Simulation of snowmelt runoff in lowland and lower Alpine by nig11470

VIEWS: 19 PAGES: 16

									Modelling Snowmelt-Induced Processes (Proceedings of the Budapest Symposium,
July 1986). IAHS Publ. no. 155,1986.



Simulation of snowmelt runoff in lowland and
lower Alpine regions of Switzerland


LUDWIG N. BRAUN & HERBERT LANG
Department of Geography, Hydrology         Section,
Swiss Federal Institute   of    Technology,
Winterthurerstr.   190, CH-8057    Zurich,
Switzerland

ABSTRACT   This study investigates the influence of
snowmelt model structure on the overall performance of
runoff modelling in basins ranging from 3.2 to 1696 km 2 .
For a given basin, melt rates as calculated by various
snowmelt models of different complexity are used as
input to the same runoff model, and the performance is
assessed by linear scale plots of measured and simulated
discharge and by a numerical efficiency criterion. In
small to medium basins (<1000 km 2 ) there is a noticeable
improvement in model performance when moving from a
temperature index method to any other method investigated.
During advection-melt conditions an energy balance method
yields the best results, while during radiation-melt
conditions index methods are sufficient. In larger
basins (>1000 km ) , a noticeable change in model
performance is observed only in individual years
characterized by an above-normal snow pack. For
operational purposes, the combination method according
to Anderson (1973) seems to be the most suitable
approach of the methods investigated.

La simulation   de l'écoulement   des eaux de fonte                        dans   les
régions helvétiques     basses et    préalpines
RESUME   Cette étude étudie l'influence de la structure
de modèles de fonte de neige sur la performance générale
de modèles de prévision d'écoulement de basins versants
d'une superficie de 3.2 à 1696 km 2 . Pour un bassin
versant donné, des vitesses de fonte calculées avec des
modèles de fonte de neige de différentes complexités
sont utilisées comme données pour le même modèle de
prevision d'écoulement. La performance est évaluée sur
la base des courbes linéaires de débit calculées et
mesurées ainsi qu'avec un critère d'efficacité numérique.
Pour de petits basins versants (<1000 km 2 ) la performance
du modèle d'écoulement augmente de manière significative
si l'on passe d'un modèle de fonte de neige basé sur la
température de l'air à tout autre modèle de fonte de
neige. Lors de conditions de fonte par advection les
meilleurs résultats sont obtenus avec un modèle basé
sur le bilan énergétique alors qu'en périodes de fort
rayonnement un modèle basé sur la température de l'air
est suffisant. Pour de grands basins versants (>1000 km 2 )
                                                                                    125
126 L.N.Braun     &   H.Lang

                la performance d'un modèle ne change de manière
                significative que pour certaines années avec une couche
                de neige supérieure à la moyenne. Pour les applications
                pratiques, la méthode combinée selon Anderson (1973)
                semble être l'approche la plus appropriée des méthodes
                analysées.

INTRODUCTION
This study evaluates the influence of snowmelt model structure on
the overall performance of runoff modelling in various sized
basins. At lower elevations, a great part of the snow cover is
typically short-lived. Special attention is given to the fact
that temporal resolution of data and steps of calculation need to
be higher in small basins as compared to larger ones for
adequately simulating runoff events.
    In the literature considerable attention is devoted to assessing
the performance of point snowmelt models of different complexities.
Significant contributions are made by Anderson (1976) and Obled &
Rosse (1977), who compared melt rates calculated by different
approaches with snowmelt measured using a snow pillow and snow
lysimeters respectively. Compared with index methods, detailed
energy balance approaches requiring additional data, such as
radiation, generally yield better simulation of snowmelt in open
area, as they can account for different meteorological conditions.
For forested sites however, temperature index methods generally
yielded acceptable results. Rosl (1976) investigated the
performance of snowmelt models as used in various runoff models at
sites located in the lowland and lower alpine regions
 (Mittelgebirge) of the Federal Republic of Germany (FDR). In his
analysis, the snow accumulation and ablation model of the NWSRFS
 (Anderson, 1973) performed best. This approach employs a
temperature index method (seasonally varied melt factor) during
fair weather conditions, and calculations of the turbulent
exchange of heat and water vapour using empirical wind functions,
of longwave radiative exchange and of heat gained by precipitation
during rainy periods.
    Assessing the performance of various snowmelt models over a
basin in conjunction with a precipitation runoff model is a more
difficult endeavour. Apart from the temporal variability, the
spatial variability of input data needs to be considered. There is
a WMO project for the intercomparison of conceptual models of
snowmelt runoff near its completion (WMO, 1982), where some 10
different models are tested over six watersheds ranging from 8.7
to 2170 km . The results of the project are to be published soon.
    Charbonneau et al. (1981) tested different snowmelt runoff
models in an alpine basin (Durance River, France, 548 km 2 ). They
conclude that the choice of particular interpolation procedures
for input data such as air temperature and precipitation is much
more crucial than the level of sophistication of individual
snowmelt models.
    Both the WMO project and Charbonneau et al. assess the overall
performance of the different snowmelt runoff models. In this study
the sole influence of the different structures of the snow
                          Simulation   of snowmelt   runoff   in Switzerland   127

