Reflectance Modeling of Snow-Covered Forests in Hilly Terrain

Seasonal snow covers large land areas of the Earth. Information about the snow extent in these regions is important for climate studies and water resource management. A linear spectral mixture model for snow-covered forests (the SnowFor model) has previously been developed for flat terrain. The SnowFor model includes reflectance components for snow, trees and snow-free ground. In this paper, the model is extended to handle radiometric effects caused by topography on mixed pixels of snow and trees through subpixel topographic reflectance modeling. Empirical reflectance models for snow and trees, based on the local solar incidence angle, are proposed (TopoSnow and TopoTree models), and integrated into the SnowFor model. Experiments with two Landsat Thematic Mapper (TM) images are carried out in hilly, forested terrain in Alptal, Switzerland with full snow cover. Results show that the calibrated TopoSnow and TopoTree models enhance the modeling of reflectance variability from snow-covered forests for visible and near-infrared wavelengths. The performance of four other topographic correction methods is evaluated for snow-covered forests. Introduction Seasonal snow covers large areas of the Earth’s land surface. Information about snow is important for climate studies (Cess, et al., 1991) and hydrological applications. Snowmelt may contribute significantly to the runoff from drainage basins, and runoff forecasts serve flood warning and hydropower production (Winther and Hall, 1999). Therefore, improving the techniques for monitoring the snow cover is of growing interest. Remote sensing techniques have far reaching monitoring potential. Methods for snow-cover mapping have been developed for optical sensors (e.g., Andersen, 1982; Hall, et al., 1995; Rosenthal and Dozier, 1996) for active microwave sensors (e.g., Koskinen, et al., 1997; Nagler and Rott, 2000) and for passive microwave sensors (e.g., Hallikainen, 1989; Foster, et al., 1997). Presently, the reflected signals from snowcovered areas recorded by optical sensors are better understood than those acquired by microwave sensors (Solberg, et al., 1997). Hence, optical images are most frequently used in operational snow-cover mapping, although clouds may obscure the mapping (Solberg and Andersen, 1994; Hall, Reflectance Modeling of Snow-Covered Forests in Hilly Terrain Dagrun Vikhamar, Rune Solberg, and Klaus Seidel et al., 2002). Snow in forests is challenging to map with optical remote sensing techniques because some of the snow is masked by the tree canopy and thereby mostly occluded from the sensor. Additionally, the trees contribute spectrally to the satellite-measured radiance. A few studies have particularly focused on handling the forest problem (Klein, et al., 1998; Metsämäki, et al., 2002), and it has been demonstrated that snow in forests is mapped with lower accuracy than in unforested regions (Hall, et al., 2001). To investigate the problems of snow-cover mapping in forests, a linear spectral mixture model (SnowFor) has previously been developed for different forest types with snowcovered ground and flat terrain (Vikhamar and Solberg, 2002 and 2003b). It is known that topography significantly influences the radiometry of the acquired satellite image (Proy, et al., 1989). There are two objectives for the work presented in this paper: 1) To study radiometric effects caused by the topography in snow-covered forested areas; and 2) To model these topographic effects in SnowFor. The SnowFor model is the main component of a snow-cover mapping method for forests currently under development. Reflectance modeling is a step on the way to understand processes, and thereby being able to improve subpixel snow cover mapping. For this study, two Landsat TM images of hilly, forested terrain in Switzerland are selected for experiments. The 30 m spatial resolution makes it possible to identify and study pure snow and forest pixels. Empirical reflectance models for snow and trees, based on the local solar incidence angle, are proposed (TopoSnow and TopoTree), and integrated into the SnowFor model. This approach, referred to as subpixel topographic reflectance modeling, was motivated by an experiment presented here which investigates the performance of four other topographic correction methods for snow-covered forests. Modeled and observed Landsat TM reflectances are compared for visible and near-infrared wavelengths. The paper first provides a short summary of how trees and topography affect the image radiometry, and how terrain effects are handled by four topographic correction methods. Then, the approach for integrating radiometric terrain effects into the SnowFor model is presented (Figure 1). Moreover, the study area, data sets, and four experiments are successively described. Finally, the result of the reflectance modeling is assessed. Background Effects influencing the measured radiance from snow-covered forests, and how terrain effects are handled by four topographic correction methods, are briefly described in the following. P H OTO G R A M M E T R I C E N G I N E E R I N G & R E M OT E S E N S I N G September 2004 1 0 6 9 D. Vikhamar is with the Department of Geosciences, University of Oslo, P.O. Box 1047 Blindern, N-0316 Oslo, Norway ([email protected]). R. Solberg is with the Norwegian Computing Center, P.O. Box 114 Blindern, N-0314 Oslo, Norway ([email protected]). Klaus Seidel is with the Remote Sensing Group, Computer Vision Lab (ETHZ), CH 8092 Zürich, Switzerland ([email protected]). Photogrammetric Engineering & Remote Sensing Vol. 70, No. 9, September 2004, pp. 1069–1079. 0099-1112/04/7009–1069/$3.00/0 © 2004 American Society for Photogrammetry and Remote Sensing 02-072.qxd 8/5/04 18:45 Page 1069