unfortunately, bulk snowpack temperature was not reported for many such models. Negative SWE and depth biases are associated with colder winter soil temperatures and shorter snow seasons. Most models simulate snow thermal conductivity with one of many relationships with density. Soil temperature bias is highly sensitive to the choice of snow thermal conductivity parameterization.</p><p>Models with many snow layers tend to show smaller errors in snowpack properties and are less likely to show cold biases in the snowpack and soil compared with composite or single layer models. However, as found in previous SnowMIPs, some single-layer models occupy the same bias range as multi-layer models. Models employing a multi-layer snowpack tend not to employ multiple “simplified parameterizations” as described above whereas the models with a single snow layer employ surface parameterizations with a range of sophistication.</p>

Predictability of Precipitation in Complex Terrain using the WRF Model with Varying Physics Parameterizations

<p>Canada’s west coast topography plays a crucial role for the local precipitation patterns, which are often shaped by orographic lifting on one side of the mountains, and rain shadows on the other side. The hydroelectric infrastructure in southwest British Columbia (BC) relies heavily on the abundant rainfall of the wet season, but long lasting and heavy precipitation can cause local flooding and make reliable precipitation forecasts crucial for resource management, risk assessment, and disaster mitigation.</p><p>This research evaluates hourly precipitation forecasts from the Weather Research and Forecasting (WRF) model over the complex terrain of southwest BC. The model data includes a full year of daily runs across three nested domains (27-9-3 km). A selection of different parameterizations is systematically varied, including microphysics, cumulus, turbulence, and land-surface parameterizations. The resulting over 100 model configurations are evaluated with observations from ground-based quality-controlled precipitation gauges. The individual model skill of the precipitation forecasts is assessed with respect to different accumulation windows, forecast horizons, grid resolutions, and precipitation intensities. Furthermore, the ensemble mean and spread provide insight to the general error growth for precipitation forecasts in WRF.</p><p>Cumulus and microphysics parameterizations together determine the total precipitation in numerical weather prediction models and this study confirms the expectation that the combination of those physics parameterizations is most decisive for the precipitation forecasts. However, the boundary-layer and land-surface parameterizations have a secondary effect on precipitation skill. The verification shows that the WSM5 microphysics parameterization yields surprisingly competitive verification scores when compared to more sophisticated and computationally expensive parameterizations. Although, the scale-aware Grell-Freitas cumulus parameterization performs better for summer-time convective precipitation, the conventional Kain-Fritsch parameterization performs better for winter-time frontal precipitation, which contributes to the majority of the annual rainfall in southwest BC.</p><p>Throughout a 3-day forecast horizon mean absolute errors are observed to grow by ~5% per forecast day. Furthermore, this study indicates that coarser resolutions suffer from larger total biases and larger random error components, however, they have slightly higher correlation coefficients. The mid-size 9-km domain yields the highest relative hit rate for significant and extreme precipitation. Verification metrics improve exponentially with longer accumulation windows: On one side, hourly precipitation values are highly prone to double-penalty issues (where a timing error can, for example, result in an over-forecast error in one hour and an under-forecast in a subsequent hour); on the other side, extended accumulation windows can compensate for timing errors, but lose information about short-term rain intensities.</p>