Comparing different techniques and algorithms for the retrieval of SWE and snow depth from space-borne microwave radiometric data over forested areas
M. Tedesco¹, J. Pulliainen², M. Takala², M. Hallikainen² and P. Pampaloni³

1) GEST – NASA Goddard Space Flight Center – Greenbelt – MD – USA
2) Helsinki University of Technology – Espoo – Finland
3) IFAC – CNR – Firenze – Italy

mtedesco@neptune.gsfc.nasa.gov

Snow represents an important component of the Global Water Energy Cycle, covering up to 53% of the northern hemisphere and up to 44% of the world land mass. The estimation of snow parameters such as Snow Water Equivalent (SWE) and snow depth at large scales represent a useful support for developing and testing hydrological models, forecasting of water storage and for climatological applications. Space-borne microwave instruments can be used for this purpouse. In particular, the SSM/I radiometer, flying on the DMSP series satellites, provides data with daily coverage of more than 80% of the Earth's surface. Data over areas of interest are available almost twice per day, with the asceding and descending orbits. The temporal continuity of data is garuanteed because at the frequencies under consideration (19 and 37 GHz) clouds weakly affect recorded brightness temperatures and solar illumination is not required. However, the pixel size of SSM/I (25x25 Km2) has the consequence that several types of surfaces (i.e. forest, water ) can influence the single-pixel value of brightness temperature. With the launch of AMSR and AMSR-E radiometers the spatial resolution has been improved, but the heterogeneity of the observed scene must still be considered.

Several numerical techniques have been developed for the extraction of SWE and snow depth from space-borne radiometric data. In this study, we apply different technoques and algorithms over forested areas in order to compare their performances by means of the Root Mean Square Error (RMSE), R2 and regression coefficients. Algorithms developed by Al Chang and its modification for forested areas are considered, as well as the Helsinki University of Technology (HUT) iterative algorithm, the Spectral Polarization Difference (SPD) algorithm and the inversion of theoretical equations of the Dense Media Theory through Artifical Neural Netwroks. A total number of twelve (12) test sites containing densely forested areas over Finland is used for the study. Ground snow data set based on observations of the Finnish Environment Institute (SYKE) and the Finnish MEteorological Institute (FMI) are used to compare retrieved parameters with ground based observations. The study is performed over data collected between the beginning of 1996 and the end of 1999.

Obtained results show that both SPD and the iterative inversion of HUT snow emission model provides good results when SWE values were lower than 150-170 mm or springtime measurements are disregarded. The formula proposed by Chang for the retrieval of snow depth gives the best results by taking into account only measurements performed until mid February. Artificial neural networks trained either with the HUT model or experimental data showed good accuracy for both parameters. Finally, the results obtained with Chang’s algorithm modified for forested areas show that improvements are obtained with respect to the algorithm without forest correction for values of forest cover percentage up to a threshold value, over which the correction term due to the forest leads to high errors.

[back]