Seasonal Variability in Spectral Reflectance for Discriminating Grasslands along a Dry-mesic Gradient in Switzerland

Dry grasslands in Switzerland are species-rich habitats resulting from a traditional agricultural land use. Almost 40% of plant and in some cases over 50% of the animal species present on dry grasslands are included in the red lists, and are classified as endangered or threatened. Furthermore, it is estimated that about 90% of dry grassland in Switzerland have been transformed to other land cover types over the past 60 years. Existing grasslands are managed differently depending on the region and the lower-altitude communities range from very dry and nutrient-poor to mesic and nutrient-rich conditions. There is a need to better understand the seasonal variation of the reflectance properties of these grassland ecosystems in order to develop efficient and reliable tools for mapping, evaluating and monitoring within the framework of a national inventory. In this study, we examined the potential use of remote sensing for monitoring the development of these grasslands during their growing season. In addition, we investigated the optimal points in time during the growing season for discriminating the different grassland types spectrally. For this purpose a field spectrometer, the Analytical Spectral Devices (ASD) FieldSpec Pro FR, was used to collect reflectance data from 12 sample fields (4 grassland types) in 12 time steps at the Cantons of Aargau and Chur. The measurements examined were from the beginning of March until the beginning of October 2004. The 4 grassland types cover the wetness / nutrient gradient. The revisiting period of the sample areas was approximately 10-14 days depending on the weather conditions. Analysis for statistically significant differences in reflectance was performed between the vegetation types during the growing season. Continuum removal analysis was used as a spectral transformation method in addition to the original reflectance spectra. After the statistical significant bands between the grassland types were found, Classification and Regression Trees (CART) were used to select the bands that could optimally be used to discriminate the different types. Finally, using the bands selected from the CART analysis, the separability of the grassland types during the season was estimated using the Jeffries-Matusita (JM) distance. Our results demonstrate that there is seasonal variation in the spectral reflectance of the grasslands. Furthermore, the potential of using spectral information for discriminating different grassland types changes during the growing period.