Use of distributed snow measurements to test and improve a snowmelt model for predicting the effect of forest clear-cutting

Abstract

Summary Modeling the effect of forest clear-cutting with a distributed hydrological model can be used to detect hydrologic changes as an alternative to paired-catchment studies, and also to estimate the hydrologic sensitivity of a catchment to assist in forest management decisions. To model the effect of clear-cutting in a snow dominated forested watershed, a model needs to be able to simulate effects of all of the main controls on snow accumulation and melt. However, most studies that used a distributed modeling approach relied on one or a few sites for model testing. In this study, we employ a stratified nested sample layout that was specifically designed to test distributed models, involving measurement of snow water equivalent (SWE) under forest and in clearcuts over a range of elevations and aspects. To test the ability of a model to simulate the main controls on the spatial distribution of SWE, spatial gradients of observed and simulated SWE in relation to topographic and vegetation controls are computed using regression analysis. Comparison of observed and simulated gradients helps to highlight model weaknesses. The approach is applied to evaluate the snow algorithms in the distributed hydrology soil and vegetation model (DHSVM) using data collected in Cotton Creek, a snow dominated forested watershed in south-eastern British Columbia. SWE measurements were made from 2005 to 2008, covering peak snow accumulation and snow melt. Albedo decay and canopy transmittance were found to be the two processes that DSHVM version 3.0 did not simulate well enough to predict basin average differences between forests and clearcuts properly. After replacing the internal albedo decay functions with functions obtained from snow albedo measurements and changing the canopy transmittance function in the model, DHSVM was able to reproduce the major spatial patterns derived from snow surveys. Model performance is better during winter up to the peak snow accumulation than during snow melt. Spatial patterns of peak snow accumulation in the snow-rich year 2006 can be modeled better than those after the warm winter of 2005. The influence of aspect on snow accumulation and snowmelt is underestimated by DHSVM. While we have focused specifically on DHSVM, the methods developed in this study should be generally useful for model testing purposes and important in the context of interpretation of modeling results, in particular when dealing with large spatial datasets.