accumulation and ablation model is investigated by using the same
interpolation procedures for meteorological data and the same
runoff model. Flemming & Gurtz (.1983) report first results of
such an investigation in two lowland to lower alpine basins
(Mittelgebirge, 4.9 and 363 km 2 ) situated in the German Democratic
Republic (GDR). They found that temperature index methods
performed as well as an energy balance method.
   Braun & Zuidema (.1982) have demonstrated that an energy balance
method yields better results in snowmelt runoff modelling than
simple temperature index methods in the Rietholzbach basin
(3.2 km2) during advection-melt situations. In this study,
published in more detail in Braun (1985), a wider variety of
meteorological conditions and various other index methods are
considered not only in this small basin, but in several other
basins ranging from 79 km to 1696 km2. We try to show to what
extent snowmelt model structure, including the choice of different
meteorological input variables, influences the overall runoff
model performance in the lowland to lower alpine environment in
eastern Switzerland using one and the same runoff model. The
results should also help to find the appropriate level of snowmelt
model complexity for the operational forecast procedures for the
the Rhine River system as used by the Hydrology Section of the
Swiss Federal Institute of Technology (ETH) in Zurich (see for
example Vischer, 1978, and Jensen, 1980).


METHODS AND DATA
The following snowmelt models of varying complexity are compared:
   (a) temperature index method (Bergstrom, 1976);
   (b) temperature and wind index method (T,u-index);
   (c) combination method (Anderson, 1973);
   (d) extended combination method (using also water vapour
pressure as an input variable)
   (e) energy balance method (calculation of the turbulent fluxes
of heat and water vapour based on Price & Dunne, 1976).
   In each case a water retention of the snow cover of 10% of its
water equivalent is assumed. Refreezing of liquid water in the
snow cover during periods with negative air temperatures is
modelled according to the snow routine of the HBV runoff model as
developed by the Swedish Meteorological and Hydrological Institute
(Bergstrom, 1976). Melt rates as calculated by the various
snowmelt models serve as input in a runoff model, and the overall
model performance is assessed comparing linear scale plots of
measured and simulated discharge and by the numerical efficiency
criterion R2 (Nash & Sutcliffe, 1970) which is defined as follows:

               F 2 - F2
        ™2 _    O




where
    2
F       = the sum of squares of the differences between measured and
128 L.N.Braun   S   H.Lang

      and computed discharge;
F2 = the sum of squares of the differences between measured
 °     discharge and mean discharge over the period investigated
       (also called initial variance or no model value of F ).
   The R2-criterion is a relative measure which facilitates
comparisons between different snowmelt routines when applied to
the same basin and the same period of time. It takes on values
ranging from -°° to +1, whereby +1 represents a complete agreement
between measured and computed hydrographs. R2-values are not
directly comparable between different basins and different time
periods as the initial variance F 2 may be quite different. The
R2-criterion has been found to reflect fairly well the general
conclusions of model performance based on visual inspection
(Bergstrom, 1976)„ Ideally, other numerical and graphical criteria
should be used to cover different aspects of model performance.
In this study, however, the emphasis lay on visual inspection of
linear scale plots of measured and simulated discharge of all
years, basins and snowmelt models investigated. A more complete
evaluation of the performances on this basis is given in Braun
(1985).
   Optimization of selected model parameters is achieved by mapping
R2-values as a function of these parameters (analysis of response
surfaces). In all periods investigated these optimized "generally-
valid" parameter values are applied with the exception of the
correction factors for rainfall and snowfall, which are optimized
for each period to yield a minimal value of accumulated
differences between simulated and measured discharge.


