Evaluation of Snow Depth and Soil Temperatures Predicted by the Hydro–Thermodynamic Soil–Vegetation Scheme Coupled with the Fifth-Generation Pennsylvania State University–NCAR Mesoscale Model

Balachandrudu Narapusetty Geophysical Institute, University of Alaska Fairbanks, Fairbanks, Alaska

Search for other papers by Balachandrudu Narapusetty in
Current site
Google Scholar
PubMed
Close
and
Nicole Mölders Geophysical Institute, University of Alaska Fairbanks, Fairbanks, Alaska

Search for other papers by Nicole Mölders in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

The Hydro–Thermodynamic Soil–Vegetation Scheme (HTSVS) coupled in a two-way mode with the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (NCAR) Mesoscale Meteorological Model (MM5) is evaluated for a typical snowmelt episode in the Baltic region by means of observations at 25 soil temperature, 355 snow-depth, and 344 precipitation sites that have, in total, 1000, 1775, and 1720 measurements, respectively. The performance with respect to predicted near-surface meteorological fields is evaluated using reanalysis data. Snow depth depends on snow metamorphism, sublimation, and snowfall. Because in the coupled model these processes are affected by the predicted surface radiation fluxes and cloud and precipitation processes, sensitivity studies are performed with two different cloud microphysical schemes and/or radiation schemes. Skill scores are calculated as a quality measure for the coupled model’s performance for a typical forecast range of 120 h for a typical spring (snowmelt) weather situation in the Baltic region. Discrepancies between predicted and observed snow-depth changes relate to the coupling. Enhanced water supply to the atmosphere, which results from water that was assumed to be open in MM5 but was actually ice covered in nature, finally leads to an overestimation of snowfall (input to HTSVS) and changes in snow depth (output). The resolution-dependent discrepancies between the terrain height in the model and real world also lead to snowfall where none occurred. For heavy snowfall the performance of the coupled model with respect to predicted snow-depth changes becomes nearly independent of the choice of the cloud microphysical and radiation schemes. As compared with observed changes in snow depth, the coupled model simulation using the Schultz scheme in conjunction with the radiation scheme from the Community Climate Model, version 2, (CCM2) predicts snow-depth changes of less than 2.5 mm considerably better than the other combinations that were tested. For thick snowpacks, the accuracy of the snow-depth decrease resulting from metamorphism strongly depends on the initial value of snow density. The coupled model acceptably captures the soil temperature diurnal cycles, the observed soil temperature increase with time, and the soil temperature behavior with depth. In general, discrepancies between simulated and observed soil temperatures decrease with soil depth. Simulations performed with the so-called CLOUD radiation scheme capture soil temperature minima and maxima better than do simulations performed with the CCM2 scheme.

Corresponding author address: N. Mölders, Geophysical Institute, University of Alaska Fairbanks, 903 Koyukuk Dr., Fairbanks, AK 99775-7320. molders@gi.alaska.edu

Abstract

The Hydro–Thermodynamic Soil–Vegetation Scheme (HTSVS) coupled in a two-way mode with the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (NCAR) Mesoscale Meteorological Model (MM5) is evaluated for a typical snowmelt episode in the Baltic region by means of observations at 25 soil temperature, 355 snow-depth, and 344 precipitation sites that have, in total, 1000, 1775, and 1720 measurements, respectively. The performance with respect to predicted near-surface meteorological fields is evaluated using reanalysis data. Snow depth depends on snow metamorphism, sublimation, and snowfall. Because in the coupled model these processes are affected by the predicted surface radiation fluxes and cloud and precipitation processes, sensitivity studies are performed with two different cloud microphysical schemes and/or radiation schemes. Skill scores are calculated as a quality measure for the coupled model’s performance for a typical forecast range of 120 h for a typical spring (snowmelt) weather situation in the Baltic region. Discrepancies between predicted and observed snow-depth changes relate to the coupling. Enhanced water supply to the atmosphere, which results from water that was assumed to be open in MM5 but was actually ice covered in nature, finally leads to an overestimation of snowfall (input to HTSVS) and changes in snow depth (output). The resolution-dependent discrepancies between the terrain height in the model and real world also lead to snowfall where none occurred. For heavy snowfall the performance of the coupled model with respect to predicted snow-depth changes becomes nearly independent of the choice of the cloud microphysical and radiation schemes. As compared with observed changes in snow depth, the coupled model simulation using the Schultz scheme in conjunction with the radiation scheme from the Community Climate Model, version 2, (CCM2) predicts snow-depth changes of less than 2.5 mm considerably better than the other combinations that were tested. For thick snowpacks, the accuracy of the snow-depth decrease resulting from metamorphism strongly depends on the initial value of snow density. The coupled model acceptably captures the soil temperature diurnal cycles, the observed soil temperature increase with time, and the soil temperature behavior with depth. In general, discrepancies between simulated and observed soil temperatures decrease with soil depth. Simulations performed with the so-called CLOUD radiation scheme capture soil temperature minima and maxima better than do simulations performed with the CCM2 scheme.

Corresponding author address: N. Mölders, Geophysical Institute, University of Alaska Fairbanks, 903 Koyukuk Dr., Fairbanks, AK 99775-7320. molders@gi.alaska.edu

Introduction

In numerical weather prediction (NWP) and climate models, typically a land surface model (LSM) determines the temperature and moisture states and heat and moisture fluxes within the soil, canopy, and/or snow as well as the fluxes of sensible and latent heat and momentum at the land–atmosphere interface (e.g., Sellers et al. 1986; Bonan 1994; Robock et al. 1997). These LSMs have been developed using the best of our scientific knowledge, and great effort has been spent in evaluating them (e.g., Henderson-Sellers et al. 1993, 1995; Chen et al. 1997; Yang et al. 1997; Wood et al. 1998; Schlosser et al. 2000; Luo et al. 2003; Mölders et al. 2003a, b). Usually these evaluations have been performed offline; that is, the LSMs were driven by observed meteorological data. These studies evidenced that LSMs usually work appropriately when run in an offline mode. The major propagating errors are associated with measurement errors, uncertainty in initial values and empirical parameters, and the assumptions made in parameterization, boundary conditions, and numerical discretization.

If an LSM is coupled with an NWP or climate model, the simulated meteorological forcing conditions will be an error source. The forcing can be wrong because of, among other things, erroneous or unknown initial conditions, boundary conditions, discretization, grid resolution, inaccurate model assumptions, or choice of inappropriate empirical parameters (e.g., Anthes et al. 1989; Slater et al. 1998; Zhong et al. 2005). It is obvious that these false forcing conditions may propagate to incorrect prediction of the state variables and fluxes by the LSM, again affecting the forecasts of meteorological quantities. Furthermore, it is typical that no site-specific empirical parameters are available at the resolution of NWP or climate models, resulting in erroneous predictions of soil and surface conditions.

For the reasons outlined above, the performance of an LSM decreases as compared with offline simulation when it is coupled to another model. Based on 5 months of area-averaged observations, Chen et al. (1996), for instance, found that in an offline mode the Oregon State University LSM (OSULSM) correctly captures the diurnal cycle in surface heat flux and that differences in simulated and observed skin temperature could be as high as 5 K. Chen and Dudhia (2001) reported that a modified version of OSULSM coupled with the fifth-generation Pennsylvania State University– National Center for Atmospheric Research (NCAR) Mesoscale Model (MM5; Dudhia 1993; Grell et al. 1994) made an excellent prediction of latent heat flux and predicted sensible heat flux to within 100 W m−2 accuracy. Visual comparison of Chen et al.’s (1996) figures showing the offline performance for 4, 5, and 6 June 1981 with the results from the coupled simulation for the same days shown in Chen and Dudhia (2001) indicates a decrease in performance for the coupled model of about 3 K for temperature and 85 W m−2 for sensible heat fluxes. This example demonstrates that the additional “error” sources introduced by running an LSM and NWP model in a coupled mode requires a reevaluation of the performance of any LSM when introduced into a coupled model.

This task is addressed in this article for the Hydro–Thermodynamic Soil–Vegetation Scheme (HTSVS; Kramm et al. 1996; Mölders et al. 2003a) that recently was coupled in a two-way mode with MM5 (Mölders 2000; Mölders and Walsh 2004). In offline evaluation studies, HTSVS predicted soil temperature to within ±2.5 K in the short term (e.g., Kramm 1995), and the water supply to the atmosphere and groundwater recharge to within 15% accuracy as well as daily mean soil temperatures to within 1–2 K in the long term (2050-day simulation; Mölders et al. 2003a, b). Note that besides being used in MM5 for permafrost and snow studies (e.g., Mölders and Walsh 2004) HTSVS has been used in one-dimensional chemistry models for air-quality studies (e.g., Kramm et al. 1996) and in the Geesthacht Simulation Model of the Atmosphere (GESIMA; Kapitza and Eppel 1992; Eppel et al. 1995) as an alternative LSM (e.g., Mölders and Rühaak 2002) to the force–restore method including vegetation processes (Claussen, 1988) usually applied. The aim of our study is to evaluate HTSVS coupled to MM5 using soil temperature and snow-depth data obtained in the Baltic Sea Experiment (BALTEX; e.g., Raschke et al. 1998) for a 5-day spring episode.

