Simulated Albedo in Needleleaf Forests is Highly Sensitive to the Treatment of Intercepted Snow: An Examination of Canopy Snow Parameterizations in the Canadian Land Surface Scheme

Abstract:

The winter albedo of boreal evergreen needleleaf forest (ENF) has been poorly simulated in climate models, with a reported range among CMIP5 models exceeding 0.25 in April, and a strong positive bias in areas with high canopy cover. Such errors have been attributed to unrealistic representation of leaf area index, snow interception and unloading, and are associated with biases in the simulated snow albedo feedback. The Canadian Atmospheric Global Climate Model has been shown to underestimate the winter albedo in boreal ENF. We present changes to the parameterization of the albedo of ENF with intercepted snow; a new relationship between interception and the fractional coverage of the canopy by snow (f_snow); and unloading based on weather conditions. The new algorithms are employed in version 3.6 of the Canadian Land Surface Scheme (CLASS) in off-line mode and the simulated daily albedo compared with observations at four ENF sites.

Default values for the visible and near-infrared albedo of snow-covered canopy were increased from 0.17 and 0.23, respectively, to 0.27 and 0.38. f_snow increased too slowly with interception, producing a damped albedo response. A new model for f_snow is based on z_I* = 3 cm, the effective depth of newly intercepted snow required to raise the canopy albedo to its maximum (corresponding to f_snow = 1). Snow unloading rates were derived from visual assessments of photographs and modeled based on relationships with meteorological variables. A model based on wind speed at the canopy top produced the best result, replacing the time-based method employed in CLASS. Model configurations were assessed based on the index of agreement, d, and the root mean squared error (RMSE). The mean d and RMSE over four sites were 0.58 and 0.058 for the default configuration of CLASS 3.6, and 0.86 and 0.038 for the best model configuration.