The   basins
The different snowmelt models are applied in the Rietholzbach basin
 (3.2 km ) and in four various-sized basins of the Thur River, all
located in northeastern Switzerland (see Fig.l). The Thur basin
above Andelfingen has an area of 1696 km2 and ranges from 356 m to
2500 m a.m.s.l. For calculation purposes it is subdivided into
seven sub-basins. Sub-basins 3, 4, 5, 6 and 7 constitute the Thur
basin above Halden with an area of 1085 km2. Sub-basins 4 and 5
constitute the Thur basin above Jonschwil with an area of 493 km2,
and sub-basin 2 is identical with the Murg basin above Wangi with
an area of 79 km . Mean yearly precipitation across the whole
Thur basin ranges from 900 mm in the lowlands of the northern part
to up to 2500 mm at elevations above 1500 m a.m.s.l. in the
southern part. Mean annual runoff at Andelfingen is 1080 mm. Below
about 700 m a.m.s.l. the mean annual temperature is around 8°C, the
mean temperature in January is about -1°C and in July about 18°C.
Coniferous and deciduous forests cover some 30% of the area. Based
on measurements in western Switzerland by Maurer et al. (1975) it
was established that a snow cover is formed and disappears again
about 8 times during one winter at elevations between 400 m and
900 m„
    The geographical location of the Rietholzbach basin within the
Thur basin is given in Fig.l. It ranges from 680 m to 960 m
a.m.s.l., and about 20% is forested. Mean annual precipitation is
                      Simulation   of snowmelt   runoff     in Switzerland       129




            FiG.l   Location   of the Thur and the Rietholzbach              basins.


1400 mm (Grebner & Goldi, 1983), and mean annual runoff is expected
to be 800 mm. Meteorological and hydrological measurements in this
basin have been taken since 1974/1975 by the Hydrology Section of
the Swiss Federal Institute of Technology in Zurich, and the aim of
this research project is to further our understanding of the water
balance and the runoff processes in this lower alpine zone (Lang,
1978). More details about the hydrology of this basin can be found
for instance in Germann et al. (1982).

Runoff   model and data used in the Rietholzbach          basin
Meteorological and hydrological data with a temporal resolution of
1 h are used for simulation in the Rietholzbach basin. Five
specific periods totalling about 300 days of marked accumulation and
ablation of snow in the winter months of the years 1976 to 1980
have been selected for simulation. No data after spring 1980 were
available at the time of analysis due to a change in the data
collection system. As a result no independent data set has been
investigated in this basin. Precipitation is measured by a heated
tipping bucket raingauge, and 1380 mm were recorded over the 300
days considered here. About 510 mm or a third of the total
precipitation fell as snow, assuming a transition air temperature
from snow to rain of 0.5°C which was found to be optimal. Total
discharge over the investigated period amounts to 1330 mm.
Meteorological data measured at one point in the Rietholzbach
basin are considered representative for the whole basin in this
analysis. A modified version of the approach used by Martinec
(1970) has been chosen as a runoff model, using a time step of 1 h.
130 L.N.Braun   & H.Lang

Runoff   model and data    used in the Thur   basin
Meteorological data used for simulation of discharge in the Thur
basins having a temporal resolution of 12 h are taken from 14
climatological stations of the Swiss Meteorological Institute.
Distribution of these data across the watershed is achieved on the
basis of a so-called "optimal" interpolation technique, where
interpolated values are a linear function of the values measured at
the stations, and where the coefficients of this function are
determined using an arbitrarily chosen functional relationship
between covariance and distance. Precipitation is determined at
the centre of gravity of each sub-basin (see Fig.l) and the
remaining meteorological variables are distributed to the various
elevation belts of 200 m vertical extent between 400 and 2200 m
a.m.s.l. Wind speed is taken as invariant across all basins and
is represented by the arithmetic mean of about 10 stations.
   The water equivalent of the snow pack is simulated for each
sub-basin and elevation belt, and calculated snowmelt and rain
inputs are used with a distributed version of the HBV-3 runoff
model to simulate daily discharge of the various Thur basins. This
conceptual runoff model described by Bergstrom (1976) has been
selected because it has a simple structure and because it is well
documented. A total of 11 hydrological years (covering the
months between October and June only) are used in this analysis,
where the years 1976/1977 to 1980/1981 served as the calibration
period, and the years 1971/1972 to 1975/1976 as well as 1981/1982 as
the verification period.
   A comparison between simulated and measured water equivalent of
the snow pack at one point can be achieved for all years
investigated, where measurements are taken twice a month between
November and April by the network of the Hydrology Section of the
Federal Institute of Technology (Schwagalp station, 1290 m
a.m.s.l., location see Fig.l). During the years 1980/1981 and
1981/1982 additional water equivalent measurements were taken along
an easily accessible snow course ranging from 400 to 1200 m
a.m.s.l. (Regelstein snow course).


