1. Abstract
2. VIC-3L Hydrologic Model
3. Model Application
4. Point Model Tests
5. Spatial Model Calibration, LSA-NC
6. Frozen Soil Effects
7. Conclusions

Frozen soils play an important role in cold regions hydrology by limiting infiltration during winter and spring melt events. Currently most large-scale hydrologic models ignore the effects of frozen soils. The Variable Infiltration Capacity (VIC) model, is a macroscale water and energy balance model that has been used to simulate streamflow in many large river basins including the Columbia, Ohio (GCIP LSA-E) and Arkansas-Red Rivers (GCIP LSA-SW). As part of the GCIP project, VIC has been modified for better performance in cold regions by adding a frozen soil algorithm and improving the snow accumulation and ablation routines. The revised model has been applied to the Upper Mississippi Basin (LSA-NC), to investigate the role of snow and frozen soils on winter and spring hydrologic processes.

Field observations at the University of Minnesota's Rosemount Agricultural Experiment Station show that spring melt is a complex spatial process even at the field scale. Melt water tends to accumulate in depressions within each field, where it may be unable to drain for several days due to the underlying frozen soils. Thaw of the soils is patchy, usually occurring after the ground surface is exposed by melting snow, but it may occur suddenly under ponds of melt water, allowing them to drain quickly through the opening.

Tests of the VIC model frozen soil algorithm shows that point model results compare favorably against data collected at Rosemount. Spatial tests have been conducted on several subbasins within the Upper Mississippi including: the Minnesota River (41,958 km²), the Mississippi above Anoka, MN (49,468 km²), and the Illinois River (69,266 km²). Each basin was run at 1/2° resolution, with meteorological data obtained from the NCDC hourly surface airways, and daily precipitation data sets. Soil parameters were derived from the Continental United States soil data set, prepared by D. A. Miller and R. A. White at Pennsylvania State University. Simulated outflow from the basins was calibrated against USGS daily flows. Comparisons of model simulations with and without the frozen soil algorithm show that the timing of melt flow, as well as peak flows change with the addition of frozen soils.

The Variable Infiltration Capacity - Three Layer (VIC-3L) Model is a macroscale hydrologic model that solves a full energy and water balance. It has been used previously to model the hydrology of large river basins such as the Ohio (LSA-E) and the Arkansas-Red (LSA-SW), as well as at a global scale.

Figure 1 shows the basic components of the VIC-3L model:

• Infiltration and runoff are controlled using a Variable Infiltration Curve (VIC);
• Baseflow is controlled by a non-linear function for high moisture content;
• Three soil moisture layers; and
• Variable vegetation coverage (1 through N-1) plus bare soil (N).

Figure 1: Schematic of the VIC-3L macroscale hydrologic model.

In order to improve the model's performance in cold regions, the following additions and improvements were undertaken:

• Improved snow accumulation and melt model;
• Added frozen soil routines.

Recent improvements in the snow accumulation and ablation model (Figure 2) include the following:

• resolves stability problems the older model had with thin snow packs,
• added interception of snow by the canopy,
• improved the calculation of aerodynamic resistance with a canopy by using an updated wind profile, and
• added ground – snow heat flux.

Figure 2: Schematic of the new snow model included in the VIC-3L.

Frozen soils are important in cold regions because they can decrease infiltration during melt events, thereby increasing the runoff, and increasing the chances of flooding downstream.

The improved VIC model determines the depth of frost penetration by computing heat fluxes at a number of nodes through the soil column (Figure 3). The resulting soil temperatures are used to compute the ice content of the VIC layers. The ice content affects infiltration, moisture fluxes in the soil column, the thermal conductivity, and the volumetric heat capacity. Thermal conductivity and volumetric heat capacity are then used to compute the thermal fluxes for the next time step.

Figure 3: Schematic of the frozen soil model incorporated into the VIC-3L Model.

The VIC-3L model with frozen soils was tested in point mode as well as spatially distributed applications in the Upper Mississippi River basin. The Upper Mississippi River basin drains an area of 444,183 km², before joining with the Missouri River in St Louis, MO. VIC models the basin as 226 1/2° x1/2° grid cells. Inputs needed by the VIC model include:

• Air temperature,
• Precipitation,
• Soil properties, and
• Vegetation parameters.

Atmospheric forcing data was taken from NCDC daily and hourly station data. Data was gridded to the 1/2° x1/2° grid resolution used by VIC. Figure 4 shows that the LSA-NC has a strong north-south gradient in mean annual air temperature, but an absence of a strong precipitation gradient.