Experimental design

Brief description of HTSVS

HTSVS (Kramm et al. 1996; Mölders et al. 2003a; Mölders and Walsh 2004) consists of a one-layer canopy model and a multilayer snow and soil model. The canopy model describes the exchanges of momentum, heat, and moisture at the vegetation–soil–atmosphere interface. It considers microscale heterogeneity by a mixture approach; that is, a grid cell can be partly covered by vegetation (e.g., Deardorff 1978; Kramm et al. 1996).

The multilayer snow model follows Fröhlich and Mölders (2002) with the modifications and additions required for the coupling to MM5 described in Mölders and Walsh (2004). Snow depth increases by snowfall, and it decreases by sublimation, outflow of meltwater, windbreak, compaction, settling, meltwater percolation, and freezing. The processes contributing to a snow-depth decrease are denoted snow metamorphism. Like in many other LSMs (e.g., Verseghy 1991; Loth and Graf 1998; Bonan et al. 2002) a minimum thickness (2 mm) is assumed for each snow model layer to consider snow metamorphism, except melting, that may occur independent of thickness.

The soil model makes use of the principles of linear thermodynamics of irreversible processes (e.g., de Groot 1951; Prigogine 1951), including the Richards equation (e.g., Philip 1957; Philip and de Vries 1957; de Vries 1958; Kramm 1995; Kramm et al. 1996), and takes into account freezing and thawing of soil (e.g., Mölders et al. 2003a). It also considers the Ludwig–Soret effect (i.e., a temperature gradient contributes to the water flux and changes the soil volumetric water content) and the Dufour effect (i.e., a moisture gradient contributes to the heat flux and alters soil temperature). These so-called cross effects have been found to be important during freezing/thawing and snowmelt (e.g., Mölders and Walsh 2004) and when chemicals are involved. For a detailed description of the parameterized algebraic equation sets and the differential equations governing the physical processes in HTSVS for soil temperature, volumetric water, and ice content, including phase transition processes and water extraction by roots, see Mölders et al. (2003a).

The atmospheric model and simulations

MM5 can be run with various physical parameterizations (Grell et al. 1994). The parameterizations of radiation and cloud microphysical processes can most strongly affect the coupled MM5–HTSVS performance. To elaborate sources of discrepancies between simulated and observed snow depth that are caused by the radiation scheme, we perform simulations using alternatively the radiation scheme from the Community Climate Model, version 2, (CCM2) with a δ-Eddington method (Briegleb 1992) and the so-called CLOUD radiation scheme (Stephens 1978, 1984; Garand 1983). For simplicity, we denote these radiation schemes as the CCM2 and CLOUD schemes, the names under which they are known in the MM5 community.

To assess the impact of the cloud microphysical scheme on snow-depth increase we perform simulations using the modified version of Reisner et al.’s (1998) mixed-phase scheme with graupel (Thompson et al. 2004) and, alternatively, Schultz’s (1995) scheme.

In the following, the simulations performed with the Schultz and CCM2 schemes or, alternatively, the Reisner and CCM2 schemes and their results are named SC and RC, respectively. Simulations and their results obtained with the Schultz and CLOUD schemes or, alternatively, the Reisner and CLOUD schemes are called SL and RL, respectively (Table 1).

Because the grid spacing chosen in our simulations (35 km) requires the use of a cumulus convection scheme, we consider Grell et al.’s (1991) cumulus scheme. Boundary layer physics is considered in accordance with Hong and Pan (1996).

Model domain

The model domain encompasses the BALTEX region from the surface to 100 hPa with a center at 60°N, 17.5°E (Fig. 1). The horizontal spacing is 35 km, with 57 × 45 grid points. Five soil layers are spaced by the same logarithmic increment so that central differences can be used in solving the coupled soil equations by a generalized Crank–Nicholson scheme in combination with a Gauβ–Seidel technique. Because the forecast range of MM5 is generally restricted to several days to a week, soil temperature, volumetric water, and ice content are held constant throughout a simulation at the lower boundary of the soil model for simplicity. In the case of snow, there are five snow layers of equal thickness that get redivided whenever snow depth changes. The simulations are performed with a time step of 105 s.

Synoptic situation

The episode chosen encompasses from 0000 UT 21 April to 0000 UT 26 April 2000 and is a typical snowmelt weather situation in the Baltic region in spring. On 21 April the eastern part of the region was influenced by a moderate high pressure system while the western part was governed by a trough associated with intensive cyclonic activity, with the center in the northern Atlantic Ocean, west of the British Islands. On 22 April, most of the region was under high pressure influence. In the northern part, cyclonic activity cut off from the trough, developed, and moved eastward. On 23 April, Estonia, Finland, and Latvia were under weak high pressure influence, whereas Norway, Sweden, and Russia were affected by a low pressure system. The latter lost its energy while the newly formed system strengthened as a result of advection of Arctic air. On 24 April, a low pressure system moved into northern Finland from the east. Germany and Norway were governed by a northward-moving low pressure system while the weakened high pressure system remained over Belarus, Lithuania, and northern Poland. On 25 April, the cyclonic activities joined and affected Sweden and Norway while the high pressure system intensified and moved eastward.

On 21 April air temperatures ranged from −3° to 0°C in Norway, Sweden, and Finland and from 9° to 0°C in Germany, Poland, and Belarus. The warm front moved northward on 22 April, followed by a cold wave front entering from the Arctic. On 23 April, the cold wave entered the southern part of the model domain while in the eastern part temperatures still reached up to 9°C. On 24 April, the cold wave quickly moved north, influencing Norway, Sweden, and Finland with near-surface temperatures between −12° and 0°C. On 25 April, warm air, with near-surface air temperatures of up to 9°C, was advected from Estonia and Latvia toward Sweden and Norway. During the episode near-surface wind speed ranged from 0 to 17 m s−1.

Initialization

To evaluate the performance of the coupled MM5–HTSVS, we ran it continuously without reinitialization for the entire episode. Doing so permits enough time for feedback between the atmospheric and land surface components of the coupled model. This means we do not use 48-h composite runs started every 12 h like Ek et al. (2003) in their evaluation of the performance of the National Centers for Environmental Prediction (NCEP) model because 1) such a procedure would reduce the amount of data that is usable for evaluation by the data needed for reinitialization and 2) the time for establishing feedback between HTSVS and MM5 would be reduced to 48 h.

The initial conditions and boundary conditions for the coupled model are taken from the NCEP–NCAR reanalysis project (NNRP). Vegetation fraction is extracted as a weighted combination of April and May mean green vegetation cover data (0.15° resolution) derived from Advanced Very High Resolution Radiometer data (Gutman and Ignatov 1998). Soil type, terrain elevation, and land use type are derived from the 10-min-resolution U.S. Geological Survey terrain and vegetation data. Soil type is constant with depth. The soil parameters used are listed in Table 2.

We initialize snow depth and soil temperatures from observation on 20 April 2000 where possible and use interpolated NNRP data elsewhere. Initial total soil moisture is interpolated from NNRP data. Total soil water is partitioned between the solid and liquid phase according to Mölders and Walsh (2004). The soil model used for the reanalysis data uses a simpler approach than that of HTSVS. Offline studies show no significant change of soil temperature and moisture with time when soil moisture is initialized within the range of the values obtained from the NNRP data. Thus, we can expect that the model results will not be strongly affected by using these soil moisture values. Initial snow temperature is assumed to be equal to initial ground surface temperature. Initial snow density is set equal to 300 kg m−3, a typical value for the Arctic and subarctic at this time of year (e.g., Sturm et al. 1997).

Datasets

The database of the BALTEX Meteorological Data Center contains snow-depth data for 355 sites in our model domain (Fig. 1). Snow depth is reported once per day. Daily averages of soil temperature measured at depths of 0.2, 0.4, 0.8, 1.6, and 3.2 m are available for five sites in Belarus. At 20 German sites, soil temperatures were recorded at depths of 0.1, 0.2, and 0.5 m at 0600, 1300, and 2000 UT. Precipitation (water equivalent) was available at 344 sites in Sweden and Finland. In total, there are 1000 soil temperature, 1775 snow-depth, and 1720 precipitation measurements.

Analysis

We evaluate the four simulations (Table 1) with reanalysis data of near-surface air temperature, humidity, wind, and sea level pressure, calculating forecast skills in accord with Anthes (1983) and Anthes et al. (1989). The performance of the coupled MM5–HTSVS is also evaluated by comparing simulated soil temperatures and snow depths with the respective observations. For comparisons, the locations of the stations are projected on the model domain under the assumption that a site is representative for the grid cell into which it falls.

To evaluate the coupled model’s ability to predict snow-depth changes, we calculate the bias scores [bias = (N1 + N2)/(N1 + N3)] and accuracy [AC = (N1 + N4)/( N1 + N2 + N3 + N4)]. These statistical scores are defined based on contingency tables in which N1 is an observed and simulated event, N2 denotes an event not observed but simulated, N3 represents an event observed but not simulated, and N4 indicates an event that is neither observed nor simulated. Each table represents the number of events for which the simulated and observed changes fall into a certain threshold class (1, 2.5, 5, and 10 mm) for a given time and simulation. The bias score measures how the coupled model simulates the frequency of occurrence for a given snow depth change, with bias = 1 indicating a perfect simulation. An averaged bias >1 or bias <1 corresponds to a systematic over- or underprediction, respectively. Accuracy AC ranges between 0 (only incorrect predictions) and 1 (only correct predictions).

