Remote sensing of snow: Factors influencing seasonal snow mapping in boreal forest region

Monitoring of snow cover in northern hemisphere is highly important for climate research and for operational activities, such as those related to hydrology and weather forecasting. The appearance and melting of seasonal snow cover dominate the hydrological and climatic patterns in the boreal and arctic regions. Spatial variability (in particular during the spring and autumn transition months) and long-term trends in global snow cover distribution are strongly interconnected to changes in Earth System (ES). Satellite data based estimates on snow cover extent are utilized e.g. in near-real-time hydrological forecasting, water resource management and to construct long-term Climate Data Records (CDRs) essential for climate research. Information on the quantitative reliability of snow cover monitoring is urgently needed by these different applications as the usefulness of satellite data based results is strongly dependent on the quality of the interpretation. This doctoral dissertation investigates the factors affecting the reliability of snow cover monitoring using optical satellite data and focuses on boreal regions (zone characterized by seasonal snow cover). Based on the analysis of different factors relevant to snow mapping performance, the work introduces a methodology to assess the uncertainty of snow cover extent estimates, focusing on the retrieval of fractional snow cover (within a pixel) during the spring melt period. The results demonstrate that optical remote sensing is well suited for determining snow extent in the melting season and that the characterizing the uncertainty in snow estimates facilitates the improvement of the snow mapping algorithms. The overall message is that using a versatile accuracy analysis it is possible to develop uncertainty estimates for the optical remote sensing of snow cover, which is a considerable advance in remote sensing. The results of this work can also be utilized in the development of other interpretation algorithms. This thesis consists of five articles predominantly dealing with quantitative data analysis, while the summary chapter synthesizes the results mainly in the algorithm accuracy point of view. The first four articles determine the reflectance characteristics essential for the forward and inverse modeling of boreal landscapes (forward model describes the observations as a function of the investigated variable). The effects of snow, snow-free ground and boreal forest canopy on the observed satellite scene reflectance are specified. The effects of all the error components are clarified in the fifth article and a novel experimental method to analyze and quantify the amount of uncertainty is presented. The five articles employ different remote sensing and ground truth data sets measured and/or analyzed for this research, covering the region of Finland and also applied to boreal forest region in northern Europe