RESULTS FOR THE RIETHOLZBACH BASIN
As mentioned in the previous section, the emphasis of the
assessment of model performance lay on visual inspection of linear
scale plots of measured and simulated discharge. It is beyond the
scope of this paper to give due attention to these plots, and for
further information the reader is referred to the more detailed
report of this study given in Braun (1985).
   Table 1 summarizes the efficiencies of the various snowmelt
models as applied in the Rietholzbach basin. Over the total 300
days of simulation using a temporal resolution of 1 h, an increase
in model performance of about 4% can be demonstrated when moving
from the temperature index to any other method. The energy
balance method performed slightly less well than the combination
method. The main reason may lie in the fact that slope and aspect
are not considered when calculating the shortwave radiation balance
                                  Simulation                of snowmelt     runoff      in Switzerland                      131

                                      2
TABLE 1 Performance    R of different     snowmelt  models over 298                                                  days
of simulation  and over a period    of predominant         advection-melt
(10 December 1979-13   February  1980) Rietholzbach       basin
                                                           Performance     R :
                                F2                                                        Comb.                 Energy
                                  O
                                (mm           h~l)z        T-index          T,u-index     method                balance


 Over 298 days                  526.1                      0.85          0.89             0.89                  0.88
 Period    of                   212.2                      0.86          0.91             0.93                  0.94
   predominant
   advection-melt
    (66 days) .




 in this version of the energy balance method. On days with
 predominant advection-melt conditions, differences in exposition
 play an insignificant role, and the superiority of the energy



                     precipitation (mm h"')




              -10
                    iii
               10 f air temperature CC)
                                                      i
                                      pA^/^^V-AAAAAfA/V
                                                                     JL-
                                                                           jrAJ^Af^r-^A
                                                                                        ,1, \\\
                                                                                                       .
                                                                                                           U
                                                                                                                .S\ ^

              200




                    16   18     20    22    24        26   28|                   10   12    14    16       18   20     22
                               February 1978                                     March 1978
              FIG.2      Performance       of snowmelt         routines   during                                  advection-
              and radiation-melt         situations,         Rietholzbach     basin                               1978.
              Radiation-melt      situations         identified       by # .
132 L.N.Braun      &   H.Lang

balance as compared to the index methods becomes evident (see
Table 1). Figure 2 illustrates a period of snow accumulation and
ablation in February and March 1978. On days with predominant
radiation-melt conditions, index methods perform adequately, and
the combination method using air temperature and wind speed only as
input variables seems to be the most appropriate approach for
operational purposes.


RESULTS FOR THE THUR BASINS

Influence       of snowmelt     model   structure
Figure 3 illustrates the performance of the various snowmelt




                                            extended    .   „             . .    combination     extended
                          T,u- index         . . ,.         T-mdex   T,u-index        1L   J      L.   .
                                          combination                              method      combination
                                            method                                               method
                FIG. 3 Performance of different             snowmelt models          as a
                function of their sophistication,             Thur basins             (individual
                years 1971/1972 to 1981/1982,                October-June).
                             Simulation        of snowmelt        runoff    in Switzerland           133

models as applied in the four Thur basins for every hydrological
year (October to June only) from 1971/1972 to 1981/1982. There is
a slight tendency to increased performance with increasing
sophistication of the model used, particularly in the smaller
basins and in years with an above-normal snow pack. Generally,
the temperature and wind index approach performs less well than the
temperature index method. It appears that a further input variable
such as wind speed needs to be included through physically based
functions and not as an index in order to increase model
performance. This is in contrast with the findings from the
Rietholzbach basin, where the temperature and wind index method
increased performance considerably over the temperature index
method. This may be due to the greater temporal resolution and the
increased representativeness of the wind data in the Rietholzbach
basin.
   The exceptionally low R2-values of the 1971/1972 simulations are
due to the very low F -values as compared to other years. Winter
1971/1972 had by far the smallest snow pack of all years
considered.
   Table 2 summarizes the results over the total calibration and
verification periods as well as for the year 1981/1982. It can be
stated that snowmelt model structure influences the overall
performance over several years of runoff simulations only to an



