The behaviour of mast-borne spectra in a snow-covered boreal forest

Forest canopy impairs the detection of snow underneath and optical interpretation methods tend to underestimate the snow coverage. The effect of measurement conditions on snow reflectance is considerably well understood, but the sensitivity of scene reflectance (when forest canopy present) to these properties is less examined. To investigate scene reflectance of a boreal forest, an Analytical Spectral Devices (ASD) Field Spec Pro JR was installed at the top of a 30-metre-high mast in Sodankylä, Northern Finland. Two sites – sparse forest and a forest opening – were monitored. The data consists of average reflectance spectra in the range 350–2500 nm, as well as of simultaneously acquired digital images. The Moderate Resolution Imaging Spectroradiometer (MODIS) is widely used for the monitoring of snow‐covered area (SCA). Therefore, ASD spectra were resampled to the spectral characteristics of MODIS. Time series of normalised difference snow index (NDSI), normalised difference vegetation index (NDVI) and the reflectance at selected wavelengths were investigated in order to understand their behaviour in a snow-covered forest in spring. In order to discriminate between the effects of illumination conditions and the actual snow characteristics, reflectance spectra were measured both under direct and diffuse illumination. A specific focus was on the behaviour of NDSI, which is typically used in SCA retrieval – e.g. in MODIS snow mapping algorithms by NASA's Goddard Space Flight Center (GSFC). We further modelled the scene reflectance as a function of surface reflectances considering areal proportions of Scots Pines, shadowed snow and directly illuminated snow.In the forest opening, NDSI was found to be relatively insensitive to illumination conditions, while in the forest stand it showed stronger dependence. This indicates that NDSI-based snow algorithms may provide a lower accuracy over forested areas. We demonstrated this by comparing the mast-observed indices and reflectances with the corresponding thresholds applied in MODIS snow mapping. The current thresholds led to underestimated snow cover in our test forest. It was also noticed that over the forest stand, variations in snow reflectance induced particularly great variation in the scene reflectance, even though visible snow-covered ground accounted for only half of the observed area as a result of the blocking effect of the trees. Comparison between the mast-borne and the modelled scene reflectance suggest that the surface reflectance observations could offer a feasible means to predict scene reflectance characteristics of snow-covered boreal forests observed by space-borne instruments.

► Snow physical properties caused clear variation in the spectra from boreal forest.
► Forest canopy dominates the signal from boreal forest even with full snow coverage.
► NDSI and NDVI from snow are rather insensitive to illumination conditions.
► Surface reflectances are feasible to describe scene reflectance from boreal forest.