Light absorbing impurities in snow and ice (LAISI) originating from atmospheric deposition enhance snowmelt by increasing the absorption of shortwave radiation. The consequences are a shortening of the snow duration due to increased snowmelt and, at the catchment scale, a temporal shift in the discharge generation during the spring melt season.
In this study, we present a newly developed snow algorithm for application in hydrological models that allows for an additional class of input variable: the deposition mass flux of various species of light absorbing aerosols. To show the sensitivity of different model parameters, we first use the model as a 1-D point model forced with representative synthetic data and investigate the impact of parameters and variables specific to the algorithm determining the effect of LAISI. We then demonstrate the significance of the radiative forcing by simulating the effect of black carbon (BC) deposited on snow of a remote southern Norwegian catchment over a 6-year period, from September 2006 to August 2012. Our simulations suggest a significant impact of BC in snow on the hydrological cycle. Results show an average increase in discharge of 2.5, 9.9, and 21.4 %, depending on the applied model scenario, over a 2-month period during the spring melt season compared to simulations where radiative forcing from LAISI is not considered. The increase in discharge is followed by a decrease in discharge due to a faster decrease in the catchment's snow-covered fraction and a trend towards earlier melt in the scenarios where radiative forcing from LAISI is applied. Using a reasonable estimate of critical model parameters, the model simulates realistic BC mixing ratios in surface snow with a strong annual cycle, showing increasing surface BC mixing ratios during spring melt as a consequence of melt amplification. However, we further identify large uncertainties in the representation of the surface BC mixing ratio during snowmelt and the subsequent consequences for the snowpack evolution.
The representation of the seasonal snowpack is of outstanding importance in
hydrological models aiming for application in cold or mountainous
environments. In many mountain regions, the seasonal snowpack constitutes a
major portion of the water budget, contributing up to 50 %, and even
more, to the annual discharge
As LAISI lower the snow albedo, the effect on the snowmelt has the potential
to alter the hydrological characteristics of catchments where snowmelt
significantly contributes to the water budget. Recent research investigates
the impact of LAISI on discharge generation in mountain regions at different
scales.
Only a few studies have developed model approaches to resolve the impact of
LAISI on snowmelt discharge generation at the catchment scale.
Despite these efforts, the direct integration of deposition mass fluxes of
light absorbing aerosols in a catchment model is still lacking. To date,
there is no rainfall–runoff model with a focus on runoff forecast at the
catchment scale that is able to consider aerosol deposition mass fluxes
alongside snowfall. On the other hand, there is evidence that including the
radiative forcing of LAISI has the potential to improve the quality of
hydrological predictions:
In this study, we address this deficiency by introducing a rainfall–runoff model with a newly developed snow algorithm that allows for a new class of model input variable: the deposition mass flux of different species of light absorbing aerosols. The model integrates snowpack dynamics forced by LAISI and allows for analysis at the catchment scale. The algorithm uses a radiative transfer model for snow to account dynamically for the impact of LAISI on the snow albedo and the subsequent impacts on the snowmelt and discharge generation. Aside from enabling the user to optionally apply deposition mass fluxes as model input, the algorithm depends on standard atmospheric input variables (precipitation, temperature, shortwave radiation, wind speed, and relative humidity). To enable a critical evaluation of the newly developed snowpack algorithm, we conduct two independent analyses: (i) a 1-D sensitivity study of critical model parameters, and (ii) a catchment-scale analysis of the impact of LAISI. In both analyses we use BC in snow from wet and dry deposition as a proxy for the impact of LAISI.
We first present an overview of the hydrological model used in this study and
our algorithm to treat LAISI in Sect.
In the following section we provide descriptions of the hydrologic model
(Sect.
For the analysis, we use Statkraft's hydrologic forecasting toolbox (Shyft;
The methods used to simulate hydrological processes are (i) a single-equation
implementation to determine the potential evapotranspiration, (ii) a newly
developed snowpack algorithm using an online radiative transfer solution for
snow to account for the effect of LAISI on the snow albedo, and (iii) a
first-order nonlinear differential equation to calculate the catchment
response to precipitation, snowmelt, and evapotranspiration. (i) and
(iii) are described in more detail herein, while (ii) is described in detail
in Sect.
