H23B-0962
A Comprehensive Assessment of SNODAS for Hydrologic Forecasting
The NOAA/NOHRSC National Snow Analysis, otherwise known as the SNow Data ASsimilation (SNODAS) product, combines a high resolution modeling system with a wide variety of surface, airborne and satellite observations. The assimilation scheme is a mix of automated and manual analysis. However, evaluation of this product is problematic because there is little snow data available that was not used to produce the SNODAS product. Streamflow is one of the few measures that can give a true indication of how well snow accumulation and ablation is replicated in SNODAS. We use two basins in the Sierra Nevada mountains near the border of California and Nevada: the North Fork of the American River and the East Fork of the Carson. These two basins are the testbed for the NWS distributed model intercomparison project (DMIP) and are quite different in how snow impacts their hydrographs. We assess the water balance of the SNODAS product, compare it's snow estimates to some local observations of snow and look the streamflow results when the products are assimilated into a distributed hydrologic model. The presentation offers guidance for increasing the utility of SNODAS in the operational realm and suggests ways of making this product more effective.
H23B-0963
Non-stationarity and statistical partitioning of snow depth fields: Independent variable controls and application to sampling design and modeling.
High resolution LiDAR data sets of snow depth, vegetation height, and elevation from the NASA Cold Land Processes Experiment (CLPX) in Colorado have previously been examined to understand correlation length scales and for examining the relationship of snow depth to governing physical variables. Previous analyses have focused on mean scaling of snow depth over areas with extent of roughly 1 square km. This study investigates the spatial non-stationarity of snow depth over these same regions. The aim of the current analysis is to understand the prior scaling relationships in a spatially explicit way and to devise a practical, data-driven method for statistically partitioning snow depth variability in space. The role of various controlling physical processes in shaping snow depth non-stationarity and prediction of statistical partitioning from governing physical variables is investigated. Results are applicable to modeling and sampling design and are evaluated in the context of sampling design; can statistical partitioning of snow depth fields improve sampling efficiency?
H23B-0964
Development and Application of a Parsimonious Snow-Hydrologic Modeling Suite: Investigating the Link Between Model Complexity and Predictive Uncertainty
The simulation and modeling of snowmelt and hydrologic drivers is desirable for prediction of different hydrologic variables, most significantly streamflow at the catchment outlet. This is particularly true of mountainous regions where snowmelt drives major hydrologic events and water resource predictability. We have developed a suite of parsimonious models of first-order snow and hydrologic processes to investigate the link between overall model complexity (both snow and hydrologic elements) and predictive performance. The use of simper models is motivated by the desire to capture first-order processes, in line with a top-down modeling philosophy. Such models have the capability to be more efficient in modeling the system by having less uncertainty with similar predictive power when compared to more complex model structures. Constructed in a modular fashion, the modeling suite has the ability to assess the interaction between each snowmelt and hydrologic base structure coupling, as well as to separate error between each component. The modeling suite was applied to the Stringer Creek watershed of Tenderfoot Creek Experimental Forest (TCEF), located in central Montana, USA. Making use of meteorological data collected at one of the two NRCS SNOTEL stations within TCEF's borders and streamflow data from the USFS Rocky Mountain Research Station (TCEF's managing agency), we compare the performance of different model combinations using 6 years of available data. Implementation of a Markov chain Monte Carlo approach to parameter estimation and uncertainty estimation provides the ability to characterize errors in the models (including non-stationarities), explore complex parameter spaces and interdependence, and incorporate multiple sources of data for model conditioning. The necessity of such abilities becomes especially critical in the application of a top-down modeling approach, where conceptual models are used that often involve highly interdependent model parameters. Further, the flexibility and design of the coupled, modular framework allows for the separation of uncertainty with regard to both snow and hydrologic process components.
H23B-0965
Parameterizing Conductive Energy Fluxes and the Internal Energy Balance of a Snowpack: Inferences from Frequency Analysis of Snow Temperature Time Series
The snow surface temperature governs the exchange of energy between the snowpack and the atmosphere
and is in turn modulated by the conductive heat flux within the snowpack below. As a consequence,
parameterizations of heat conduction are important for snowmelt models that do not explicitly calculate
internal heat fluxes using finite difference or similar approaches. We compared three parameterizations of
heat conduction: equilibrium gradient, force-restore, and a modified force-restore. The modification
consisted of combining the force-restore for diurnal variability with the equilibrium gradient approach for lower
frequency variations in conduction. For the comparison, we inferred snowpack thermal properties from a
fourier analysis of vertically distributed temperature time series in the snowpack. The modified force-restore
parameterization could be calibrated, guided by the information on snow thermal properties, to match both
the snowpack energy content and conductive fluxes at the surface better than either the equilibrium gradient
or force-restore parameterizations alone. Analysis of the thermal data combined with ground heat flux
measurements revealed that a lag parameterization of the ground heat flux improved the agreement between
the modeled and observed ground heat flux for input to the complete model.
