A fractional snow cover mapping method for optical remote sensing data, applicable to continental scale

This thesis focuses on the determination of fractional snow cover (FSC) from optical data provided by satellite instruments. It describes the method development, starting from a simple regionally applicable linear interpolation method and ending at a globally applicable, semi-empirical modeling approach. The development work was motivated by the need for an easily implementable and feasible snow mapping method that could provide reliable information particularly for forested areas. The contribution of the work to the optical remote sensing of snow is mainly associated with accounting for boreal forest canopy effect to the observed reflectance, thus facilitating accurate fractional snow retrievals also for ground beneath the tree canopies. The first proposed approach was based on a linear interpolation technique, which relies on a priori known reference reflectances at a) full snow cover and b) snow-free conditions for each calculation unit-area. An important novelty in the methodology was the utilization of a forest sparseness index determined from AVHRR reflectance data acquired at full dry snow cover conditions. This index was employed to describe the similarity between different unit-areas. In practice, the index was used to determine the reference reflectances for such unit-areas for which the reflectance level could not be determined otherwise, e.g. due to frequent cloud cover. This approach was found to be feasible for Finnish drainage basins characterized by fragmented landscape with moderate canopies. Using a more physical approach instead of linear interpolation would allow the model parameterization using physical quantities (reflectances), and would therefore leave space for further model developments based on measuring and/or modeling of these quantities. The semi-empirical reflectance model-based method SCAmod originates from radiative transfer theory and describes the scene-level reflectance as a mixture of three major constituents: opaque forest canopy, snow and snow-free ground, which are interconnected through transmissivity and snow fraction. Transmissivity, in turn, can be derived from reflectance observations under conditions that highlight the presence of forest canopy, namely the presence of full snow cover on the ground. Thus, SCAmod requires a priori information on transmissivity, but given that it can be determined with the appropriate accuracy, it enables consideration of the obstructing effects of forests in fractional snow estimation. In continental-scale snow mapping, determination of the transmissivity map becomes a key issue. The preliminary demonstration of transmissivity generation using global land cover data was a part of this study. The first implementations and validations for SCAmod were presented for AVHRR data at Finnish drainage basin scale. In subsequent work, determination of the feasible reflectance constituents was addressed, followed by a sensitivity analysis targeting at selection of optimal spectral bands to be applied with SCAmod. Feasibility of the NDSI-based approach in FSC-retrievals over boreal forests is also discussed. Finally, the implementations and validations for MODIS and AATSR data are presented. The results from relative (using high-resolution Earth Observation data to represent the truth) and absolute validation (using in situ observations) indicate a good performance for both forested and non-forested regions in northern Eurasia. Accounting for the effect of forest canopy in the FSC-retrievals is the key issue in snow remote sensing over boreal regions; this study provides a new contribution to this research field and provides one solution for continental scale snow mapping.