To determine the potential evapotranspiration,
The catchment response to precipitation and snowmelt is determined using the
approach of
We assume that the sensitivity function,
To account for snow in the model, we developed a snow algorithm to solve the
energy balance:
The main addition provided in the algorithm described herein is the
implementation of a radiative transfer solution for the dynamical calculation
of snow albedo,
Below, we introduce the radiative transfer calculations required to represent
LAISI (Sect.
Both the albedo of clean snow and the effect of LAISI on the snow albedo
strongly depend on the snow optical grain radius
In our model, we compute the evolution of
To solve for the effect of light absorption from LAISI on the snow albedo, we
have integrated a two-layer adaption of the Snow, Ice, and Aerosol Radiative
(SNICAR) model
The absorbing effect of LAISI is most efficient when the LAISI reside at or
close to the snow surface
To represent the evolution of LAISI mixing ratios near the snow surface, we
treat LAISI in two layers in our model. The surface layer has a
time-invariant maximum thickness (further called the maximum surface layer
thickness). The mixing ratio of each LAISI species in this layer is
calculated from a uniform mixing of the layer's snow with either falling snow
with a certain mixing ratio of aerosol (wet deposition) or aerosol from
atmospheric dry deposition. The second layer (bottom layer) represents the
snow exceeding the maximum thickness of the surface layer. Following
To allow for melt amplification in the model, we include LAISI mass fluxes
between the two layers during snow accumulation and snowmelt. Generalizing
In order to allow for explicit treatment of snow layers while representing
sub-grid snow variability, we follow
We selected the unregulated upper Atna catchment for our analysis. The
catchment is located in a high-elevation region of southern Norway
(Fig.
For the meteorological model input of precipitation, temperature, relative
humidity, and wind speed we use daily observations from the Norwegian Water
Resources and Energy Directorate (NVE) and the Norwegian Meteorological
Institute (MET). Four meteorological stations are located in the watershed at
elevations between 701 and 780 m a.s.l. along the Atna river, two of these
measuring precipitation and two measuring temperature. Observations of
relative humidity and wind speed originate from two stations at locations
close by the catchment (not shown in Fig.
Location of the Atnsjoen catchment in Norway (black box in the left map) and overview map of the Atnsjoen catchment (right).
Information about observational stations.
The wet and dry deposition
rates of BC for the study area are generated using the REMO-HAM regional
aerosol-climate model
For the simulations, we follow the approach of
REMO-HAM was used for the same European domain as in
Model parameters used in the sensitivity and case study. Parameters
optimized during calibration are marked with
Our analysis is conducted in two parts. First, in a 1-D sensitivity study, we
investigate the sensitivity of parameters and variables specific to the LAISI
algorithm presented in Sect.
We assume uncertainties of the LAISI radiative forcing in snow to originate
mainly from the model representation of surface layer thickness, melt
scavenging of BC, and uncertainties in the deposition input data. To account
for the uncertainties, we declare minimum (min), central (mid), and maximum
(max) effect estimates for each of the critical parameters, outlined together
with further model parameters in Table
The results of the 1-D sensitivity
study are presented in Sect.
Our approach evaluates these parameters and the evolution of the snowpack
under constant melting conditions. We run the 1-D simulations with model
parameters as outlined in Table
To investigate the impact of the maximum surface layer thickness (parameter
To explore the sensitivity to scavenging ratio, we apply different BC
scavenging ratios in the range of the uncertainty of hydrophilic BC, which
covers a wide range from very efficient to inefficient scavenging. The
scavenging ratios applied are based on the analysis conducted by
Hydrophilic BC absorbs more strongly than hydrophobic BC under the same
conditions due to an increased MAC for hydrophilic BC resulting from ageing
of the aerosol during atmospheric transport
We investigate the impact of BC aerosol deposition on the catchment hydrology
of a Norwegian catchment over a study period of 6 years, from September 2006
to August 2012. The station-based input data described above
(Sect.
