Forward-Looking Assimilation of MODIS-Derived Snow-Covered Area into a Land Surface Model

Abstract

Snow cover over land has a significant impact on the surface radiation budget, turbulent energy fluxes to the atmosphere, and local hydrological fluxes. For this reason, inaccuracies in the representation of snow covered area (SCA) within a land surface model (LSM) can lead to substantial errors in both offline and coupled simulations. Data assimilation algorithms have the potential to address this problem. However, the assimilation of SCA observations is complicated by an information deficit in the observation-SCA indicates only the presence or absence of snow, and not snow volume-and by the fact that assimilated SCA observations can introduce inconsistencies with atmospheric forcing data, leading to non-physical artifacts in the local water balance. In this paper we present a novel assimilation algorithm that introduces MODIS SCA observations to the Noah LSM in global, uncoupled simulations. The algorithm utilizes observations from up to 72 hours ahead of the model simulation in order to correct against emerging errors in the simulation of snow cover while preserving the local hydrologic balance. This is accomplished by using future snow observations to adjust air temperature and, when necessary, precipitation within the LSM. In global, offline integrations, this new assimilation algorithm provided improved simulation of SCA and snow water equivalent relative to open loop integrations and integrations that used an earlier SCA assimilation algorithm. These improvements, in turn, influenced the simulation of surface water and energy fluxes both during the snow season and, in some regions, on into the following spring.2

Why Do Global Reanalyses and Land Data Assimilation Products Underestimate Snow Water Equivalent?

“…MERRA, MERRA-Land, and ERA-I/Land do not assimilate any observational snow information (Rienecker et al 2011;Reichle et al 2011;Balsamo et al 2015). The GLDAS models only assimilate information about snow cover from the Moderate Resolution Imaging Spectroradiometer (MODIS; Rodell and Houser 2004;Zaitchik and Rodell 2009). …”

DEnKF–Variational Hybrid Snow Cover Fraction Data Assimilation for Improving Snow Simulations with the Common Land Model

“…The overestimation is caused by the large positive innovation − , which may be related to the predicted SCF observation underestimated by the simplified observation operator [20]. It is expected to alleviate the overestimation by improving the simplified observation operator in a method similar to the SCF parameterization scheme [10,61]. At the Jimunai site, the DEnVar-Beta0.5 experiment obtained the best analysis performance, but it tends to melt the snowpack too early and rapidly in Figure 3D.…”

“…Most studies have mainly used rule-based direct insertion (e.g., [10][11][12][13]), ensemble-based (e.g., ensemble Kalman filter (EnKF) [14][15][16][17], ensemble square root filter (EnSRF) [18], ensemble adjustment Kalman filter (EAKF) [19], and deterministic ensemble Kalman filter (DEnKF) [20]), or Bayesian (e.g., [21][22][23]) DA methods to assimilate SCF (e.g., [24,25]), SWE (e.g., [5,16,26]), SD [12], or passive microwave brightness temperature [27] observations into hydrological and land surface models to improve snow estimates. All existing studies show that these DA methods improve the SD and SWE estimates when available snow-related observations are assimilated into hydrological and land surface models.…”

Operational hydrological data assimilation with the recursive ensemble Kalman filter

“…The observations used to update model states can include river flows (Seo et al, 2003), soil moisture (Brocca et al, 2010;Flores et al, 2012), snow-covered area and snow water equivalent Zaitchik and Rodell, 2009;Andreadis and Lettenmaier, 2006), and satellite observations of discharge (Neal et al, 2009;Andreadis et al, 2007). Sequential use of different observation types is also possible.…”