International Commision for the Hydrology of the Rhine basin - logo

Secretariat: PO Box 2232 | 3500 GE Utrecht | The Netherlands | T +31 887 985 233 | info@chr-khr.org

You are here

The snow and glacier melt components of the streamflow of the River Rhine and its tributaries considering the influence of climate change

00-
03
Subtitle
Final report to the International Commission for the Hydrology of the Rhine basin (CHR)
Author
Stahl, K.; Weiler, M.; Freudiger, D.; Kohn, I.; Seibert, J.; Vis,M.; Gerlinger,K.; Böhm, M.;
Date of publication
2017
Summary

In the year 2012, the International Commission for the Hydrology of the Rhine basin (KHR/CHR)
initiated the research project “The snow and glacier melt components of the streamflow of the River
Rhine and its tributaries considering the influence of climate change” (ASG Rhine). The objective of
the project was to determine the daily and monthly fractions of rain, snowmelt, and glacier ice melt
that comprised the streamflow of the River Rhine during the study period 1901–2006. One particular
interest was the analysis of the flow composition during extreme low flow conditions in the mid and
lower reaches of the River Rhine based on the modelled daily streamflow components. The long study
period may elucidate trends and changes in streamflow as a result of a combined influence of climate
change, glacier retreat, and an increase in reservoir storage and regulation during the 20th Century. The
governing idea for the study was that all analyses and modelling should be based on as observed data
as much as possible.

Hence, the study first assembled a long-term dataset of water cycle observations and gridded (interpolated)
data products for the Rhine basin. Amongst these, the HYRAS data, a gridded dataset of interpolated
meteorological variables provided by the German Weather Service (DWD) and the Federal
Institute of Hydrology (BfG) provided an important input. However, as HYRAS only covers the period
1951–2006, a first important development in the project was a climate reconstruction for the early
part of the study period 1901–1950. This HYRAS-REC dataset was obtained by resampling daily
grids from HYRAS based on daily weather analogues determined from a selection of available longterm
data records in the study area. The combination of HYRAS-REC and HYRAS then provided
consistent meteorological input to the project’s hydrological modelling experiments.

Besides the meteorological variables, the project’s dataset of observations also includes long time
series of streamflow along the Rhine River and its tributaries. In particular, all available streamflow
records from smaller scale glacierized headwater catchments in the Alps were included, even if the
measurements only covered a few years within the study period. Further data assembled include snow
water equivalent information, glacier length, glacier area, and thickness distributions at various times
in the past. While many of these data were made available to the project from other researchers or
from the agencies, glacier extents at the start of the study period had to be digitized manually from the
Swiss “Siegfried maps”. This additional layer of areal extent of glaciers in the early 1900s then provided
the initial conditions for the hydrological modelling over the entire 20th Century.

Before the modelling, the project carried out a statistical analysis of climate sensitivity and trends
of streamflow based on a set of 25 long-term streamflow observations from alpine catchments across
Switzerland. Most of the selected catchments are relatively unregulated and partially glacierized, but
some were selected to serve as non-glacierized references. The empirical analysis investigated these
long-term streamflow records together with long-term climate data observations for different subperiods
of the study period. The climate sensitivity, which was defined by the contribution of climate variables
to the variability of weekly flows, was calculated by regression models. The resulting coefficients
from all 25 records showed clear relations to catchment glacier coverage and mean catchment
elevation. Below an elevation of ca. 1000–1500 m a.s.l. flow variability in winter is strongly dominated
by precipitation, whereas above 1500–2000 m a.s.l. flow variability in summer is driven by temperature.
The sign of the temperature control depends on the presence of a glacier, i.e. a dominance of
evapotranspiration processes reduces flow with temperature, while the dominance of ice melt increases
flow with temperature.

The empirical analyses showed how important a correct description of runoff generation mechanisms
is particularly in mountain headwater catchments. Hence, in the setup of hydrological models for the
project particular attention was paid to the determination of climate variable lapse rates as well as to
basin glacier coverage and its changes with time. The analysis further showed a negative trend for July
and August in streamflow for the last few decades in several basins with small glacier coverage. These
trends suggest that the temporary increase of ice melt that is generally expected in a warming climate
may have already passed its peak.

Considering these findings, a model chain was developed for the Rhine basin. Glacierized headwater
catchments were modelled with the semi-distributed conceptual HBV-light model that allows a coupling
of rainfall-runoff hydrology with glacier mass balance and glacier change simulation. The entire
Rhine basin was then modelled with different distributed LARSIM water balance models ("LARSIMHochrhein"
at 1km resolution and "LARSIM-ME-Rhein" at 5km resolution).