Figure 4: Basin distribution of (a) mean annual air temperature, and (b) average annual precipitation.

Land cover types were obtained from the North America land cover characteristics database compiled as part of the Global Land Cover Characterization (GLCC) project. Coverage types were reduced from 20 SiB classifications to 5 for use with VIC:

• deciduous trees,
• coniferous trees,
• grasslands,
• agriculture, and
• bare soil

Figure 5 shows the fractional grid cell coverage of trees (both deciduous and coniferous), and agriculture within the Upper Mississippi Basin.

Figure 5: Basin distribution of fractional grid cell coverage by (a) trees, and (b) agriculture.

VIC uses many parameters to control the hydrologic processes it represents. Many of these parameters can be estimated from commonly measured soil properties. Soil parameters for the Upper Mississippi were taken from a database maintained by the Earth System Science Center at Pennsylvania State University. Figure 6 shows the distribution of two of these parameters, percent sand and percent clay.

Figure 6: Basin distribution of the measured soil properties (a) percent sand, and (b) percent clay.

VIC solves each grid cell in the basin independently, the runoff and baseflow results are then input to an external routing model. This model uses a specified routing network to determine the discharge at the basin outlet, and at user defined locations within the basin. These discharge simulations can then be compared against records from gauging stations. The routing network for the Upper Mississippi is shown in Figure 7.

Figure 7: Routing network for the LSA-NC. Colored nodes indicate the locations of calibration subbasins.

Point model tests of the frozen soil model were conducted using data collected at the University of Minnesota Rosemount Agriculture Experiment Station. Figure 8 compares the VIC simulation with observations for the winter of 1994-95.

Figure 8: Point model comparison of model simulation versus observations for the Rosemount, MN field site.

Snow cover serves as an insulation layer, so changes in frost depth are slower under thicker snow packs. Therefore the better simulations of snow cover will yield better simulations of frozen soils.

Due to the size of the Upper Mississippi it was broken into five subbasins to help with the calibration. The subbasins shown in Figure 7 are:

• the Mississippi River Above Anoka, MN,
• the Minnesota River,
• the Wisconsin River,
• the Iowa River, and
• the Illinois River.

Figure 9 shows weekly outflow from the gauged catchments of the Illinois the Minnesota, and the Mississippi above Anoka. These subbasins were selected for further analysis because they have the best calibrations at this time.

Figure 9: Observed versus simulated runoff for three of the calibration subbasins. Black lines indicate observations, red dashed line indicate simulations. Simulations run without frozen soil model.

The presence of frozen soils changes the timing of outflow peaks (Figure 10), as well as their volume. Baseflow from the frozen soil model is lower during the winter, since less of the mid-winter melt infiltrates. During spring melt the frozen soil model produces narrower peaks, since more of the melt water runs off directly and does not contribute much to baseflow. The frozen soil model has less of an effect in the Illinois since that basin is warmer and receives less snow.

Figure 10: Simulated outflow for three of the calibration subbasins. The blue line show model results for the frozen soil model, the red dashed lines show results for the model without frozen soils.

Figure 11 is a scatter plot of peaks over threshold for the Mississippi above Anoka and the Illinois. Both rivers show a couple of increased peak flow events with the frozen soil model, but also some events where the frozen soil model produces smaller peaks. Most of the smaller peaks are due to the timing changes, where the frozen soils model produces higher runoff over several days of melt, but does not get the baseflow response for the last days of melt, or post-melt storm events.

Figure 11: Scatter plot of peaks over threshold for two calibration subbasins (a) the Illinois, and (b) the Mississippi above Anoka, MN. Vertical axis is for the model with frozen soils, while the horizontal axis is for the model without frozen soils.

Frozen soils change the response of basins during spring melt. In some cases causing higher peaks flows, and in other cases changing the timing of events. Frozen soils are clearly important to cold regions, as is better representation of the snow pack. Future work will look at verifying the simulated snow cover, and frozen soil extent by using in situ and remote sensing measurements.

Figure 1: VIC-3L Schematic

Figure 2: VIC Snow Model Schematic

Figure 3: VIC Frozen Soil Model Schematic

Figure 7: LSA-NC Routing Network and Calibration Subbasins

Figure 9: Calibration Runs – observations vs. simulations

Figure 10: with Frozen Soils vs. without Frozen Soils