In the model, snow depth increases by snowfall and decreases by outflow of meltwater, windbreak, compaction, settling, and sublimation. Accurate prediction of precipitation is important for simulating snow-depth increase. Therefore, we compare the results obtained by simulations using different cloud microphysical schemes with each other and also with observations to get insight into how the cloud treatment affects the coupled model’s prediction of snow-depth changes. Accurate prediction of alterations in the radiation budget and atmospheric cooling/heating rates is important for simulating snow-depth decrease. Thus, the results gained by simulations using different radiation parameterization schemes are also compared with each other. Because discrepancies in snow depth may result from malfunctioning of different parts of the coupled model, we calculate the skill scores for decrease in snow depth (DSD) and increase in snow depth (ISD) in addition to the scores for total change in snow depth (CSD) (overall performance). Investigation of CSD evaluates the overall performance of the coupled model. ISD is dominated by the cloud microphysical scheme, and DSD is governed by snow metamorphism and the surface energy budget.

Because the numerical scheme of HTSVS requires equal logarithmic grid spacing with respect to the soil layers, HTSVS calculates soil temperatures at depths of approximately 0.1, 0.23, 0.54, 1.27, and 2.95 m. For comparison, we interpolate predicted soil temperatures to the observational levels using a distance-weighted approach. In the following, soil layers are counted from the surface to the bottom of the soil model and are addressed with respect to the model; that is, the uppermost model layer is called layer 1, and the deepest layer is layer 5.

The coupled model’s ability to simulate the diurnal cycle of soil temperature is evaluated by the observations at the 20 German sites that record data 3 times per day. We use all 20 German and all 5 Belarusian sites in the overall evaluation of daily soil temperatures predicted by the coupled model. Root-mean-square errors (rmse) of soil temperatures are determined for each site at each depth. Considering all of the sites collectively, the total rmse is calculated for each day and for the entire episode for the Belarusian and German sites.

Results and discussions

General remarks

MM5 applies the so-called strategy of dominant land use and soil type, which means only one vegetation type and one soil type exist per grid cell. Nevertheless, the model considers heterogeneity on the microscale by a so-called mixture approach (e.g., Deardorff 1978; Kramm et al. 1996); that is, bare soil and one vegetation type are homogeneously distributed within the grid cell and no extended patches of either one (macroscale heterogeneity) exist. Subgrid-scale heterogeneity on the macroscale with respect to different vegetation or soil types and terrain elevation within a grid cell is not considered, nor is subgrid-scale variability of the snow cover taken into account. The latter would require the possibility of three different surface types (vegetation, bare soil, snow) in a grid cell. As a consequence, errors will arise from the dominant surface type approach, especially in areas of very heterogeneous soil and vegetation distribution and/or complex terrain. Because these errors have been discussed elsewhere in detail, we will not address them here (e.g., Avissar and Pielke 1989; Leung and Ghan 1995; Mölders and Raabe 1996; Giorgi and Avissar 1997; Boone et al. 2004).

Another source of discrepancies results from the comparison of simulated and observed quantities itself. Observed soil temperatures or snow depths are point measurements, whereas simulated soil temperatures and snow depths represent volume averages (35 × 35 km2 × soil layer thickness) and area averages (35 × 35 km2), respectively, for the entire grid cell. This kind of error is well known (e.g., Avissar and Pielke, 1989; Seth et al., 1994; Boone et al., 2004; Zhong et al., 2005) and therefore is not further examined here.

Primary differences between the simulations using the Schultz and Reisner schemes result from the different parameterization of cloud microphysical processes. Because comparisons of cloud microphysical schemes have been discussed elsewhere in detail (e.g., Mölders et al. 1994; Mölders and Laube 1994; Kotroni and Lagouvardos 2001) and our study focuses on evaluation of soil temperature and snow depth predicted by the coupled MM5–HTSVS, we restrict the discussion of the atmospheric part to the aspects that are important for the surface forcing, that is, 1) the performance with respect to surface pressure, wind, near-surface air temperature, and humidity prediction and 2) the differences in cloud properties as they affect surface fluxes and states indirectly through absorption, transmission, scattering, and radiative cooling/heating and directly through precipitation.

In the Schultz scheme, falling ice particles are assumed to be aggregates of nonrimed ice crystals, whereas they are ice crystals of hexagonal type in the Reisner scheme. As a consequence, for the same snow mixing ratios, snow settles up to 1.55 m s−1 quicker (Fig. 2) and the water load is removed more rapidly in the simulations with the Schultz scheme than with the Reisner scheme. Therefore, the accumulated precipitation distributions predicted by the two schemes differ considerably (e.g., Fig. 3). The altered terminal velocity also produces differences in cloud thickness and water particle transport and hence in insolation and soil heat fluxes, which cause slight soil temperature differences.

The two different radiation schemes slightly differ with respect to the assumed profiles of absorbing gases, the spectral bands considered under clear-sky conditions, and the treatment of cloud–radiation interaction under cloudy conditions. These different assumptions primarily cause slight differences in simulated radiative cooling/heating rates. The resulting differences in vertical air temperature profiles affect stability, buoyancy, vertical mixing, cloud formation/depletion, and, finally, the energy and water budget at the surface. Precipitation (Fig. 3) and latent heat flux (not shown), for instance, slightly differ in response to the radiation scheme chosen. The two fluxes mentioned affect snow-depth changes and soil temperatures.

The major differences among the four simulations are as follows. Maximum accumulated precipitation is higher in the simulations with the Schultz scheme than in those with the Reisner scheme (SC: 39.5 mm; SL: 39.5 mm; RC: 30 mm; RL: 37 mm). Precipitation starts and stops earlier in SC and SL than in RC and RL.

Atmospheric near-surface quantities

All simulations capture well the evolution of the pressure field until the end of the simulation (e.g., Fig. 4). Then, the ridge reaching from the North Sea over the Baltic Sea is slightly overestimated in intensity (∼1 hPa) but is underestimated in eastward extension.

During the entire episode, the coupled model tends to overestimate (underestimate) near-surface air temperatures in the snow-covered (snow free) areas. It is obvious that snow albedo decreases too quickly in MM5–HTSVS. Thus, reflected shortwave radiation is reduced and near-surface air temperatures increase. In MM5–HTSVS, albedo depends on vegetation fraction and soil moisture. Thus, in snow-free areas, discrepancies between the assumed 5-yr-averaged weighted and actual vegetation fraction, as well as the assumed mean and actual albedo of the vegetation, may cause this underestimation. Moreover, incorrect initial soil moisture may play a role.

Near-surface specific humidity was overestimated (up to 6 g kg−1) under snowmelt regimes, whereas the opposite is true under other conditions (e.g., Fig. 4). This fact means sublimation and evaporation are overestimated during snowmelt. Here the overestimated air temperatures contribute to an overestimation of specific humidity because relatively warm air can take up more water vapor. Over the Baltic Sea, near-surface specific humidity is overestimated up to 2.4 g kg−1 for the following reason: MM5 uses a mean sea ice distribution, under which some areas covered by sea ice in nature are ice free in the simulations. Because open water has a much lower albedo (<0.05) than does sea ice (>0.35) and because the surface temperature of open water is usually warmer than that of ice, near-surface air is relatively warmer over the seemingly open water in the model than in nature. Moreover, evaporation requires less energy than sublimation; more moisture can be supplied to the atmosphere over the seemingly nearly ice free Baltic Sea in the simulations than can be supplied from the respective frozen area in nature.

As is typical for mesoscale forecasts (e.g., Anthes et al. 1989; Colle et al. 2000, 2003), rmse and bias gradually increase with simulation time (e.g., Fig. 5). In the first layer above ground, the rmse of air temperature and u (east–west) and υ (north–south) components of the wind vector are about 11 K, 9 m s−1, and 10 m s−1, respectively, after 120 h of simulation for all four simulations. The simulations using the CLOUD scheme have lower rmse for air temperature and wind speed than do those using the CCM2 scheme (Fig. 5).

All four simulations capture well the area of high precipitation in northern Finland and the areas of lower precipitation in southern Finland close to the Russian border (around 62°N, 27°E) (Fig. 3). Both cloud microphysical schemes predict snowfall for some Finnish sites located between 60° and 65°N and between 20° and 25°E, where no snowfall was observed. In this area, they also often predict notable snowfall when only a trace or slight snowfall was observed. Note that even traces of snowfall can notably affect the surface energy and water budget and hence can affect soil temperature. The higher-than-observed accumulated precipitation in southern Finland (Fig. 3) results from the aforementioned discrepancies between the real and mean sea ice distributions.

All simulations have an offset of up to 2° in the position of the precipitation field in southern Sweden (around 57.5°N between 12° and 15°E) (Fig. 3). This offset may be explained by the grid resolution. In the model, the average terrain height within a grid cell represents elevation. Thus, the Scandinavian Ridge is flatter in the model than at its highest points in nature. Therefore, under southwestern flow over the barrier, saturation and snowfall occur later and farther northeastward in the model than in nature; water vapor is supplied to the atmosphere by sublimation and large-scale lifting in the model rather than by forced lifting at the mountain barrier. As a consequence, grid cells that correspond to sites in Sweden often receive no snowfall in the model, although snowfall is actually observed in those locations. This effect leads to an underprediction of 120-h-accumulated snowfall (Fig. 3). Note that Colle et al. (2000) and Zhong et al. (2005) reported that grid resolution significantly affects the precipitation bias.

