1. Development and Evaluation of a Forward Snow Microwave Emission Model
Kostas Andreadis1, Dennis Lettenmaier1 and Eric Wood2
1 Civil and Environmental Engineering, University of Washington
2 Civil and Environmental Engineering, Princeton University
EGU General Assembly
Vienna, 3 April 2006

2. Motivation
Passive microwave remote sensing offers opportunity for observing snow properties over large areas
Has many limitations (e.g. wet snow, saturation effects)
Data assimilation provides framework for merging model and satellite-based information
Direct assimilation of satellite SWE is problematic

3. Motivation (cont'd)
Need for an accurate and robust snow microwave emission model
Development of Land Surface Microwave Emission Model (LSMEM)
Initial validation and sensitivity over data intensive sites (CLPX)
Small-scale data assimilation experiment with a coupled hydrology-radiative transfer model

4. LSMEM Description
Extension of warm season LSMEM (Gao et al. 2003) to include a snow microwave emission component
Snow module based primarily on semi-empirical HUT model (Pulliainen et al. 1999)
Assumes that scattering is mostly concentrated in the forward direction
Extinction coefficient computed empirically
Dielectric constants estimated from strong fluctuation theory

5. Dataset Description
Cold Land Processes Experiment (CLPX) during winters 2002 & 2003 at several sites in Colorado, USA
Multi-sensor, multi-scale measurements
Ground-based radiometer (GBMR) and snowpits (Local Scale Observation Site) located within the 1x1 km Fraser Intensive Study Area
GBMR footprint free of vegetation
Satellite data (AMSR-E, SSM/I, MODIS etc) and aircraft data (PSR etc)

6. SNTHERM LSOS Validation
Evaluate ability of SNTHERM to reproduce snow conditions over site of interest
Validation data from snowpits over the LSOS (3 Feb-29 Mar 2003)
Calibrate SNTHERM over entire period and use as benchmark simulation

7. Comparison with Ground-based Tb

8. Sensitivity and Error Estimation
Sensitivity of Tb model prediction to various snow parameters
x-axis values refer to difference from the nominal value
y-axis values show difference between predicted and observed Tb
Linear dependence might be an artifact of the model

9. Data Assimilation Experimental Design
Benchmark simulation: calibrated SNTHERM over LSOS, with baseline forcing data (same with one used for GBMR validation)
Prior simulation: uncalibrated SNTHERM without assimilation, with “wrong” forcings and initial state
Filter simulation: uncalibrated SNTHERM with AMSRE assimilation from LSMEM, with “wrong” forcings and initial state
“Wrong” forcings created from perturbing Precip, Tair, SW Rad, LW Rad and RH
Ensemble Kalman Filter used as assimilation technique

10. Comparison with AMSR-E Data
Scale discrepancy between model and observations
Compare variability of percentiles
19 GHz (H)
37 GHz (H)

11. Assimilation Results State variables: snow depth, SWE and grain size
EnKF allows for the representation of uncertainty in important parameters, i.e. grain size
19 GHz (V)

12. Assimilation Results
Evaluate effects of assimilating different sets of microwave channels
Very similar performance (RMSE: and respectively, versus when using all channels)
Data assimilation offers framework for evaluating and optimizing potential observation missions

13. Conclusions
Data assimilation of microwave Tb shows potential for estimation of snow properties
Limitations in snow microwave emission model
Develop a more physically-based model, combining approaches from LSMEM, DMRT and MEMLS
Develop multi-layer snow model component for the macroscale Variable Infiltration Capacity model
Conduct further validation of LSMEM in different snow conditions as part of ongoing model inter-comparison

14. Questions?