http://www.fs.fed.us/rm/boise/AWAE/scientists/profiles/AWALuce.shtml
H23B-0966
Incorporating Radiation Inputs into an Operational Snow Model: A Bayesian Model Averaging Approach
The primary snow accumulation and ablation model in the US National Weather Service operational streamflow prediction system is the temperature-based SNOW17. Future advancements in streamflow prediction systems will likely include ensemble prediction approaches and new data streams, which will require modifications to the current hydrologic forecasting system. We have created an energy balance version of the SNOW17 model (SNOW17-EB) which computes the snowpack heat exchange and melt based on observed radiation inputs, while the effect of wind, water retention and water release are parameterized according to the original SNOW17 structure. The SNOW17-EB will be used to test remotely sensed albedo and radiation values for snow monitoring and prediction in future studies. In the present study, three versions of the SNOW17-EB with varying albedo formulations and the original SNOW17 are applied within a Bayesian multi-model framework. Each model is calibrated via the Shuffle Complex Evolution (SCE) using three different objective functions, resulting in 12 models with varying parameters or structures. Bayesian Model Averaging (BMA) is used to compute model weights based on observed SWE, which are then combined to create an expected SWE simulation with an uncertainty estimate from the model ensemble. Development and testing is conducted for select SNOTEL sites in the western U.S. Initial results show that the average weights assigned to each model range from 3-18%. The model ensemble shows little spread during the accumulation period and increasing spread during melt at most sites, which indicates a higher likelihood of capturing events during the most variable period over the use of a single model.
H23B-0967
Assessing Forecasting Uncertainties for Improved Snow Model Predictions
Snowpack is a significant fraction of available water resources in many regions of the nation. Studies demonstrate that atmospheric warming over the past 50 years has led to a steady decline in snowpack depth as well as altered melt patterns across large regions of the western U.S. Changes in snowpack volume and spring melt timing obviously alter the distribution of water resources. The National Weather Service currently applies the SNOW17 model for operational forecasting of snow accumulation and melt in snow-dominated areas in the nation, relying on calibrated parameters and manual assimilation of observed model states including snow water equivalent (SWE) measurements. The current research aims to improve the predictability of the SNOW17 model by addressing forecasting uncertainties (parameters, forcing, SWE measurements, initial conditions, etc.) and using the established uncertainties to formulate an operational data assimilation framework. Initial work focuses on six SNOTEL sites in the Western U.S with a range of climatic and geographic conditions. A generalized sensitivity analysis (GSA) method is applied to identify sensitive parameters, identifiable parameter ranges, and correlations for parameters of the SNOW17 model. Impacts of uncertainties in precipitation forcing on parameter sensitivity are also investigated. Potential influences of climatic and geographic factors on parameter sensitivities are assessed. Based on the sensitivity and uncertainty analysis results, an ensemble Kalman filter framework is formulated and tested at several SNOTEL sites. Preliminary results indicate parameter sensitivity generally varies with climatic and geographic conditions, and that introducing uncertainties in precipitation forcing reduces overall sensitivity and identifiability of parameters. Improved snow state and snowmelt estimates are expected to improve short- and long-term streamflow predictions. The sensitivity and uncertainty analysis method as well as the proposed filter framework will assist the NWS in advancing the current forecasting system and reinforcing current operational forecasting skill.
H23B-0968
Characterization of Uncertainty in Modeling Snow Processes in High Mountains Using a Multiscale Kalman Filtering-based Data Assimilation Paradigm
Assimilation of remotely sensed retrievals into a snow model is an important method to reduce uncertainties associated with the estimates of snow states. To use data assimilation in an advantageous way it is critical that the data assimilation scheme accounts for the uncertainty structure of both the physical model and observations. To this end, we employ the Multiscale Kalman Filtering (MKF)-based data assimilation paradigm. We also examine the impacts of uncertainties associated with the meteorological forcing data on the snow water equivalent (SWE) estimates. The MKF-based data assimilation framework is fully coupled with the snow module of the Variable Infiltration Capacity (VIC) land surface model, into which the remote sensing information of the SNODAS and AMSR-E SWE is assimilated. Uncertainties related to meteorological forcing data, imperfect knowledge of snow physics, and modeling scales are analyzed and presented through an application to the Sierra Mountain region of California.
