Dry snow extent monitoring in strong topography conditions

A new method to discriminate dry snow in alpine regions is presented. Due to the wide variety of Alpine environments, the scene under study is first segmented into surface and forested area classes from L and C-band summer polarimetric SAR data. Dry snow is then discriminated over each class using adapted methods. A new polarimetric multi- temporal optimization procedure, named PCVE, is proposed to increase the slight polarimetric contrast due to the presence of snow over surfaces. Snow covered forests are discriminated using performing polarimetric indicators resulting from a decomposition of incoherent matrix representations. The effectiveness of this method is demonstrated over a French alpine test site using SIR-C L and C-band polarimetric SAR data. I. INTRODUCTION The localization of dry snow in alpine environments using intermediate frequency SAR data (L and C-bands) still remains a problematic application (1). Indeed, at such frequencies, dry snow is a low attenuation medium and only slightly affects the backscattered signal amplitude. Moreover alpine areas are characterized by a wide variety of underlying media with changing characteristics and important topography that may strongly affect a scene response. Dry snow mapping is an important product for global snow monitoring, widely used in the frame of hydrological applications, like Snow Water Equivalent determination. This paper presents a polarimetric method to map dry snow extent in alpine areas using multi-frequency and multi- temporal polarimetric SAR data. Due to the variability of alpine environments the method is decomposed into three steps. A first part is dedicated to the classification of the scene into surface and forest types from summer data sets. The classification may be applied over both L and C-band data sets. The main advantage of the C-band summer classification is the possibility to lead the dry snow discrimination analysis at a single frequency band. Each media is then processed separately. The presence of snow is then detected over surfaces by means of a new optimization method, based on a Polarimetric Contrast Variation Enhancement (PCVE) (2). Snow covered forests are analyzed from summer to winter variations of polarimetric decomposition results at C-band. Merged discrimination results are finally analyzed through a quantitative estimation of the detection performance. II. TEST SITE AND SAR DATA The test region is located in southern French alps (N 44°.15' / E 7°.15') and was measured during the SIR-C campaign in April and October 1994. Multi-temporal fully polarimetric SAR data sets were acquired at L and C-bands in both snow free (October) and snow (April) conditions. The test sites, Risoul (300 km²) and Izoard (800 km²), are composed of three main alpine environments: high altitude unvegetated surfaces, medium to high altitude forested zones and low altitude valleys (Fig. 1). Simultaneously to radar acquisitions, ground truth measurements were carried out over the test sites (automatic and manual snow sample network). They are summarized in Table I. Various types of underlying media may be encountered over the considered sites: rocks, bare soils, forests and pastures. Both test sites are partly covered by frozen spring snow, due to the early morning acquisition time. The snow cover altitude ranges from 1200 m up to 3000 m where its depth reaches 2 m. In this paper, only the Risoul site results are presented (Fig. 1).