Snow depths

Observed snow depth ranges between 0 and 1 m in Finland and between 0 and 1.97 m in Sweden. For all four simulations the following behavior is found. On average, accumulated snow depths between 0.4 and 1.2 m are overestimated, and the opposite is true for snow depth lower than 0.4 and higher than 1.6 m (Fig. 6). On the high extreme of snow depths, no snowfall occurred at these sites for which the underestimate can only result from a too-strong metamorphism. On the lower end, snow depth grew by snowfall in nature while no snowfall was predicted on the first day in many cases because of model spinup. Note that MM5 starts without clouds and it takes some time for clouds and precipitation to form in the model. Thus, this discrepancy is caused by the cloud microphysical schemes rather than by the HTSVS snow model.

Because for the same snow mixing ratio the Reisner scheme provides a lower terminal velocity (Fig. 2), snow is accumulated farther downstream in the simulations than in the simulations with the Schultz scheme (Fig. 3).

The correlation coefficients for simulated and observed snow-depth amounts are 0.96, 0.97, 0.98, and 0.98 for RC, SC, RL, and SL, respectively. For all simulations the rmse in snow depth increases with time. After the main snowfall event (22 April), averaged absolute differences remain nearly constant for all simulations (Fig. 7). The best prediction of snow depth with respect to the averaged absolute differences and rmse is provided by SL, followed by RL. Based on the (maximum) absolute differences and rmse, the coupled model will provide better snow-depth predictions if the CLOUD scheme is used instead of the CCM2 scheme.

Total change in snow depth

Changes in snow depth encompass the effects of both increases and decreases in snow depth. Comparison of simulated and observed CSD permits assessment of the coupled model’s overall performance with respect to snow predictions.

Overall, the number of grid cells with no CSD observed but with CSD predicted (N2) and with CSD observed but not predicted (N3) decreases when the simulations are performed with the Reisner scheme (Fig. 8). Note that, for SC, RC, SL, and RL, N2 is 8, 8, 8, and 7 and N3 is 25, 19, 23, and 23, respectively. The reasons for these discrepancies are manifold. The assumption on ice crystal type affects terminal velocity (Fig. 2), for which the timing of precipitation onset and ending differs between the real and model worlds. The use of a mean sea ice distribution overestimated precipitation for some Finnish sites or even predicted precipitation when none occurred. Because HTSVS neglects subgrid-scale heterogeneity with respect to snow, discrepancies between the model and real world occur during snowmelt. In nature, radiative effects enhance snowmelt around obstacles that stick out of the snow. Their low albedos lead to a warming of the ambient air and increased melt rates. Once an area becomes snow free, appreciable free convection can be generated in nature, and the different heating rates and near-surface temperatures of snow-free and adjacent snow-covered patches can produce air circulation that is similar to sea or vegetation breezes (e.g., Baker et al. 1999). This phenomenon explains, for instance, the N3 values of SC and RC in the areas around 62°N from 25° to 30°E (Fig. 8). On occasion, discrepancies also result from blowing snow. In the coupled model, resuspension and redeposition are assumed to occur in the same grid cell, whereas in nature the horizontal transport may cause a decrease/increase at a site that falls into another grid cell with respect to the model, that is, N2 or N3 may change.

Both RL and SL only slightly differ with respect to the bias for all threshold values of changes in snow depths (Fig. 9). Here, RL (SL) differs notably from RC (SC), which shows a slightly higher bias than RL (SL) for the 1-mm threshold but shows a much lower bias for the 5-mm threshold. Both SL and RL have similar accuracy, and accuracy is the highest for low and high thresholds of CSD. Because in SC snow falls much quicker than in RC, no precipitation is falsely predicted for sites downwind of the main precipitation area, for which SC provides the highest accuracy for the total changes of intermediate thresholds of CSD. Overall, RC shows the weakest performance.

Increase in snow depth

A decrease of snow depth by snow metamorphism is usually superimposed on any snow-depth increase by snowfall. Thus, discrepancies in ISD may result from snowfall that is incorrectly predicted by the cloud microphysical scheme plus incorrect simulation of metamorphism processes by the HTSVS snow model.

For the reasons discussed before, snowfall is predicted in Finland in areas where no or only a trace of snow was observed (Fig. 3). This error is the most prominent for SC and the least for RC. It is obvious that the Reisner scheme represents the snow crystal type occurring in this event better than does the Schultz scheme and hence captures better the precipitation distribution and amount and consequently ISD.

In general, the bias of ISD is greater for low (1 or 2.5 mm) than for high (5 or 10 mm) thresholds (Fig. 9), because a slight precipitation event is much harder to predict than a notable one. In accord with findings by various authors (e.g., Chaumerliac et al. 1991; Mölders et al. 1995; Rassmussen et al. 2002), the onset of precipitation strongly differs depending upon the assumptions made about ice crystal type and microphysical processes. Therefore, the bias of low ISD is dominated by these systematic errors from the cloud microphysical scheme. The generally high bias for 1-mm thresholds for SL and SC, which is slightly higher than that for RL, indicates that often precipitation begins too early because of the high terminal velocities in the Schultz scheme. It is obvious that RC provides the highest bias for ISD, whereas the other simulations perform similarly, in broad terms. For intermediate thresholds of ISD the bias of RC is approximately 2 times that for the other simulations, with SC having the lowest bias. Note that ISD results from precipitation reduced by sublimation that is weaker than the precipitation. It is obvious that RC overestimates sublimation for which the bias is high for low thresholds of ISD.

All simulations provide similar accuracy for threshold values greater than 10 mm of ISD (Fig. 9). ISD of this magnitude can only be achieved if cloud thickness is great and precipitation is strong, because then differences in sublimation become comparatively small relative to the overall increase. In general, RC has the lowest accuracy and SC has the highest accuracy for ISD. For accuracy, RL and SL only slightly differ. According to our findings, the inaccuracy in predicting ISD by the coupled model is caused by the cloud microphysical schemes rather than by the feedback of processes simulated by HTSVS.

Decrease in snow depth

Overestimation/underestimation of snowfall, clouds that are too thick (thus reducing incoming radiation and hence sublimation), and overestimation of snow metamorphism are potential reasons for false prediction of DSD. In general, great (absolute) snow-depth decreases correspond to thick snowpacks where compaction and settling dominate. The coupled model underestimates snow depth for great values of snow depth (Fig. 6); that is, it overestimates snow metamorphism. Sensitivity studies show that an overestimate of initial snow density may be the reason. The rate of DSD by compaction depends on the weight of the overlying snowpack. An initial snow density that is too low consequently yields a DSD that is too large (Fig. 10) and can explain the differences found in our study. Note that the relative DSD (change in snow depth normalized with the total snow depth) is greater in thin than in thick snowpacks.

In thin snowpacks, the contribution of sublimation can be of the same order of magnitude as compaction and settling or can be greater. This situation means predicted DSD is more sensitive to incorrect prediction of the surface energy budget for thin snowpacks than for thick snowpacks. Thin snowpacks are also more sensitive to incorrectly predicted traces of snowfall because the relative change in snow depth is higher than for thick snowpacks. As already mentioned, the coupled model predicts snowfall at some sites in Finland where none was observed. This shortcoming will propagate to incorrectly predicted DSD.

As the threshold for DSD approaches 5 mm, the bias of SC, RL, and SL approaches 1, indicating a nearly perfect prediction (Fig. 9). All four simulations slightly overestimate DSD for thresholds higher than 5 mm. Because the process of snow metamorphism is captured well by HTSVS for the intermediate range of DSD, we may conclude that initialization of thick snowpacks with a snow density of 300 kg m−3 is inappropriate. As can be derived from Fig. 10, higher initial values of snow density provide lower DSD and should be favored for initialization of thick snowpacks.

The accuracy for DSD is the highest for SC and the lowest for RC for all thresholds (Fig. 9). Accuracy increases with increasing threshold of snow-depth decrease. A slight decrease in snow is hard to capture. In RC, snow crystals sediment slower than in SC, and sometimes traces of snowfall are predicted, superimposed on a decrease by snow metamorphism, when there is no snow falling in nature.

Generally high DSD rates coincide with low cloudiness or no clouds. Thus, at high thresholds of DSD, differences in accuracy decrease among the simulations using the different cloud microphysical schemes (Fig. 9). Sublimation depends, among other things, on solar radiation reaching the ground, which can be appreciably reduced by clouds. Thus, the slow removal of the hydrometeor load with the higher water supply to the atmosphere in RC is responsible for the lower accuracy at small thresholds as compared with the other simulations. Around 5 mm, DSD, RL, SC, and SL show similar accuracy.

Soil temperature

Because no three-dimensional datasets on soil type are available at the resolution of NWP models, these models assume the same soil parameters for the entire soil column. Thus, the comparison performed here evaluates the typical practice in NWP modeling. Some of the discrepancies found between simulated and observed soil temperatures stem from this assumption. In the southern Baltic region, layers of different soil material frequently occur because most soils are sediments deposited during the ice ages.

For most of the 25 sites that are available, the coupled model tends to underestimate soil temperatures slightly (Fig. 11). Here the underestimate of near-surface air temperatures is the main reason. Correlation coefficients of simulated and observed soil temperatures are 0.917, 0.936, 0.955, and 0.949 for RC, SC, RL, and SL, respectively.