To calibrate the model against observed discharge, we first run a
split-sample calibration
Model calibration is run with mid estimates for all model parameters
impacting the handling and effect of LAISI and aerosol depositions as
simulated from REMO-HAM during model calibration. Those parameters and
further model parameters, including the parameters estimated during
calibration, are listed in the left column of
Table
Snow albedo (top row of graphs; solid lines) and melt rate (top row
of graphs; dashed lines), BC mixing ratio in the surface layer and factor
increase in the mixing ratio during melt compared to the pre-melt BC mixing
ratio (central row of graphs), and snowpack SWE (bottom row of graphs) for
simulations forced with synthetic data based on average meteorological
conditions during the melt season from mid March until mid July of the
Atnsjoen catchment and different model configurations:
Figure
In the range of investigated scavenging ratios, we find the sensitivity of
the surface BC mixing ratio, the albedo, and the subsequent snowmelt to this
parameter (Fig.
The changes in the scavenging ratio lead to a considerable effect on albedo and snowmelt. Meltout is delayed by circa 0.5 (purple lines), 3 (red lines), and 8 days (yellow lines) for scavenging ratios of 0.02, 0.2, and 2.0, respectively, compared to no scavenging (green lines). Compared to the no-ARF experiment (black lines), our simulations show that the presence of BC causes an earlier meltout of about 9.5, 7, and 2 days for scavenging ratios of 0.02, 0.2, and 2.0, respectively.
The column of graphs in Fig.
The isolated effect of the stronger absorption of hydrophilic BC leads to an
earlier meltout by circa 2 days compared to hydrophobic BC (purple and green
curves in graphs of Fig.
In the split-sample test, the model performance is acceptable during both
calibration and validation, with Nash–Sutcliffe model efficiencies of 0.86
during the calibration period (green line in
Fig.
Simulated (green and red curves) and observed (black curve) daily
discharge from the Atnsjoen catchment.
Comparison of observed and simulated daily discharge
For the min and mid estimate, the model simulates an average annual surface
BC mixing ratio of about 18 and 71 ng g
With the start of the melt season, the difference in albedo between model
experiments becomes increasingly larger over time. During the melt season,
the mid estimate spatially averaged surface BC mixing ratio increases from 49
to about 250 ng g
The radiative forcing in snow (RFS) induced by the presence of BC is
calculated from the average radiative forcing over snow-bearing tiles only.
The RFS represents the additional uptake of energy from solar radiation per
area snow cover due to the presence of BC in the snow compared to clean snow
with the same properties. Figure
Catchment snow-covered fraction (SCF; dashed lines),
However, most relevant for discharge generation (see
Sect.
Figure
Average change in discharge during the early (22 March to 29 May) and late (30 May to 10 August) melt seasons of min, mid, and max estimates and average change in SWE during the melt season (22 March to 10 August) compared to the no-ARF scenario.
Min, mid, and max estimates all show the change from higher to lower
discharge compared to the no-ARF scenario approximately at the same time (at
the end of May; see the blue marker in Fig.
The differences in discharge among the scenarios can be explained by
understanding the evolution of the snowpack. In all scenarios the catchment
SWE (Fig.
The objective of this work is to provide a mechanism to assess the impact of
light absorbing aerosols on runoff at the catchment scale in a
rainfall–runoff modelling context. Prior investigations into LAISI indicate
potentially significant impacts on the cryosphere
To assess the sensitivity of the newly introduced algorithm and parameters, we conducted a sequence of 1-D sensitivity studies. In this context, we are able to remove complexities that arise when conducting distributed simulations at the catchment scale.