TABLE 2        Performance    R     of different        snowmelt     models,     Thur       basins



                                      F2
                                                                                           Extended
                   Size                o                 T-         T,u-       comb.       comb.
Basin             (kmz)                (mm day     ,)     index     index      method      method

Murg                 79      C        6002               0.85       0.86       0.87         0.87
(Wangi )                     V        3470               0.86       0.86       0.86         0.87
                          1981/82     2176               0.85       0.83       0.87         0.88
 Thur               493      C        25566              0.85       0.85       0.86         0.86
 (Jonschwil)                 V        14982              0.85       0.83       0.86         0.86
                          1981/82      5306              0.77       0.73       0.79         0.81
 Thur             1085       C        15355              0.84       0.84       0.85         0.85
 (Maiden)                    V         9950              0.85       0.83       0.85         0.84
                          1981/82      3094              0.78       0.78       0.81         0.82
 Thur             1696       C          8959             0.82       0.81       0.82         0.82
 (Andelfingen)               V          5582             0.81       0.80       0.80         0.80
                          1981/82       2008             0.77       0.75       0.78         0.78


C = calibration  period,         1976/1977-1980/1981
V = verification   period,        1971/1972-1975/1976
    Murg basin: October-April        (212 days)
    all other basins:      October-June    (274 days)
    1981/82: verification;       maximal snow pack in all                      the      years
    investigated
134 L.N.Braun          S H.Lang

insignificant degree. In other words, the temperature index
approach using only precipitation and air temperature as input
variables works almost as well as the extended combination method
using precipitation, air temperature, wind speed and water vapour
pressure as inputs. This conclusion applies for runoff simulations
over extended periods of time (October-June and over several years)
and using meteorological input data of a temporal resolution of 12
h, and for discharge data having a temporal resolution of one day.
    During individual years, snowmelt model structure may have a
noticeable influence in the smaller Thur basins investigated,
particularly during the years 1974/1975, 1980/1981 and 1981/1982,
when the snowpack was above average. In these years an increase in
model performance by about 5% can be demonstrated when moving from
the temperature and wind index method to the extended combination
method. Figure 4 shows part of the simulation of discharge of the
Thur above Jonschwil in 1981/1982, when the snowpack was the
greatest of all the 11 years investigated. The T,u-index method
(It2 = 0.73) overestimates melt at the beginning of January 1982,
whereas it underestimates melt in April and May 1982. The extended
combination method (R2 = 0.81) gave the best results in this case
here as well as in all other basins.

            precipitation      ( m m / 1 2 h)                  sub-basin      5 (Nesslau)

                20 +

                 0 j •• **    J l M n jili ..I, . .Mi ni. .,.1, 11 j. Ai\
                        N            0          J         F          M          A           M

                daily mean air temperature (°C)
             20 T                                                    500 m      .1300 m level
             10
              0
            -10
                                                                             é
                                                                             2100 m

            water equivalent SP (mm)                  —       simulated (ext.comb method)
                                                              simulated     (T,u-index)
            measured • Schwàgalp 1290 m a.S-l.
                         jn    1000 m Regelstein snow course

            1000




                discharge Q (mm d1)                 Thur/Jonschwil                              Q ( m J s1)
                30 +                                                     measured
                                                                          T, u - index              W0
                                                                          extended combination
                                                                                       method




                                                                     M
                              1981         '                       1982

            FIG.4      Comparison of measured and simulated     discharge                                      and
            water equivalent      of the snow cover, Thur basin above
            Jonschwil,      1981/1982.
            Simulated    water equivalent    values refer to the centre                                       of
            gravity    of sub-basin    5   (Nesslau).
                              Simulation        of snowmelt     runoff      in Switzerland       135

Measured       and simulated      water         equivalent
As an example, Fig.4 also compares measured and simulated water
equivalent of the snow cover at different elevations in sub-basin
5 (Nesslau) for winter 1981/1982 (for locations see Fig.l).
Simulated values are compared with Schwagalp data (1290 m a.m.s.l.)
and data collected at the Regelstein snow course at different
elevations. Generally speaking, the simulated water equivalents
at different elevations correspond fairly well with the Regelstein
data below the 1000 m level. However, simulated values above this
level are significantly lower than the Regelstein and the Schwagalp
data. This latter finding also applies to all other years
investigated where Schwagalp data are available» This discrepancy
may be due to the specific location of the Schwagalp station
(northern exposure, shading effects). It seems, however, also
connected to the interpolation procedure as used for precipitation.
The annual precipitation gradient as suggested by the interpolated
values is considerably lower than what can be expected for this
region (Zeller & Gensler, 1980). On the other hand, areal
precipitation and water equivalent values may not be directly
compared with point values. Despite the discrepancy between
measured and simulated water equivalent values above 1000 m,
simulated discharge of the Thur at Jonschwil corresponds fairly
well with the measured one.