H23B-0969
Evaluation of Snow Cover Depletion to Support Snowmelt Runoff Prediction for the Cache la Poudre River, Colorado
The Cache la Poudre River in northeastern Colorado is a source of water for many agricultural, municipal,
and industrial users. Most runoff in the basin is generated from snowmelt, but snow measurements are
sparse, located only at a few high elevation SNOTEL stations and snow courses. Over much of the
watershed, no snow measurements are available to support runoff forecasts. For this study we analyzed
snow covered area (SCA) depletion characteristics to evaluate whether SCA data could improve snowmelt
runoff prediction. Moderate Resolution Imaging Spectroradiometer (MODIS) 8-day snow-cover products were
obtained for the Cache la Poudre basin from 2000 to 2006 for March through June of each year. We
analyzed snow cover depletion characteristics for spatial subsets of the basin, including sub-basins and
elevation bands. Regression analyses compare the 8-day SCA images to 8-day average stream flow at the
USGS canyon mouth gauge (the forecasting location). Results from regression analyses show a wide range
of relationships between SCA and streamflow (0.03
H23B-0970
Estimating terrain adjusted daily temperatures for snow modelling
In order to assess the impacts of climate change on those watersheds, where most of the streamflow is generated by snow melt from high elevation areas, the most important elements in simulating the hydrological behaviour is how much snow falls and what the snowmelt dynamics are. A hydrological simulation research project is under way in the upper St. Mary's River watershed, which partly includes Glacier National Park in Montana. The ACRU agro-hydrological modelling system is used to simulate all major elements of the hydrological cycle, including snow pack developments and transpiration. The approach towards spatial representation of hydrological variables is to divide the watershed into hydrological response units (HRUs), and to estimate daily minimum and maximum temperatures for each HRU using lapse rate adjusted temperatures from a base station with a complete and long record. Successful snow modelling depends on partitioning incoming precipitation into rain and snow. A traditional approach to estimate temperatures is to use lapse rates, often locally derived using climate stations at different elevations. As snow melt is dependent on ground temperatures, which differ significantly along north and south-facing slopes, further terrain-based temperature adjustments are required. To model the effect of slope and aspect on temperatures, the lapse rate derived temperatures are adjusted for incoming radiation and leaf area index. Incoming radiation is estimated in a GIS from a digital elevation model and extrapolated estimates of mean monthly transmittivity and the mean monthly diffuse proportion of incoming radiation from nearest climate station records. Results show an intuitively correct distribution of daily ground temperatures, where north facing areas have much lower temperatures than south facing slopes, with profound effects on snow melt. Currently unresolved problems include mixing of air along the ridges, inversions, and katabatic winds in mountain terrain. Results are verified with one SNOTEL site within the watershed and a series of snow surveys carried out in Glacier National Park by the USGS. As most snow models use some type of temperature based degree-day melt routine, this described concept is expected to enhance the spatial distribution of snow melt. Potential and actual evapotranspiration estimates are also expected to be more realistically distributed over the watershed.
H23B-0971
Most Critical Surface Meteorological Measurements for Modeling Distributed Snowmelt in the Sierra Nevada, California
Accurate forecasts of snowpack ablation in the Sierra Nevada are important to flood forecasters and water managers, but adequate surface meteorological observations required for snow models are often not readily available. In the Sierra Nevada, most energy available for snowpack ablation is provided by net radiation, but other energy fluxes can also contribute considerable energy to the snowpack during rain-on-snow storms, which have been associated with significant floods in the mountainous regions of the western United States. It can be difficult to obtain the necessary forcing data to run a distributed snow model with adequate temporal and spatial resolution to accurately calculate physical ablation processes on the basin scale. This is due largely to minimal observation stations, discontinuities in the data, erroneous observations, and the intrinsic heterogeneity of atmospheric processes in complex terrain. Many distributed snow models require continuous records of surface meteorological measurements, and thus there is a need for reliable methods of estimating these data when they are missing or unavailable. In this study we compare methods of estimating forcing data from other types of surface observations and contrast methods of interpolating meteorological parameters between stations for the purpose of running a distributed snow model. Specifically, methods for estimating air temperature, relative humidity, wind speed, precipitation, and solar radiation are evaluated in the well-instrumented American River and Tuolumne River Basins in the Sierra Nevada, California from Oct 2002 through May 2008. Estimated values are compared with distributed observations from both standard meteorological stations and small, self recording instruments at over 30 sites within each basin. The impacts of parameter estimations on snowmelt modeling, for both dry and rain-on-snow events, are assessed with the Utah Energy Balance Model.