Rmse (Table 3) and differences between simulated and observed soil temperatures are the highest in the uppermost layer because variability is greater close beneath the surface than deeper in the soil and is hence more difficult to predict (Mölders et al. 2003b). Also, the uppermost layer can be strongly affected by incorrectly simulated atmospheric forcing. Furthermore, the chances that the soil temperature field is affected by the sensor installation are higher close beneath the surface than deeper in the soil. The impact of the atmospheric forcing and the diurnal variability decrease with depth, as do the differences between simulated and observed soil temperatures and rmse, on average.

In general, rmse and differences slightly increase with time. Over all soil temperature sites, rmse will be appreciably lower (up to 0.7 K) if the CLOUD scheme is chosen. For the weather situation in our study, there is no obvious advantage for any cloud microphysical scheme. However, our finding that radiation has a greater impact than cloud microphysical scheme on soil temperature prediction may be due to the fact that no precipitation occurred at the soil temperature sites. If snow occurred at the soil temperature sites, the accuracy of the snowfall prediction would influence the insulation and the heat flux into the soil, finally affecting soil temperature prediction. Note that a 10% error in snow depth alters rmse by ±0.1 K, on average, over an entire winter (Mölders and Romanovsky 2005, manuscript submitted to J. Geophys. Res.), whereas a delay of snow cover onset by 10 days, for instance, can decrease maximum soil temperature by 9, 2.9, 2, and 1.1 K at depths of 0, 0.5, 1, and 2 m (Ling and Zhang 2003).

At the five Belarusian sites, daily soil temperature averages are predicted correctly to within a range from −1 to 1.4 K. In layers 2 to 5, rmse is about 0.3 K. At the 20 German sites, soil temperatures are underestimated by about 0.5–1 and 2 K in layers 2 and 3, respectively.

At some of the 25 sites, discrepancies are greater in layers 3 and 4 than at other layers. These discrepancies may result from assuming vertically constant soil characteristics in HTSVS, whereas in nature the soil characteristics may vary with depth at these sites. At some sites, rmse and differences remain greater in the lowermost layer than in the layer above. This behavior has to be considered as a model artifact that results from the constant soil temperature boundary condition at the bottom of the soil model. Sensitivity studies performed with a zero flux boundary condition showed lower rmse in layer 5 than did the simulations using constant temperature and moisture conditions.

Diurnal course

According to the observations at the 20 German sites for which soil temperatures were reported at 0600, 1300, and 2000 UT, the coupled model acceptably captures the diurnal course of soil temperature (e.g., Fig. 12). The rmse and absolute differences are higher at noon than early in the morning or evening. On average, all simulations slightly underestimate soil temperature for layer 2 at 0600 UT, the value at 1300 UT is underestimated at most by 2.5 K, and the value at 2000 UT is predicted almost exactly. The simulations with the CLOUD scheme capture the nighttime minimum soil temperatures better than do those with the CCM2 scheme.

Conclusions

The aim of our study is to evaluate HTSVS coupled to MM5 using soil temperature, snow-depth, and precipitation data obtained during a snowmelt episode of BALTEX. Because snow-depth changes depend, among other things, on the accurate prediction of precipitation and radiation, sensitivity studies are performed alternatively using two different radiation and cloud microphysical schemes.

In general, discrepancies result from the assumption of a mean sea ice distribution and coarse grid resolution. Water that is assumed to be open in the model but that is ice covered in the real world increases the atmospheric water supply and finally leads to an overestimate of snowfall in the lee of the Baltic Sea and incorrect changes in snow depth. The resolution-dependent discrepancies between the terrain in the model and the real world lead to snowfall where none or only traces occurred and consequently lead to incorrect input for the HTSVS snow model and incorrect changes in snow depth.

The main differences in performance with respect to snow-depth changes occurring between the simulations with the Reisner and Schultz schemes are related to differences in simulated onset and end of precipitation. Once an area is influenced by high pressure, the radiation scheme will gain importance.

The simulation using the Reisner and CLOUD schemes shows the best overall performance. The accuracy of the forecast of low snow-depth changes (1 or 2.5 mm) is considerably higher for the Schultz and CCM2 radiation scheme than for all other combinations. The performance of the coupled model in predicting great snow-depth changes (10 mm) is nearly independent of the cloud microphysical and radiation schemes used. For thick snowpacks, the accuracy of snow-depth decrease by metamorphism strongly depends on the initial value of snow density.

In general, the bias for snow-depth increase is greater for low (1 or 2.5 mm) than for high (5 or 10 mm) thresholds and is much greater than for a decrease in snow depth. This means that the bias of low snow-depth changes is dominated by systematic errors from the cloud microphysical scheme.

In general, the discrepancies in simulated and observed soil temperatures decrease with soil depth. The decrease rate is not of the same order for the bottom layer as for the other layers, because soil temperature and moisture are kept constant at the bottom of the model. Thus, a flux boundary condition should be used in long-term simulations like climate studies.

The coupled model tends to underestimate daily soil temperatures (by up to 2.5 K), because MM5 underestimates near-surface air temperatures slightly in Belarus and noticeably in Germany. It successfully captures the diurnal courses of soil temperatures, the increase of soil temperature with time, and temperature behavior with depth. The simulations with the CLOUD radiation scheme better predict the minima and maxima of diurnal soil temperature cycle than do those with the CCM2 scheme. We have to expect that soil temperature prediction may be more strongly affected by the cloud microphysical schemes in snow-covered areas because snowfall prediction errors in MM5 will have a greater impact for locations with little or no snow. Such investigations should be made as soon as a great database of soil temperature measurements under snow with concurrently measured snow depth becomes available during BALTEX.

Acknowledgments

We thank U. Bhatt, F. Chen, J. Dudhia, A. Ebel, H. Elbern, M. Jankov, A. Klioutchnikova, G. Kramm, Z. Li, F. Toussiant, J. E. Walsh, and the anonymous reviewers for fruitful discussions and helpful comments. BMBF and NSF financially supported this study under Contracts 01LD0036 and OPP-0327664. Narapusetty was financially supported by a graduate research scholarship of the University of Alaska Fairbanks Graduate School. NCAR provided computational support.

REFERENCES

  • Anthes, R. A. 1983. Regional models of the atmosphere in middle latitudes. Mon. Wea. Rev. 111:13061335.

  • Anthes, R. A., Y. H. Kuo, E. Y. Hsie, S. Low-Nam, and T. W. Bettge. 1989. Estimation of skill and uncertainty in regional numerical models. Quart. J. Roy. Meteor. Soc. 115:763806.

    • Search Google Scholar
    • Export Citation
  • Avissar, R. and R. A. Pielke. 1989. A parameterization of heterogeneous land surface for atmospheric numerical models and its impact on regional meteorology. Mon. Wea. Rev. 117:21132136.

    • Search Google Scholar
    • Export Citation
  • Baker, J. M., K. J. Davis, and G. C. Liknes. 1999. Surface energy balance and boundary layer development during snow melt. J. Geophys. Res. 104D:1961119621.

    • Search Google Scholar
    • Export Citation
  • Bonan, G. B. 1994. Comparison of two land surface process models using prescribed forcings. J. Geophys. Res. 99:2580325818.

  • Bonan, G. B., K. W. Oleson, M. Vertenstein, S. Levis, X. Zeng, Y. Dai, R. E. Dickinson, and Z-L. Yang. 2002. The land surface climatology of the Community Land Model coupled to the NCAR Community Climate Model. J. Climate 15:11151130.

    • Search Google Scholar
    • Export Citation
  • Boone, A. Coauthors 2004. The Rhône-Aggregation Land Surface Scheme intercomparison project: An overview. J. Climate 17:187208.

  • Briegleb, B. P. 1992. Delta-Eddington approximation for solar radiation in the NCAR community climate model. J. Geophys. Res. 97:76037612.

    • Search Google Scholar
    • Export Citation
  • Chaumerliac, N., E. Richard, and R. Rosset. 1991. Impact of two microphysical schemes upon gas scavenging and deposition in a mesoscale meteorological model. J. Appl. Meteor. 30:8897.

    • Search Google Scholar
    • Export Citation
  • Chen, F. and J. Dudhia. 2001. Coupling an advanced land surface hydrology model with the Penn State/NCAR MM5 modeling system. Part 1: Model description and implementation. Mon. Wea. Rev. 129:569586.

    • Search Google Scholar
    • Export Citation
  • Chen, F. Coauthors 1996. Modeling of land surface evaporation by four schemes and comparison with FIFE observations. J. Geophys. Res. 101D:72517268.

    • Search Google Scholar
    • Export Citation
  • Chen, T. H. Coauthors 1997. Cabauw experimental results from the Project for Intercomparison of Land-Surface Parameterization Schemes. J. Climate 10:11941215.

    • Search Google Scholar
    • Export Citation
  • Clapp, R. B. and G. M. Hornberger. 1978. Empirical equations for some soil hydraulic properties. Water Resour. Res. 14:601604.

  • Claussen, M. 1988. On the surface energy budget of coastal zones with tidal flats. Contrib. Phys. Atmos. 61:3949.

  • Colle, B. A., C. F. Mass, and K. J. Westrick. 2000. MM5 precipitation verification over the Pacific Northwest during the 1997–99 cool seasons. Wea. Forecasting 15:730744.

    • Search Google Scholar
    • Export Citation
  • Colle, B. A., J. B. Olson, and J. S. Tongue. 2003. Multiseasonal verification of the MM5. Part II: Evaluation of high-resolution precipitation forecasts over the United States. Wea. Forecasting 18:458480.