Influence       of basin        subdivision
The influence of basin subdivision on model performance is
assessed for the Thur basin above Halden. Simulations using a
division into five sub-basins as given in Fig.l are compared to
simulations treating the basin as one unit with the same vertical
division into 200 m elevation belts (see Table 3)„ For both


TABLE 3      Influence         of basin  subdivision         on overall     model
performance,      Thur      basin  above Halden,         temperature    index     method,
October-June


                                     o                       5 sub-basins          1       sub-basin
Period                              (mm day'1)2              R2                    R2

Calibration       period           15     355                0.84                   0.83
Verification       period           9     950                0.85                   0.83
1974/1975                           2     333                0.86                   0.82
1978/1979                           3     428                0.83                   0.84



calibration and verification periods, a decrease of the efficiency
R2 by 0.02 (about 2%) is found when using only one unit. In
individual years, performance decreases by 4% in the year 1974/1975
(second highest maximal water equivalent at Schwagalp in all of
the 11 years considered) and increases by 1% in 1978/1979 (below-
136 L.N.Braun       Si     H.Lang

normal snowpack). It can be concluded that a more detailed
subdivision of a basin may improve runoff simulations by the same
order of magnitude as a more detailed snowmelt model.

Influence       of temporal             resolution              of meteorological   data
Here, an example of the influence of the temporal resolution of
meteorological data is given. Between 16 and 20 February 1978 a
cold polar air mass was stationary over northeastern Switzerland,
while warm maritime air was moving from the southwest across
Switzerland. This caused heavy precipitation falling as snow above
about 2000 to 2500 m a.m.s.l., as rain below that elevation, and
partly as freezing rain as it entered the cold bottom air layer in
northeastern Switzerland. All snowmelt models as applied in the




                air temperature (°C)   at 0700 and 1900 hours




                FIG.5     Simulation     of discharge of the Thur River at
                Halden, January-March        1978.
                Failure of all snowmelt models to simulate       the peaks on
                18 and 20 February 1978 was due to insufficient          resolution
                of meteorological      data and inadequacy of a fixed   air
                temperature       divider.
                    Simulation   of snowmelt runoff   in Switzerland   137
Thur basins recognized this situation as an accumulation period,
increasing the simulated water equivalent by about 80 mm (Fig.5).
At an elevation of 700 m a.m.s.l. a maximal water equivalent of the
snow pack of about 130 mm resulted in sub-basin 4 (Wasserfluh,
location see Fig.l) on February 20, which corresponds well with the
Rietholzbach basin using data of the Buel station located at about
750 m a.m.s.l. (see Fig.2). However, the snow depth data as given
in Fig.5 indicate a much smaller increase of the snow cover.
   Two distinct peaks of discharge of the Rietholzbach. and of the
Thur river are observed on 17 (18) and 20 February. These are
simulated fairly well by the energy balance method in the
Rietholzbach basin, but are not simulated at all by any of the
methods applied in the Thur basins. As a result of the
overestimation of accumulation, simulated discharge is generally
too high in early March.
   This failure can be partly ascribed to the use of the readings
of air temperature, wind speed and vapour pressure at 0700 and
1900 h as mean values for the first and second half of the day in
this analysis in the Thur basins, while in the Rietholzbach basin
hourly values of meteorological data are available. As a result,
the rise of air temperature above 0°C for a few hours around noon
on February 17 and 20 is recognized in the Rietholzbach
simulations, but not in the Thur simulations.
   Another reason for the poor performance of all snowmelt models
in this case may also be the use of a fixed transition air
temperature as the sole criterion to determine the precipitation
type. In exceptional situations such as those described above,
the temperature distribution in the free atmosphere may provide
better information to determine whether precipitation falls as
snow or rain, as suggested by Grebner (1978) for example.


