Assimilation of point SWE data into a distributed snow cover model comparing two contrasting methods

Abstract

In alpine and high-latitude regions, water resource decision making often requires large-scale estimates of snow amounts and melt rates. Such estimates are available through distributed snow models which in some situations can be improved by assimilation of remote sensing observations. However, in regions with frequent cloud cover, complex topography, or large snow amounts satellite observations may feature information of limited quality. In this study, we examine whether assimilation of snow water equivalent (SWE) data from ground observations can improve model simulations in a region largely lacking reliable remote sensing observations. We combine the model output with the point data using three-dimensional sequential data assimilation methods, the ensemble Kalman filter, and statistical interpolation. The filter performance was assessed by comparing the simulation results against observed SWE and snow-covered fraction. We find that a method which assimilates fluxes (snowfall and melt rates computed from SWE) showed higher model performance than a control simulation not utilizing the filter algorithms. However, an alternative approach for updating the model results using the SWE data directly did not show a significantly higher performance than the control simulation. The results show that three-dimensional data assimilation methods can be useful for transferring information from point snow observations to the distributed snow model.

Assessing the benefit of snow data assimilation for runoff modeling in Alpine catchments

“…These values were replaced using a stochastic gap filling model that accounts for data from the same station before and after the gap, as well as for data from neighboring stations at similar elevations. Temperature data were obtained from 220 stations and interpolated using an inverse distance weighting approach as described in Magnusson et al (2014), which considers both horizontal and vertical distances between measurement stations and interpolated grid cells. A variable weighting factor was used to determine the influence of horizontally near but vertically distant stations.…”

“…While some details on the external snow model framework are given below, a full description of model and data assimilation methods is available in Magnusson et al (2014). We applied three versions of this model, denoted M1 to M3.…”

“…However, the models allow for mixed precipitation in a range close to the threshold temperature (see Eq. (10) and the corresponding description in Magnusson et al, 2014). Second, fractional snow-covered area (SCF) is parameterized using modeled snow depth and terrain parameters that were derived from a 25 m digital elevation model according to Helbig et al (2015).…”

“…Recently, various methods to assimilate observational snow data for snow cover models have been developed. At the point scale, model improvements due to assimilation of snow water equivalent data from observations were already shown (Magnusson et al, 2014). At the catchment scale and for operational purposes, several studies evaluated the effect of additional information from snow observations with different approaches.…”

A Particle Filter scheme for multivariate data assimilation into a point-scale snowpack model in Alpine environment

“…Unlike the KF and EKF schemes, this method does not require a model linearization since the error estimates are evaluated from an ensemble of possible model realizations using the Monte Carlo approach (Evensen, 2003). In the recent past, an increasing number of 15 studies on snow hydrology have contributed to confirm the EnKF as a well-performing technique enabling to enhance the accuracy of hydrological simulations by consistently updating model predictions through the assimilation of snow-related observations (Andreadis and Lettenmaier, 2005;Durand and Margulis, 2006;Clark et al, 2006;Slater and Clark, 2006;Su et al, 2008;Durand and Margulis, 2008;Su et al, 2010;De Lannoy et al, 2012;Magnusson et al, 2014;Griessinger et al, handling systems nonlinearities, PF schemes are currently garnering a growing attention for snow modelling applications. Leisenring and Moradkhani (2011) compared the performances of common sequential EnKF-based DA methods and PF variants at assimilating synthetic SWE measurements to improve its seasonal predictions and to estimate some sensitive parameters in a small-scale snowpack model.…”