We found the greatest sensitivity to lie in the parametrization of
scavenging, as it relates to how likely the aerosol is to remain at the snow
surface during melt. Field measurements indicate that only a fraction of BC
is flushed out with the meltwater and BC can accumulate near the snow surface
Further complicating the effect is the fact that hydrophilic BC (which
undergoes more efficient melt scavenging) has a larger MAC (enhanced
absorption) compared to hydrophobic BC
The 1-D model experiments further show that the definition of at least two
layers in the snowpack model is important to allow for accumulation of
impurities at the snow surface. This result in itself is not original:
numerous prior studies have identified the importance of having multiple
layers
Observations of BC in melting snow support the accumulation of BC near the
surface
We argue therefore the importance of providing, at a minimum, a separate
surface layer, but recognize that simulated surface mixing ratios of BC are
highly sensitive to the thickness of this layer. Since evaluation of model
predictions for BC in snow is commonly performed by comparing simulated with
observed BC mixing ratios in surface snow
We are interested in addressing the impact of BC deposition – and
potentially other light absorbing aerosols – on the hydrology of snowfall
dominated catchments. Studies have shown the potential impact LAISI may have
on the timing of snowmelt
The impact resulting from BC deposition in our study is seen in the timing of
the annual water balance. Inclusion of ARF generally increases early season
melt and causes the snowpack to melt out earlier. Comparing the ARF and
no-ARF scenarios, we see a general shift in the discharge, with the ARF
scenario producing greater discharge early in the season, and having less
discharge after June. Such a shift in seasonal water balance will potentially
have impacts on soil moisture and agriculture
Albedo is a critical parameter in any snowmelt model, with significant
control over the energy balance. During the accumulation period, the average
albedo of each scenario lies within the range of albedo of fresh snow with
small optical grain radius combined with a high solar zenith angle
At the end of the melt season, the evolution of surface
BC yields reductions in albedo relative to the no-ARF case of about 0.03,
0.1, and over 0.3 for the min, mid, and max estimates, respectively. This has
two reasons: first, with increasing grain radius during the melt season, the
absorbing effect of BC gets more efficient due to deeper penetration of
radiation into the snowpack, leading to a stronger effect of the BC
deposition on albedo. Snow of larger grains has a larger extinction
coefficient and more effective forward scattering properties
For the max estimate, the increase is from roughly 100 to over
2500 ng g
We recognize our max estimate results in a strong increase in surface BC
mixing ratios mostly due to low BC scavenging with melt (note the strong
increase from the end of March on in Fig.
The strong increase in RFS (Fig.
The annual mean surface radiative forcings in this study are 0.284, 0.844,
and 1.391 W m
We mention a shift in the seasonal water balance, with more melt early in the
melt season resulting from enhanced RFS. However, from mid May the melt
enhancement reduces and the differences in catchment SWE between the ARF and
no-ARF scenarios decrease (Fig.
Similar shifts in the annual water balance due to the impact from LAISI are
reported for the Upper Colorado River Basin
Compared to observations, all simulations (ARF and no-ARF) tend to
underestimate discharge during early melt season and overestimate discharge
during late melt season (Fig.
Season mean volume error in discharge during the early (22 March to
29 May) and late (30 May to 10 August) melt seasons of the no-ARF, min, mid,
and max scenarios compared to observed discharge. The percentage change shows
an increase (
There are numerous challenges associated with the development of an algorithm
that mixes conceptual hydrologic parametrizations with physically based
approaches. Both the literature and our analysis highlight aspects that
warrant a deeper investigation of ARF-induced uncertainty. The intent with
this work is to introduce a new algorithm; however, as indicated in
With respect to the implementation of a physical albedo model, the treatment
of the darkening effect of LAISI adds additional degrees of freedom to the
parameter space of the model due to the introduction of new parameters
(scavenging ratios, surface layer thickness, BC input scaling factor; see the
bottom four parameters in Table
The 95 % confidence interval of simulated discharge due to parameter uncertainty when allowing for ARF (red) and disregarding ARF (grey), calculated using the GLUE method and averaged over the 6-year simulation period. The shaded box marks the period of the melt season, where observations tend to lie outside the uncertainty bounds of the no-ARF simulations.
In our case study, further uncertainties result from mixing ratios of BC in
the snowpack due to prescribed BC deposition, and LAISI other than BC not
accounted for in the simulations:
prescribed BC deposition In the approach presented here, we use prescribed BC deposition mass fluxes.
Even though this is common practice LAISI other than BC By including only BC deposition in our simulation, we likely underestimate
the additional effect of further LAISI species such as mineral dust
With respect to our study, we acknowledge that including only BC is a
shortcoming with respect to the overall effect of LAISI. However, by
demonstrating the significant effect of BC on accelerating snowmelt and
discharge generation, our study gives a conservative estimate of the effect
of LAISI and urges a more detailed investigation.
Herein we presented a newly developed snow algorithm for application in
hydrologic models that allows a new class of model input variable: the
deposition rates of light absorbing aerosols. By coupling a radiative
transfer model for snow to an energy-balance-based snowpack model, we are
providing a tool that can be used to determine the effect of various species
of LAISI at the catchment scale. In this analysis we have focused solely on
BC and acknowledge it therefore likely represents a conservative estimate.
