Distributed snowpack simulation using weather radar with an hydrologic-land surface scheme model

Abstract

Snow serves as a reservoir of water for the earth. In fact, the volume of water stored in the snowpack and the timing of the snowmelt runoff are crucial to water supply. To determine runoff volumes, empirical relationships have been developed that relate snow accumulations at point locations to downstream flows. However, modelling of the snowmelt process is required to generate peak flow estimates. Typically an hydrological model is initiated when the snowpack accumulation has reached a maximum. Another approach has been to model the snow processes through the winter, from accumulation through to melt.

Continuous hydrological modelling requires precipitation and other meteorological data that are representative of the study watershed. Point measurements from precipitation gauges are often spatially unrepresentative and there are often systematic biases, such as wind undercatch when measuring snowfall. Weather radar has been used to overcome issues related to gridding point precipitation data, however to date weather radar has not provided winter precipitation estimates for hydrological modelling. In this research, weather radar is used as the winter precipitation input to model the snowpack throughout the snow season in order to produce streamflow estimates.

The King City Radar, which is located north of Toronto and operated by the Meteorological Service of Canada (Environment Canada), covers a majority of central southwestern Ontario. The radar images used in this research were derived from the hourly average of 2 by 2 km 10 minute Constant Altitude Plan Projection Images obtained from the conventional scan. Five watersheds are located to the west of the radar: Thames, Maitland, Saugeen, Nottawasaga, and the Grand Rivers. The focus of this research is primarily the Upper Grand River for a five year study period from 1993 through 1997.

The snowpack is modelled continuously over the winter using the linked WATFLOOD and CLASS models (WAT_CLS3). The WATFLOOD model, developed at the University of Waterloo, provides the horizontal water balance and streamflow routing, while the CLASS model, developed by Environment Canada, provides the vertical water and energy balance.

To improve the streamflow and snowpack property estimates, the radar data have been adjusted to consider a local scaling phenomenon, the occurrence of mixed precipitation at warmer than freezing air temperatures, and the shape of snow particles occurring at colder temperatures. Six snow processes have been modified or added to the CLASS model, including the disaggregation of soil properties for snowcovered and snow-free areas, the occurrence of mixed precipitation near the surface, the variation of fresh snow density as a function of temperature, the variation of the maximum attainable snowpack density as a function of landcover type, snowfall canopy interception, and redistribution of blowing snow. The radar data adjustments are tested in terms of accumulation comparisons with what is considered to be accurate gauge estimates, and runoff volumes generated from WAT_CLS3 simulation. The modelling improvements are compared in terms of simulated streamflows, snowpack and shallow soil layer properties.

While it was difficult to choose an optimal radar adjustment strategy, it was shown that addressing the local scaling issue yielded accumulation underestimates and the subsequent consideration of mixed precipitation provided improved estimates. The subsequent consideration of snow particle shape did not further improve the radar accumulation estimates.

From the hydrological modelling using the different radar datasets, the raw and the scaling removed radar datasets produced, on average, the best simulated streamflow volume estimates, yet the annual variability was greater than for the temperature adjustment schemes. The consideration of particle shape yielded a slight increase in runoff volume over the consideration of mixed precipitation. Since the phase and shape of falling precipitation can vary rapidly temporally and spatially, further investigation is required to examine the applicability of hourly averages for sub-hourly physical processes such as mixed precipitation and changes in the shape of snow crystals.

Since the water balance closed within 5% over a four year period using the WAT_CLS3 model, the model translates precipitation into runoff and evapotranspiration appropriately without losing or storing excess water. Once the meteorological data were corrected, the most significant improvement to the streamflow hydrograph occurred as a result of the splitting of the soil state variables. The redistribution of snow from the bare or crop/low vegetation areas to the forest resulted in a delay in the streamflow snowmelt peak and altered the depth and snow water equivalence (SWE) of either snowpack. While the delaying of the peak is appropriate, the simple distribution technique did not represent the actual transport of snow, and in most cases overestimated movement. Variation in the maximum allowable snowpack density altered streamflow and snowpack depth, but not the SWE. The addition of mixed precipitation into CLASS provided a more realistic representation of the precipitation near the surface, however, the impact on the streamflow was small. For the variation in the fresh snow density, the actual depth of snow was only different immediately after an event. The introduction of a snowfall specific canopy interception formulation had minimal impact on the streamflow, and estimation of the composition of the forests and the canopy parameters was more important to snowpack development. Secondary results from this research included the development of a method to estimate the daytime cloud cover fraction for computing longwave radiation, the derivation of a relationship for the relative specific surface area as a function of formation temperature, and the collection of fresh snow density data in order to assess the applicability of several existing functions.

This thesis has illustrated that ground based weather radar can be used as the winter precipitation input to a hydrological model, however, the radar data should be adjusted to consider the variability in phase and shape of winter hydrometeors. One new snow process has been added and five other processes have been enhanced in the CLASS model. All of these processes should be incorporated into hydrological models.