    • Search Google Scholar
    • Export Citation
  • Cosby, B. J., G. M. Hornberger, R. B. Clapp, and T. R. Ginn. 1984. A statistical exploration of the relationships of soil moisture characteristics to the physical properties of soils. Water Resour. Res. 20:682690.

    • Search Google Scholar
    • Export Citation
  • Deardorff, J. W. 1978. Efficient prediction of ground surface temperature and moisture, with inclusion of a layer of vegetation. J. Geophys. Res. 83C:18891903.

    • Search Google Scholar
    • Export Citation
  • de Groot, S. R. 1951. Thermodynamics of Irreversible Processes. Interscience, 242 pp.

  • de Vries, D. A. 1958. Simultaneous transfer of heat and moisture in porous media. Trans. Amer. Geophys. Union 39:909916.

  • Dudhia, J. 1993. A non-hydrostatic version of the Penn State NCAR Mesoscale Model: Validation tests and simulation of an Atlantic cyclone and cold front. Mon. Wea. Rev. 121:14931513.

    • Search Google Scholar
    • Export Citation
  • Ek, M. B., K. E. Mitchell, Y. Lin, E. Rogers, P. Grunmann, V. Koren, G. Gayno, and J. D. Tarpley. 2003. Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta model. J. Geophys. Res. 108.8851, doi:10.1029/2002JD003296.

    • Search Google Scholar
    • Export Citation
  • Eppel, D. P., H. Kapitza, M. Claussen, D. Jacob, W. Koch, L. Levkov, H. T. Mengelkamp, and N. Werrmann. 1995. The non-hydrostatic mesoscale model GESIMA: Part II. Parameterizations and applications. Contrib. Atmos. Phys. 68:1541.

    • Search Google Scholar
    • Export Citation
  • Fröhlich, K. and N. Mölders. 2002. Investigations on the impact of explicitly predicted snow metamorphism on the micro climate simulated by a meso-β/γ-scale non-hydrostatic model. Atmos. Res. 62:71109.

    • Search Google Scholar
    • Export Citation
  • Garand, L. 1983. Some improvements and complements to the infrared emissivity algorithm including a parameterization of the absorption in the continuum region. J. Atmos. Sci. 40:230244.

    • Search Google Scholar
    • Export Citation
  • Giorgi, F. and R. Avissar. 1997. Representation of heterogeneity effects in earth system modeling: Experience from land surface modeling. Rev. Geophys. 35:413438.

    • Search Google Scholar
    • Export Citation
  • Grell, G., Y-H. Kuo, and R. J. Pasch. 1991. Semi-prognostic tests of cumulus parameterization schemes in the middle latitudes. Mon. Wea. Rev. 119:531.

    • Search Google Scholar
    • Export Citation
  • Grell, G., J. Dudhia, and D. Stauffer. 1994. A description of the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5). NCAR/TN-398+STR, 122 pp.

  • Gutman, G. and A. Ignatov. 1998. The derivation of the green vegetation fraction from NOAA/AVHRR data for use in numerical weather prediction models. Int. J. Remote Sens. 19:15331543.

    • Search Google Scholar
    • Export Citation
  • Henderson-Sellers, A., Z-L. Yang, and R. E. Dickinson. 1993. The Project for Intercomparison of Land-surface Parameterization Schemes. Bull. Amer. Meteor. Soc. 74:13351349.

    • Search Google Scholar
    • Export Citation
  • Henderson-Sellers, A., B. Henderson-Sellers, D. Pollard, J. M. Verner, and A. J. Pitman. 1995. Applying software engineering metrics to land surface parameterization schemes. J. Climate 8:10431059.

    • Search Google Scholar
    • Export Citation
  • Hong, S-Y. and H-L. Pan. 1996. Nonlocal boundary layer vertical diffusion in a medium-range forecast model. Mon. Wea. Rev. 124:23222339.

    • Search Google Scholar
    • Export Citation
  • Kapitza, H. and D. P. Eppel. 1992. The non-hydrostatic mesoscale model GESIMA: Part I. Dynamical equations and tests. Contrib. Phys. Atmos. 65:129146.

    • Search Google Scholar
    • Export Citation
  • Kotroni, V. and K. Lagouvardos. 2001. Precipitation forecast skill of different convective parameterization and microphysical schemes: Application for the cold season over Greece. Geophys. Res. Lett. 20:19771980.

    • Search Google Scholar
    • Export Citation
  • Kramm, G. 1995. Zum Austauch von Ozon und reaktiven Stickstoffverbindungen zwischen Atmospheare und Biosphäre. (On the Exchange of Ozone and Reactive Nitrogen Compounds between Atmosphere and Biosphere). Maraun-Verlag, 268 pp.

    • Search Google Scholar
    • Export Citation
  • Kramm, G., N. Beier, T. Foken, H. Müller, P. Schroeder, and W. Seiler. 1996. A SVAT scheme for NO, NO2, and O3—Model description. Meteor. Atmos. Phys. 61:89106.

    • Search Google Scholar
    • Export Citation
  • Leung, L. R. and S. J. Ghan. 1995. A subgrid parameterization of orographic precipitation. Theor. Appl. Meteor. 47:95118.

  • Ling, F. and T. Zhang. 2003. Impact of the timing and duration of seasonal snow cover on the active layer and permafrost in the Alaskan Arctic. Permafrost Periglac. Process. 14:141150.

    • Search Google Scholar
    • Export Citation
  • Loth, B. and H-F. Graf. 1998. Modeling the snow cover in climate studies 1. Long-term integrations under different climatic conditions using a multilayered snow-cover model. J. Geophys. Res. 103D:1131311327.

    • Search Google Scholar
    • Export Citation
  • Luo, L. Coauthors 2003. Effects of frozen soil on soil temperature, spring infiltration, and runoff: Results from the PILPS 2(d) experiment at Valdai, Russia. J. Hydrometeor. 4:334351.

    • Search Google Scholar
    • Export Citation
  • Mölders, N. 2000. HTSVS—A new land-surface scheme for MM5. Proc. 10th PSU/NCAR Mesoscale Model Users’ Workshop, Boulder, CO, PSU/NCAR, 33–35.

  • Mölders, N. and M. Laube. 1994. A numerical study on the influence of different cloud treatment in a chemical transport model on gas phase distribution. Atmos. Res. 32:249272.

    • Search Google Scholar
    • Export Citation
  • Mölders, N. and A. Raabe. 1996. Numerical investigations on the influence of subgrid-scale surface heterogeneity on evapotranspiration and cloud processes. J. Appl. Meteor. 35:782795.

    • Search Google Scholar
    • Export Citation
  • Mölders, N. and W. Rühaak. 2002. On the impact of explicitly predicted runoff on the simulated atmospheric response to small-scale land-use changes—An integrated modeling approach. Atmos. Res. 63:338.

    • Search Google Scholar
    • Export Citation
  • Mölders, N. and J. E. Walsh. 2004. Atmospheric response to soil frost and snow in Alaska in March. Theor. Appl. Climatol. 77:77115.

  • Mölders, N., H. Hass, H. J. Jakobs, M. Laube, and A. Ebel. 1994. Some effects of different cloud parameterizations in a mesoscale model and a chemistry transport model. J. Appl. Meteor. 33:527545.

    • Search Google Scholar
    • Export Citation
  • Mölders, N., M. Laube, and G. Kramm. 1995. On the parameterization of ice microphysics in a mesoscale α weather forecast model. Atmos. Res. 38:207235.

    • Search Google Scholar
    • Export Citation
  • Mölders, N., U. Haferkorn, J. Döring, and G. Kramm. 2003a. Long-term investigations on the water budget quantities predicted by the hydro-thermodynamic soil vegetation scheme (HTSVS)—Part I: Description of the model and impact of long-wave radiation, roots, snow and soil frost. Meteor. Atmos. Phys. 84:115135.

    • Search Google Scholar
    • Export Citation
  • Mölders, N., U. Haferkorn, J. Döring, and G. Kramm. 2003b. Long-term investigations on the water budget quantities predicted by the hydro-thermodynamic soil vegetation scheme (HTSVS)—Part II: Evaluation, sensitivity, and uncertainty. Meteor. Atmos. Phys. 84:137156.

    • Search Google Scholar
    • Export Citation
  • Pielke, R. A. 1984. Mesoscale Meteorological Modelin. g. Academic Press, 612 pp.

  • Philip, J. R. 1957. Evaporation, and moisture and heat fields in the soil. J. Meteor. 14:354366.

  • Philip, J. R. and D. A. de Vries. 1957. Moisture in porous materials under temperature gradients. Trans. Amer. Geophys. Soc. 18:222232.

    • Search Google Scholar
    • Export Citation
  • Prigogine, I. 1951. Introduction to Thermodynamics of Irreversible Processes. Interscience Publishing, 242 pp.

  • Raschke, E. Coauthors 1998. The Baltic Sea experiment BALTEX: A brief overview and some selected results of the authors. Geophysics 19:122.

    • Search Google Scholar
    • Export Citation
  • Rassmussen, R. M., I. Geresdi, G. Thompson, K. Manning, and E. Karplus. 2002. Freezing drizzle formation in stably stratified layer clouds: The role of radiative cooling of cloud droplets, cloud condensation nuclei, and ice initiation. J. Atmos. Sci. 59:837860.