This work presents a novel analysis of the impact of BC deposition to snow on
the hydrologic cycle through 1-D sensitivity studies and catchment-scale
hydrologic modelling. From a 1-D model study, presented in
Sect. the implementation of at least two layers (a thin surface layer and a
bottom layer) is of outstanding importance to capture the potential effect of
melt amplification on the near-surface LAISI evolution. The parametrization
of the surface layer has only a small effect on the snow albedo and melt rate
as long as the surface layer thickness (in SWE) is sufficiently thin (e.g.
thinner than the penetration depth of shortwave radiation). However, the
evolution of the LAISI surface mixing ratio is highly sensitive to the
surface layer thickness. For this reason, we suggest including a surface
layer thickness variation in model studies when comparing simulated and
observed LAISI mixing ratios sampled in the top few centimetres of snow. The determination of how LAISI are washed out of the snowpack
with meltwater has a great effect on the evolution of LAISI concentration
near the surface, snow albedo, and melt rate. Due to rare observations of
this effect under controlled conditions, the uncertainties are high and our
findings show the need for more detailed understanding of the processes
involved due to the high importance of the overall effect of LAISI on the
snowpack evolution.
To demonstrate the significance of BC radiative forcing for the hydrologic cycle at the catchment scale, we demonstrated the effect of BC deposition and the subsequent implications for snowmelt and discharge generation on a remote Norwegian mountain catchment. The study indicates that inclusion of BC in snow is likely to have a significant impact on melt timing, and that the effect on the discharge generation leads to a shift in the annual water balance. Our simulations further suggest that melt amplification can have severe implications for both the snowpack evolution and the discharge regime of a catchment, which means that the seasonal cycle of the surface BC mixing ratio is of great importance. However, large uncertainties are connected with the representation of surface enrichment of BC. A more robust understanding of the fate of BC in melting snow is essential to fully assess impacts on the hydrologic cycle.
Including radiative forcing from BC in the simulations leads to a reduction in volume error during the early and late melt season in our simulations. We conclude from our study that hydrological modelling can potentially be improved by including the effect of LAISI, especially when the model approach implies a physically based representation of the snowpack in general and the snow albedo in particular. However, more research in the area of catchment-scale impact of LAISI is needed to support this. The approach and algorithm presented in this analysis provide a tool to target this in future applications.
Meteorological observations for the Atnsjoen
catchment are provided by the Norwegian Meteorological Institute (MET) and
the Norwegian Water Resources and Energy Directorate (NVE). Observed
streamflow data are provided by NVE. The data are published in Matt (2018),
alongside gridded BC wet and dry deposition data and SHyFT model
configuration files. The source code of SHyFT is available on GitHub
(
In order to calculate radiative forcing in snow (RFS) from surface
concentrations during melt reported in snow optical grain radius: 500–1000 snow density: 400–600 kg m BC mixing ratio: 50–200 ng g
We determine parameter uncertainty using the GLUE method
Model parameter bounds used in the uncertainty estimation with the
GLUE method. Parameters used to determine ARF are marked with
The authors declare that they have no conflict of interest.
This work was conducted within the Norwegian Research Council's INDNOR programme under the Hydrologic sensitivity to Cryosphere-Aerosol interaction in Mountain Processes (HyCAMP) project (NFR no. 222195). We thank the Mitigation of Arctic warming by controlling European black carbon emissions (MACEB) project for their help concerning the REMO-HAM simulations. Furthermore, we thank the International Institute for Applied System Analysis (IIASA), especially Kaarle Kupiainen and Zbigniew Klimont, for providing the emissions data. Sigbjorn Helset and Statkraft AS, in general, have been vital resources in the development of the algorithm and, in particular, the implementation in Shyft. The ECHAM-HAMMOZ model is developed by a consortium composed of the ETH Zurich, the Max Planck Institut für Meteorologie, the Forschungszentrum Jülich, the University of Oxford, and the Finnish Meteorological Institute, and is managed by the Center for Climate Systems Modeling (C2SM) at the ETH Zurich. Edited by: Jan Seibert Reviewed by: three anonymous referees