Data assimilation is an ideal framework for merging multi-frequency remote sensing observations and a snow physics model, because it provides a means of weighing the tremendous uncertainty of meteorological data (such as precipitation measurements from snow gages) and remote sensing observations. The objective of this study is to assess the feasibility of using the Ensemble Kalman Filter (EnKF) to characterize snowpack properties. The EnKF follows a Monte Carlo simulation approach in which all parameters, forcing variables, and initial conditions are treated as random variables. By simulating the evolution of an ensemble of state variables, an estimate of the uncertainty of the state variables is obtained. The relative uncertainty of the state variables is weighed against the uncertainty in the remote sensing observations through the EnKF update equation. The advantage of the EnKF data assimilation framework is that it easily allows for the incorporation of observations from several different portions of electromagnetic spectrum; each observation may contain different information about the snowpack, is available at a different spatial and temporal resolution, and has a different level of uncertainty.
In this study, a season-long, one-dimensional experiment is performed in which synthetic observations from the microwave, visible/near-infrared, and thermal infrared spectra are used to update snowpack states in a land surface model. Specifically, synthetic microwave observations from AMSR-E and SSM/I, as well as synthetic broadband albedo observations, and surface temperature observations from MODIS are assimilated. The land surface model has been modified to compute estimates of the average grain diameter and the snowpack albedo. Various parameters which affect the evolution of the snowpack were treated as random variables in order to make the LSM compatible with ensemble forecasting. Results from the assimilation are compared to those from a pure modeling approach and from a remote sensing inversion approach. Tradeoffs between ensemble size, estimation error, and computational expense are explored.