TimeSpec — A software tool for analyzing time-series of spectral data

With the introduction of LANDSAT 8 and the future Sentinel missions as well as several other national satellite programs, the analysis of multi-temporal remote sensing data will be of main importance for detecting and analyzing different landcover classes. The utilization of MODIS/MERIS data within the last decade led to the development of a variety of tools for time-series analysis of Earth Observation data. The potentially higher spatial resolution of the new generation of multi-temporal sensors provides the feasibility to use these tools to classify finer hierarchies of land-cover classes. This accounts especially for vegetation types or habitats with a high inter-annual and intra-annual variation. Based on these assumptions, the tool TimeSpec is introduced. TimeSpec calculates out of repeated measurements of spectroradiometers or satellite sensors a spectral-temporal library (STL) and extracts training values of any given date and class for a further classification processes. First tests of the utilization of TimeSpec for a grassland area in north-east Germany show that STL-generated training samples can provide sufficient classification results. However, an application of on-site training data is still more accurate. First test on the sensitivity of training data out of a STL show that there is a saturation in accuracy with approximately 20 training samples per class. Influences of changing variance or outlier settings for the training sample generation were limited.