    • Search Google Scholar
    • Export Citation
  • Reisner, J., R. J. Rasmussen, and T. Bruintjes. 1998. Explicit forecasting of supercooled liquid water in winter storms using the MM5 mesoscale model. Quart. J. Roy. Meteor. Soc. 124B:10711107.

    • Search Google Scholar
    • Export Citation
  • Robock, A., K. Y. Vinnikov, and C. A. Schlosser. 1997. Evaluation of land-surface parameterization schemes using observations. J. Climate 10:377379.

    • Search Google Scholar
    • Export Citation
  • Schlosser, C. A. Coauthors 2000. Simulation of a Boreal grassland hydrology at Valdai, Russia: PILPS phase 2(d). Mon. Wea. Rev. 128:301321.

    • Search Google Scholar
    • Export Citation
  • Schultz, P. 1995. On explicit cloud physics parameterization for operational numerical weather prediction. Mon. Wea. Rev. 123:33213343.

    • Search Google Scholar
    • Export Citation
  • Sellers, P., Y. Mintz, Y. Sud, and A. Dalcher. 1986. A Simple Biosphere Model (SiB) for use within general circulation models. J. Atmos. Sci. 43:505531.

    • Search Google Scholar
    • Export Citation
  • Seth, A., F. Giorgi, and R. E. Dickinson. 1994. Simulating fluxes from heterogeneous land surfaces: Explicit subgrid method employing the Biosphere–Atmosphere Transfer Scheme (BATS). J. Geophys. Res. 99:1865118667.

    • Search Google Scholar
    • Export Citation
  • Slater, A. G., A. J. Pitman, and C. E. Desborough. 1998. The validation of a snow parameterization designed for use in general circulation models. Int. J. Climatol. 18:595617.

    • Search Google Scholar
    • Export Citation
  • Stephens, G. L. 1978. Radiation profiles in extended water clouds. II: Parameterization schemes. J. Atmos. Sci. 35:21232132.

  • Stephens, G. L. 1984. The parameterization of radiation for numerical weather predication and climate models. Mon. Wea. Rev. 112:13301339.

    • Search Google Scholar
    • Export Citation
  • Sturm, M., J. Holmgren, M. König, and K. Morris. 1997. The thermal conductivity of seasonal snow. J. Glaciol. 43:2641.

  • Thompson, G., R. M. Rasmussen, and K. Manning. 2004. Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part I: Description and sensitivity analysis. Mon. Wea. Rev. 132:519542.

    • Search Google Scholar
    • Export Citation
  • Verseghy, D. L. 1991. CLASS—A Canadian land surface scheme for GCMs. I. Soil model. Int. J. Climatol. 11:111133.

  • Wood, E. F. Coauthors 1998. The Project for Intercomparison of Land-surface Parameterization Schemes (PILPS) Phase 2(c) Red-Arkansas River basis experiment: 1. Experiment description and summary intercomparisons. Global Planet. Change 19:115135.

    • Search Google Scholar
    • Export Citation
  • Yang, Z-L., R. E. Dickinson, A. Robock, and K. Ya Vinnikov. 1997. Validation of the snow submodel of the Biosphere–Atmosphere Transfer Scheme with Russian snow cover and meteorological observational data. J. Climate 10:353373.

    • Search Google Scholar
    • Export Citation
  • Zhong, S., H-J. X. Bian, J. Charney, W. Heilman, and B. Potter. 2005. Evaluation of real-time high-resolution MM5 predictions over the Great Lakes region. Wea. Forecasting 20:6381.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Terrain data as used in the model and schematic view of model domain location as well as locations of snow sites, indicated by asterisks, soil temperature sites in Belarus (53°–57°N, 25°–30°E), indicated by gray dots, and soil temperature sites in Germany (50°–55°N, 5°–15°E), indicated by black dots, used for evaluation. Contour lines of elevation are at 0, 200, 600, and 1000 m, respectively.

Citation: Journal of Applied Meteorology 44, 12; 10.1175/JAM2311.1

Fig. 2.
Fig. 2.

Comparison of terminal velocity as obtained by the Schultz and Reisner schemes at various snow mixing ratios.

Citation: Journal of Applied Meteorology 44, 12; 10.1175/JAM2311.1

Fig. 3.
Fig. 3.

Comparison of observed (gray dots) and predicted (contour lines) 120-h-accumulated precipitation given as water equivalent (mm), as obtained by (a) SC, (b) RC, (c) SL, and (d) RL. Light gray: 0 mm; medium gray: greater than 0 mm but less than 1 mm; dark gray: greater than or equal to 1 mm but less than 2 mm; and black: equal to or greater than 2 mm. White areas mean that there were no precipitation data available.

Citation: Journal of Applied Meteorology 44, 12; 10.1175/JAM2311.1

Fig. 4.
Fig. 4.

Comparison of (a) sea level pressure distribution (hPa), (b) surface air temperature distribution (K), and (c) specific humidity (g kg−1) as obtained for RL (solid lines) and from reanalysis (dashed lines) on 0000 UT 26 Apr. Note that distributions for SC, RC, and SL look similar.

Citation: Journal of Applied Meteorology 44, 12; 10.1175/JAM2311.1

Fig. 5.
Fig. 5.

Temporal evolution of rmse for (a) air temperature and (b) wind speed in the first layer above ground as obtained for the four simulations.

Citation: Journal of Applied Meteorology 44, 12; 10.1175/JAM2311.1

Fig. 6.
Fig. 6.

Comparison of simulated and observed snow depth as obtained for RL. Note that scatterplots for SC, RC, and SL hardly differ from this figure.

Citation: Journal of Applied Meteorology 44, 12; 10.1175/JAM2311.1

Fig. 7.
Fig. 7.

Temporal evolution of (top) rmse and (bottom) average absolute difference in simulated and observed snow depth as obtained by the four simulations. Note that scores for water equivalents are about one order less than those of snow depth shown in the graphs.

Citation: Journal of Applied Meteorology 44, 12; 10.1175/JAM2311.1

Fig. 8.
Fig. 8.

Event scores of CSD determined over the entire episode as obtained for (a) SC, (b) RC, (c) SL, and (d) RL. In the legend, the first and second logical correspond to the observations and the simulations, respectively. Yes–yes: change in snow depth is observed and simulated (N1), no–yes: change is not observed but is simulated (N2), yes–no: change is observed but is not simulated (N3), and no–no: change is not observed and (correctly) not simulated (N4).

Citation: Journal of Applied Meteorology 44, 12; 10.1175/JAM2311.1

Fig. 9.
Fig. 9.

(left) Bias and (right) accuracy of (top) total change in snow depth, (middle) increase in snow depth, and (bottom) decrease in snow depth as obtained for the four simulations.

Citation: Journal of Applied Meteorology 44, 12; 10.1175/JAM2311.1

Fig. 10.
Fig. 10.

Temporal decrease of snow depth by compaction and settling only as obtained for various initial snow densities of a 2-m-thick snowpack.

Citation: Journal of Applied Meteorology 44, 12; 10.1175/JAM2311.1

Fig. 11.
Fig. 11.

Comparison of simulated and observed daily averaged soil temperatures as obtained with RL. Results for SL, RL, and RC marginally differ from the ones shown here. Note that the relatively higher temperatures correspond to soil layers close beneath the surface and that relatively lower temperatures occur deeper in the soil.

Citation: Journal of Applied Meteorology 44, 12; 10.1175/JAM2311.1

Fig. 12.
Fig. 12.

Comparison of simulated and observed diurnal pattern of soil temperatures averaged over all 20 German stations as obtained by (a) SC and (b) RL. Note that figures for RC and SL look similar to SC and RL, respectively.

Citation: Journal of Applied Meteorology 44, 12; 10.1175/JAM2311.1

Table 1.

Summary of the parameterization and initialization methods used in the different simulations of this study.

Table 1.
Table 2.

Soil characteristics assumed in our study. Here, ks, ηs, b, ψs, cSρS, and εg are the saturated hydraulic conductivity, volumetric water content at saturation (porosity), pore-size distribution index, water potential at saturation, volumetric heat capacity of the dry soil material, and emissivity of the ground, respectively. Parameters are from Clapp and Hornberger (1978), Cosby et al. (1984), Pielke (1984), and Chen and Dudhia (2001).

Table 2.
Table 3.

The rmse as obtained for the various simulations. Results include the 20 German (three observations per site per day) and five Belarusian (daily average per site each day) sites. The rmse are valid for the 5-day episode.

Table 3.
Save
  • Anthes, R. A. 1983. Regional models of the atmosphere in middle latitudes. Mon. Wea. Rev. 111:13061335.

  • Anthes, R. A., Y. H. Kuo, E. Y. Hsie, S. Low-Nam, and T. W. Bettge. 1989. Estimation of skill and uncertainty in regional numerical models. Quart. J. Roy. Meteor. Soc. 115:763806.

    • Search Google Scholar
    • Export Citation
  • Avissar, R. and R. A. Pielke. 1989. A parameterization of heterogeneous land surface for atmospheric numerical models and its impact on regional meteorology. Mon. Wea. Rev. 117:21132136.

    • Search Google Scholar
    • Export Citation
  • Baker, J. M., K. J. Davis, and G. C. Liknes. 1999. Surface energy balance and boundary layer development during snow melt. J. Geophys. Res. 104D:1961119621.

    • Search Google Scholar
    • Export Citation
  • Bonan, G. B. 1994. Comparison of two land surface process models using prescribed forcings. J. Geophys. Res. 99:2580325818.