All models required substantial adaptations to be suitable for the project’s challenge of modelling the
long time period of over 100 years. A particularly interesting modelling task was the accounting of
the fractions of rain, snowmelt and glacier ice melt from runoff generation (input) through the conceptualized
hydrological modelling system into the modelled output (streamflow). Several options to
account for and follow these runoff contributions through the various models’ soil and groundwater
storages, lakes, and along the river network were developed and tested. Finally, a method with storage-
corresponding mixing tanks was employed. In each spatial model element, all input fluxes are
perfectly mixed with previously present components in a storage-corresponding mixing tank and then
passed on to the next model unit according to the unit’s water balance. The method can be implemented
separately for each model store (soil, groundwater, lakes) or as one integrative mixing tank per
spatial model unit (LARSIM grid box at the respective resolution). The mixing tanks hence serve as a
model run parallel to the hydrological model in order to calculate the mixing of the input components
with the previously stored mix and release the components accordingly. Subsequent the LARSIM
models then separately routes these components along the river network.

As a result of a perfect and complete mixing in the Rhine’s rather large lakes and groundwater reservoirs,
however, flow components became almost constant year-round. Therefore, the mixing tanks’
capacities were limited compared to the hydrological models’ maximum storage capacities. This step
was necessary for the models to allow the study to answer the question of quantitative effects of
changes in the input components, e.g. a reduced snow and ice melt input, on the streamflow response
considering runoff generation, runoff concentration and routing in the basin. Sensitivity analyses
showed that the limited mixing tank approach can achieve this objective.

Another requirement for the long-term modelling was to correctly simulate the coupled glacier
change and streamflow change. To avoid any long-term accumulation of snow outside real glacier
accumulation areas, a new snow redistribution method was implemented in HBV-light. The glacier
area adjustment with the so-called “delta-h-Method” had to be modified to account for glacier growth
in the early 20th Century besides the commonly modelled retreat of glaciers thereafter. Further model
developments include a seasonally varying melt factor and glacier reservoir outflow coefficient.
Finally, HBV-light’s multiple criteria model calibration features were extended. The objective function
that was specifically designed for the project weighs the differences between modelled and observed
snow water equivalent and snow cover area, glacier volume change between available observations,
and streamflow dynamics at various temporal scales. For the modelling of ungauged basins, a
parameter regionalisation was combined with a recalibration to available snow and glacier observations.
The automatic calibration with many parameter sets in HBV-light allowed the determination of
the range of parameter uncertainty. In the end, only the results of the ‘best’ simulation, i.e., the rain,
snowmelt, and ice melt component of streamflow were eventually transferred from the modelled
headwater basin as lateral input into the LARSIM model grid.

Modifications in the LARSIM models also concerned the snow routine, as well as an additional elevation
discretization scheme in the lower resolution LARSIM-ME model, to improve snow modelling
and other elevation-dependent processes. The increasing reservoir storage for hydropower generation
throughout the 20th century was incorporated through the implementation of lumped reservoirs, each
representing a number of smaller reservoirs upstream. Reservoir capacities change throughout the
modelling period according to a generalized timeline of historical hydropower development. The regulation
of several pre-alpine lakes is also considered in the model.

Modelled streamflow components were then assessed at three different spatial scales: the glacierized
headwater catchments of the Rhine basin, modelled by HBV-light, the basin of the Alpine Rhine and
High Rhine upstream of the gauging station in Basel (CH), modelled by LARSIM-Hochrhein, and the
Rhine basin below Basel (without the delta region) down to the gauging station of Lobith (NL), modelled
by LARSIM-ME.

The “glacierized headwater catchments” hereby represent a model and project-specific delineation
of high-elevation, subbasins with considerable glacier cover at the start of the study period. The total
area covered by these headwater basins is only 4152km2. However, an analysis at this scale, which
corresponds to the hydrological meso-scale, allows a comparison with many other studies at similar
scales. In addition, it is the scale where glacier melt water contributions are locally important for hydropower
generation.

For the entire period 1900–2006, modelled ice melt amounts to a mean fraction of about one tenth of
the average annual flow from the aggregated streamflow of all glacierized headwater catchments. In
August and September the mean ice melt component is one fifth of the total flow. During warm summers,
these fractions have been much higher, for example, in the years of 1921 and 1947 or more recently
in 2003. In these years, monthly means of ice melt flow for all the HBV-modelled glacierized
headwater catchments together amounted to 160–195 m3/s, which corresponded to fractions of up to
50%. When streamflow in such years receded strongly during late summer, the relative contributions
of ice melt were higher in September than in August. This result differs from previous studies and can
be attributed to the more accurate runoff generation and routing considered in this study.
The separate analysis of the glacierized headwater catchment scale also allowed the spatial mapping of
the ice melt contributions within the Rhine basin as well as temporal changes in that contribution.
Interestingly, over the long-term period modelled streamflow components indicate little change. The
reason glacier retreat is compensated by increased glacier melt. According to the modelled time series,
the ice meltwater production at the end of the period requires nearly a doubling of the specific negative
glacier mass balance compared to the beginning of the period. It should be noted, however, that
the last few years of the modelling period (ca. from 2000 to 2006) showed an increase in the ice melt
component of streamflow. Regarding the entire Rhine basin there is no conclusion yet on when the ice
melt component of streamflow will finally decline.