CONCLUSIONS
The aim of this study is to investigate the influence of snowmelt
model structure on the overall performance of a runoff model for
different-size basins in the lower alpine to lowland environment
of Switzerland. The following conclusions can be drawn:
   (a) Major snowmelt flood events usually occur during advection-
melt situations associated with heavy rainfall. Sensible and
latent heat fluxes are the primary sources for melt in these cases.
In small-scale to medium-scale basins (up to about 1000 km2) the
combination method according to Anderson (1973) brings an
improvement of about 5% over index methods in years with an above-
normal snow pack. A further slight improvement can be achieved
by including measured water vapour pressure (extended combination
method) or by using an energy balance method. No assessment of the
statistical significance of these improvements has been attempted,
                                     s
however, recent results of Cavadias 8 Morin (1985) indicate that
differences in R2-values need to be larger than 5% in order to
claim that they are significantly different.
   (b) In basins well over 1000 km2 (such as the Thur basin above
Andelfingen) the influence of the snowmelt model structure on
runoff simulations over several months cannot be demonstrated,
138 L.N.Braun   &    H.Lang

Further investigations of individual melt events using a higher
temporal resolution of hydrometeorological data are needed because
snowmelt model structure must play a more significant role on a
shorter time scale.
   (c) Preliminary results of the 1M0 project for the
intercomparison of conceptual models of snowmelt runoff seem to
indicate that differences in the oyerall performance of various
snowmelt runoff models are larger than differences in performance
using the same runoff model in connection with different point
snowmelt models as investigated here. In other words: the choice
of a particular runoff model may influence the results more than
the choice of a particular snowmelt model.
   (d) Simulated areal values and measured point values of the
water equivalent of the snow pack compare fairly well at elevations
below 1000 m a.m.s.l., but at higher elevations there is a
considerable discrepancy between simulated and measured values.
Despite this discrepancy the overall model performance of snowmelt
runoff simulations can be considered satisfactory in most cases.
Further investigations should lead to a better understanding of the
areal distribution of precipitation and other meteorological
variables in mountainous regions.
   Ce) A more detailed subdivision of a basin may improve runoff
simulations to the same extent as a more detailed snowmelt model.
   (f) A fixed air temperature divider to determine the state of
aggregation of precipitation is generally adequate in medium-scale
basins. However, under certain meteorological conditions
considerable errors in runoff simulations may occur, and in these
cases data from upper air soundings should be incorporated into
snowmelt models.
   (g) More research should be aimed at the link between large-
scale air mass characteristics and snowmelt. Moore & Owens (1984)
have demonstrated the utility of large-scale indices in a maritime
alpine environment for forecasting purposes. Large-scale factors
are closely connected with the turbulent exchange of heat and
water vapour over snow and are more predictable than surface air
characteristics.


ACKNOWLEDGEMENTS        D.Grebner made valuable comments on the
manuscript.


REFERENCES
Anderson, E.A. (1973) National river forecast system - snow
   accumulation and ablation model.           NOAA Tech.   Memo.   NWS
   HYDRO-17, US Dept of Commerce, Silver Spring, Maryland.
Anderson, E.A. (1976) A point energy and mass balance model of a
   snowcover.       NOAA Tech.   Rep.   NWS HYDRO--19, US Dept of Commerce,
   Silver Spring, Maryland.
Bergstrom, S. (1976) Development and application of a conceptual
   runoff model for Scandinavian catchments. Dept of Water
   Resources Engineering,  Lund Institute of Technology/University            of
   Lund, Bulletin  Series A no.52.
                      Simulation   of snowmelt   runoff   in Switzerland   139