  • Bonan, G. B., K. W. Oleson, M. Vertenstein, S. Levis, X. Zeng, Y. Dai, R. E. Dickinson, and Z-L. Yang. 2002. The land surface climatology of the Community Land Model coupled to the NCAR Community Climate Model. J. Climate 15:11151130.

    • Search Google Scholar
    • Export Citation
  • Boone, A. Coauthors 2004. The Rhône-Aggregation Land Surface Scheme intercomparison project: An overview. J. Climate 17:187208.

  • Briegleb, B. P. 1992. Delta-Eddington approximation for solar radiation in the NCAR community climate model. J. Geophys. Res. 97:76037612.

    • Search Google Scholar
    • Export Citation
  • Chaumerliac, N., E. Richard, and R. Rosset. 1991. Impact of two microphysical schemes upon gas scavenging and deposition in a mesoscale meteorological model. J. Appl. Meteor. 30:8897.

    • Search Google Scholar
    • Export Citation
  • Chen, F. and J. Dudhia. 2001. Coupling an advanced land surface hydrology model with the Penn State/NCAR MM5 modeling system. Part 1: Model description and implementation. Mon. Wea. Rev. 129:569586.

    • Search Google Scholar
    • Export Citation
  • Chen, F. Coauthors 1996. Modeling of land surface evaporation by four schemes and comparison with FIFE observations. J. Geophys. Res. 101D:72517268.

    • Search Google Scholar
    • Export Citation
  • Chen, T. H. Coauthors 1997. Cabauw experimental results from the Project for Intercomparison of Land-Surface Parameterization Schemes. J. Climate 10:11941215.

    • Search Google Scholar
    • Export Citation
  • Clapp, R. B. and G. M. Hornberger. 1978. Empirical equations for some soil hydraulic properties. Water Resour. Res. 14:601604.

  • Claussen, M. 1988. On the surface energy budget of coastal zones with tidal flats. Contrib. Phys. Atmos. 61:3949.

  • Colle, B. A., C. F. Mass, and K. J. Westrick. 2000. MM5 precipitation verification over the Pacific Northwest during the 1997–99 cool seasons. Wea. Forecasting 15:730744.

    • Search Google Scholar
    • Export Citation
  • Colle, B. A., J. B. Olson, and J. S. Tongue. 2003. Multiseasonal verification of the MM5. Part II: Evaluation of high-resolution precipitation forecasts over the United States. Wea. Forecasting 18:458480.

    • Search Google Scholar
    • Export Citation
  • Cosby, B. J., G. M. Hornberger, R. B. Clapp, and T. R. Ginn. 1984. A statistical exploration of the relationships of soil moisture characteristics to the physical properties of soils. Water Resour. Res. 20:682690.

    • Search Google Scholar
    • Export Citation
  • Deardorff, J. W. 1978. Efficient prediction of ground surface temperature and moisture, with inclusion of a layer of vegetation. J. Geophys. Res. 83C:18891903.

    • Search Google Scholar
    • Export Citation
  • de Groot, S. R. 1951. Thermodynamics of Irreversible Processes. Interscience, 242 pp.

  • de Vries, D. A. 1958. Simultaneous transfer of heat and moisture in porous media. Trans. Amer. Geophys. Union 39:909916.

  • Dudhia, J. 1993. A non-hydrostatic version of the Penn State NCAR Mesoscale Model: Validation tests and simulation of an Atlantic cyclone and cold front. Mon. Wea. Rev. 121:14931513.

    • Search Google Scholar
    • Export Citation
  • Ek, M. B., K. E. Mitchell, Y. Lin, E. Rogers, P. Grunmann, V. Koren, G. Gayno, and J. D. Tarpley. 2003. Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta model. J. Geophys. Res. 108.8851, doi:10.1029/2002JD003296.

    • Search Google Scholar
    • Export Citation
  • Eppel, D. P., H. Kapitza, M. Claussen, D. Jacob, W. Koch, L. Levkov, H. T. Mengelkamp, and N. Werrmann. 1995. The non-hydrostatic mesoscale model GESIMA: Part II. Parameterizations and applications. Contrib. Atmos. Phys. 68:1541.

    • Search Google Scholar
    • Export Citation
  • Fröhlich, K. and N. Mölders. 2002. Investigations on the impact of explicitly predicted snow metamorphism on the micro climate simulated by a meso-β/γ-scale non-hydrostatic model. Atmos. Res. 62:71109.

    • Search Google Scholar
    • Export Citation
  • Garand, L. 1983. Some improvements and complements to the infrared emissivity algorithm including a parameterization of the absorption in the continuum region. J. Atmos. Sci. 40:230244.

    • Search Google Scholar
    • Export Citation
  • Giorgi, F. and R. Avissar. 1997. Representation of heterogeneity effects in earth system modeling: Experience from land surface modeling. Rev. Geophys. 35:413438.

    • Search Google Scholar
    • Export Citation
  • Grell, G., Y-H. Kuo, and R. J. Pasch. 1991. Semi-prognostic tests of cumulus parameterization schemes in the middle latitudes. Mon. Wea. Rev. 119:531.

    • Search Google Scholar
    • Export Citation
  • Grell, G., J. Dudhia, and D. Stauffer. 1994. A description of the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5). NCAR/TN-398+STR, 122 pp.

  • Gutman, G. and A. Ignatov. 1998. The derivation of the green vegetation fraction from NOAA/AVHRR data for use in numerical weather prediction models. Int. J. Remote Sens. 19:15331543.

    • Search Google Scholar
    • Export Citation
  • Henderson-Sellers, A., Z-L. Yang, and R. E. Dickinson. 1993. The Project for Intercomparison of Land-surface Parameterization Schemes. Bull. Amer. Meteor. Soc. 74:13351349.

    • Search Google Scholar
    • Export Citation
  • Henderson-Sellers, A., B. Henderson-Sellers, D. Pollard, J. M. Verner, and A. J. Pitman. 1995. Applying software engineering metrics to land surface parameterization schemes. J. Climate 8:10431059.

    • Search Google Scholar
    • Export Citation
  • Hong, S-Y. and H-L. Pan. 1996. Nonlocal boundary layer vertical diffusion in a medium-range forecast model. Mon. Wea. Rev. 124:23222339.

    • Search Google Scholar
    • Export Citation
  • Kapitza, H. and D. P. Eppel. 1992. The non-hydrostatic mesoscale model GESIMA: Part I. Dynamical equations and tests. Contrib. Phys. Atmos. 65:129146.

    • Search Google Scholar
    • Export Citation
  • Kotroni, V. and K. Lagouvardos. 2001. Precipitation forecast skill of different convective parameterization and microphysical schemes: Application for the cold season over Greece. Geophys. Res. Lett. 20:19771980.

    • Search Google Scholar
    • Export Citation
  • Kramm, G. 1995. Zum Austauch von Ozon und reaktiven Stickstoffverbindungen zwischen Atmospheare und Biosphäre. (On the Exchange of Ozone and Reactive Nitrogen Compounds between Atmosphere and Biosphere). Maraun-Verlag, 268 pp.

    • Search Google Scholar
    • Export Citation
  • Kramm, G., N. Beier, T. Foken, H. Müller, P. Schroeder, and W. Seiler. 1996. A SVAT scheme for NO, NO2, and O3—Model description. Meteor. Atmos. Phys. 61:89106.

    • Search Google Scholar
    • Export Citation
  • Leung, L. R. and S. J. Ghan. 1995. A subgrid parameterization of orographic precipitation. Theor. Appl. Meteor. 47:95118.

  • Ling, F. and T. Zhang. 2003. Impact of the timing and duration of seasonal snow cover on the active layer and permafrost in the Alaskan Arctic. Permafrost Periglac. Process. 14:141150.

    • Search Google Scholar
    • Export Citation
  • Loth, B. and H-F. Graf. 1998. Modeling the snow cover in climate studies 1. Long-term integrations under different climatic conditions using a multilayered snow-cover model. J. Geophys. Res. 103D:1131311327.

    • Search Google Scholar
    • Export Citation
  • Luo, L. Coauthors 2003. Effects of frozen soil on soil temperature, spring infiltration, and runoff: Results from the PILPS 2(d) experiment at Valdai, Russia. J. Hydrometeor. 4:334351.

    • Search Google Scholar
    • Export Citation
  • Mölders, N. 2000. HTSVS—A new land-surface scheme for MM5. Proc. 10th PSU/NCAR Mesoscale Model Users’ Workshop, Boulder, CO, PSU/NCAR, 33–35.

  • Mölders, N. and M. Laube. 1994. A numerical study on the influence of different cloud treatment in a chemical transport model on gas phase distribution. Atmos. Res. 32:249272.

    • Search Google Scholar
    • Export Citation
  • Mölders, N. and A. Raabe. 1996. Numerical investigations on the influence of subgrid-scale surface heterogeneity on evapotranspiration and cloud processes. J. Appl. Meteor. 35:782795.

    • Search Google Scholar
    • Export Citation
  • Mölders, N. and W. Rühaak. 2002. On the impact of explicitly predicted runoff on the simulated atmospheric response to small-scale land-use changes—An integrated modeling approach. Atmos. Res. 63:338.

    • Search Google Scholar
    • Export Citation
  • Mölders, N. and J. E. Walsh. 2004. Atmospheric response to soil frost and snow in Alaska in March. Theor. Appl. Climatol. 77:77115.