The LARSIM models then propagated and modelled the streamflow components downstream. The
resulting long-term mean of the modelled annual fractions of the ice melt component were estimated
at approx. 2% of the streamflow in Basel and approx. 1% of the streamflow in Lobith. These numbers
agree well with estimates in previous modelling studies. The mean annual fractions of the snowmelt
component are much higher and thus more important. They were estimated at 39% of the streamflow
in Basel and at 34% in Lobith. Thanks to the daily resolution of the modelling, for the first time, this
study could quantify specifically the ice melt component during extreme low flow years such as 1921,
1947, 2003. These events were characterised by spatially coherent summer heat and drought conditions
in western and central Europe affecting the Rhine from the alpine headwaters all the way down
to the mouth.

The summer peak of the glacier melt season coincides with or closely follows the overall peak of the
annual streamflow regimes upstream of Basel, whereas the downstream reaches of the Rhine have
hydrological regimes with late summer to autumn low flows. Hence, in the lower Rhine this coincidence
of the peak glacier melt with the time of recession to lower flows results in relatively high fractions
of ice melt from August to October. While the long-term mean of the ice melt component in
August and September was estimated at 4.5%/6% in Basel and 2.6%/4.2% in Lobith, maximum daily
fractions in the aforementioned record years were much higher. In the hot summer of 2003, for example,
the maxima of the ice melt component were a quarter in Basel and still 17% in Lobith.

Hydrological modelling always has uncertainties stemming from a number of sources from measurement
errors to parameter uncertainties. The project considered in particular model parameter uncertainty
but concluded that it did not affect the main conclusions of the study. This can mainly be attributed
to the consideration of the observations of glacier changes and the dynamics of snow cover
and streamflow, which effectively helped to constrain the models. Therefore, the model chain developed
in the project provides a reliable tool to follow the effect of runoff contributions from rain,
snowmelt, and ice melt through the basin and to determine the consequent changes in the respective
streamflow components. This tool also provides the potential to test scenarios of future climate change
and water use and regulation.

Table of contents

List of abbreviations ............................................................................................................................................ 3
Summary ............................................................................................................................................................. 7
1 Background and objectives ...................................................................................................................... 11
1.1 Streamflow components and climate variability ................................................................................. 11
1.2 Data, analyses, and models: Outline of the project and the report .................................................... 13
2 Reconstruction of daily climate input grids for the period 1901–1950 .................................................... 16
2.1 Objectives ............................................................................................................................................ 16
2.2 Analogue method ................................................................................................................................ 16
2.3 Results ................................................................................................................................................. 19
2.4 Compilation and validation of climate input datasets 1901–2006 ...................................................... 24
2.5 Conclusions for the use as input data for hydrological models ........................................................... 25
3 Alpine catchments: statistical data analyses ........................................................................................... 27
3.1 Introduction and objectives ................................................................................................................. 27
3.2 Seasonal climate sensitivity of alpine streamflow ............................................................................... 27
3.3 Streamflow trends ............................................................................................................................... 33
3.4 Glacier coverage in the headwater catchments of the River Rhine .................................................... 35
3.5 Conclusions for the modelling of the glacierized headwater catchments .......................................... 38
4 Streamflow components: definitions and modelling ............................................................................... 41
4.1 Background and objectives .................................................................................................................. 41
4.2 Methodical approaches for quantifying streamflow components ...................................................... 43
4.3 Application and intercomparison ........................................................................................................ 45
4.4 Conclusions for the realisation within the project modelling framework ........................................... 51
5 Modelling of the glacierized headwater catchments ............................................................................... 52
5.1 Objectives ............................................................................................................................................ 52
5.2 Delineation and model setup .............................................................................................................. 52
5.3 Specific developments for glacierized headwater catchment models ................................................ 54
5.4 Model application and multi‐criteria‐calibration ................................................................................ 58
5.5 Modelled components of streamflow (1901–2006) ............................................................................ 66
5.6 Modelled streamflow components in selected extreme years ........................................................... 70
5.7 Long‐term behaviour of modelled streamflow components .............................................................. 79
5.8 Conclusions .......................................................................................................................................... 85
6 Modelling of the entire Rhine basin ......................................................................................................... 87
6.1 Objectives and description of the LARSIM model ............................................................................... 87
6.2 Model adaptations and extensions ..................................................................................................... 91
6.3 Modelled components of streamflow (1901–2006) .......................................................................... 100
6.4 Modelled streamflow components in low flow years ....................................................................... 108
6.5 Temporal changes of streamflow components ................................................................................. 119
6.6 Conclusions ........................................................................................................................................ 121
7 Model validation and uncertainty analyses ........................................................................................... 123
7.1 Validation of selected modelling results ........................................................................................... 123
7.2 Model parameter uncertainty ........................................................................................................... 128
7.3 Conclusions ........................................................................................................................................ 136
8 Conclusions from the project considering climate change ..................................................................... 137
References ....................................................................................................................................................... 139
Acknowledgement ........................................................................................................................................... 145