Braun, L.N. (1985) Simulation of snowmelt-runoff in lowland and
   lower alpine regions of Switzerland. Zuricher              Geographische
   Schriften     21, Geogr. Inst, der Eidg. Tech. Hochschule, Zurich.
Braun, L.N. & Zuidema, P.K. (1982) Modelling snowmelt during
   advection-melt situations in a small basin (Rietholzbach). In:
   Hydrological     Research Basins and their Use in Water         Resources
   Planning (Sonderheft Landeshydrologie, Bern), vol.3, 771-780.
Cavadias, G. & Morin, G. (1985) Approximate confidence intervals
   for numerical verification criteria used in hydrological models.
   Application to the WMO intercomparison of conceptual models of
   snowmelt runoff. Preliminary Report, Dept of Civil Engineering
   and Applied Mechanics, McGill University, Montreal, and
   Université du Québec, INRS-Eau, Sainte-Poy, Canada.
                                        s
Charbonneau, R., Lardeau, J.-P. 8 Obled, C. (1981) Problems of
   modelling a high mountainous drainage basin with predominant
   snow yields. Hydrol. Sci. Bull. 26(4), 345-361.
Flemming, G. & Grutz, J. (1983) Berechnung der Schmelzwasserabgabe
   aus der Schneedecke in Einzugsgebieten des Mittelgebirges
   (computation of meltwater from snow cover in basins of the
   Mittelgebirge). Wasserwirtschaft           - Wassertechnik  2, 64-66.
Germann, P., Schadler, B. & Meier, R. (1982) The variation of some
   hydrological elements within a préalpine basin. In:
   Hydrological     Research Basins and their Use in Water          Resources
   Planning     (Sonderheft Landeshydrologie, Bern), vol.2, 301-310.
Grebner, D. (1978) Zur Berechnung der Hohenlage des Ueberganges
   vom festen in fliissigen Niederschlag (On computing the altitude
   of the transition of solid to liquid precipitation).                Arbeiten
   aus der Zentralanstalt      fur Météorologie     und Geodynamik, Heft
   31(22), Wien, 1-6.
Grebner, D. & Goldi, Chr. (1983) Niederschlagskarten fur den
   Kanton Zurich (Precipitation maps for the canton of Zurich).
   Schw. Ing. Arch. 48, 1141-1150.
Jensen, H. (1980) Abflussvorhersagen in der Schweiz. Operationelle
   Wasserstands- und Abfluss-Vorhersagen (Runoff forecasts in
   Switzerland). Schriftenreihe         des Deutschen Verbandes      fur
   Wasserwirtschaft      und Kulturbau,    Heft 51, 127-140.
Lang, H. (1978) Untersuchungen iiber den Wasserhaushalt and îlber den
   Abflussprozess im hydrologischen Einzugsgebiet Rietholzbach
   (Investigations on the water balance and on the runoff processes
   in the Rietholzbach basin). 1. Tagungsbericht "Forschung in
   Mitteleuropaischen Nationalparken". Schriftenreihe             des Bayer.
   Staatsministeriums      fur Ernahrung, Landwirtschaft      und    Forsten,
   108-117.
Martinec, J. (1970) Study of snowmelt runoff process in two
   representative watersheds with different elevation range. In:
   Symposium on the Results       of Research in Representative       and
   Experimental     Basins (Wellington, 1970), 29-39. IAHS Publ. no.
   96.
Maurer, R., Kunz, S. & Witmer, U. (1975) Niederschlag, Hagel, Schnee.
   Die Niederschlagsverhàltnisse in der Region Bern. Beitrage               zum
   Klima der Region Bern, Beitr.        No.4.    Geographisches Inst, der
   Universitat Bern.
Moore, R.D. & Owens, I.F. (1984) Controls on advective snowmelt in
   a maritime alpine basin. J. Climate Appl. Met. 23, 135-142.
140 L.N.Braun   & H.Lang

Nash, J.E. & Sutcliffe, J.V. (1970) River flow forecasting through
   conceptual models. Part I - A discussion of principles.
   J. Hydrol. 10C3), 282-290.
Obled, Ch. & Bosse, B. (1977) Mathematical models of a melting snow
   pack at an index point. J. Hydrol.         32(1/2), 139-163.
Price, A.G. & Dunne, T. (1976) Energy balance computations of
   snowmelt runoff in a subarctic area. Wat. Resour. Res. 12,
   686-694.
Rosl, G. (.1976) Mathematisches Modell zur Berechnung des Abflusses
   aus einem schneebedeckten Einzugsgebiet (Mathematical model for
   the calculation of runoff from a snow covered basin).
   Vortragsvexanstaltung    vom 12.6.1976.     SFB 81,   Technische
   Unlversitat    Munchen, 63-94.
Vischer, D. (1978) Abflussprognosen fiir den Rhein in Rheinfelden
   CRunoff forecasts for the Rhine River in Rheinfelden). In:
   Die Rheinwasserstrasse. Beitr.       Geol. Schweiz -     Hydrologie,
   Nr. 24, 35-48.
WMO C1982) WMO project for the intercomparison of conceptual
   models of snowmelt runoff.   In: Hydrological   Aspects   of   Alpine
   and High-Mountain Areas (Proc. Exeter Symp., July 1982),
   193-202. IAHS Publ. no.138.
Zeller, J. & Gensler, G. (.1980) Starkregenprofile durch die
   ostlichen Schweizer Alpen (Profiles of heavy precipitation
   through the eastern Alps). Interpraevent   1980 (Bad Ischl),
   Bd 1, 203-213.

								
To top