Bernert, Hendrik (2015)

Evaluation of MODIS-derived phenological parameters using ground based observations

In this thesis, phenological descriptors for Germany have been extracted from a remotely sensed vegetation index time series using a spline-modelling approach. The results have been compared to ground based observations from the DWD, revealing a strong dependency not only on the land cover, but also the greenness parameter g, which is necessary to determine the onset of greening during the growing season. It has become clear, that this value is species-dependent and remains crucial for the estimation of suitable phenological parameters. For forest stands, especially for deciduous trees, the results for the start and end of the growing season are in good agreement with the reference data, with differences of 5 to 10 days for the former. Winter crops cause problems in deriving the appropriate growing season, because they are often associated with intermediate crops or fallow, which results in distorted vegetation index signals for that respective growing cycle.

The spatial distribution of the phenological parameters can represent land use classes, which may be used in future classification frameworks to distinguish not only certain land use, but also land cover types (meaning dominant plant species). Variations of the descriptors within these classes are mostly very high, probably being the result of the very heterogeneous study area and the estimation of the actual plant species at the reference stations. Likewise, the latter has shown to be the biggest issue of this study, which leaves the most room for improvements. The data from the pan-European LUCAS surveys, which map the land cover of Europe in a 2x2 km point-grid, can be used for single years, when appropriately combined with the DWD data.

The splits framework, which was used for the spline-modelling approach, has proven to be a very handy tool to set up an appropriate spline model for a vegetation index time series as well as to extract phenological parameters from it. Though splines with different properties have to be evaluated visually concerning their appropriateness or representative quality for the raw data, the tool has proven to be able to achieve overall reasonable results. Compared to other studies, which mostly dealt with small, homogeneous regions and in-situ reference measurements, the outcomes of this very general approach for a large-scale, heterogeneous study area are very promising and might become even better with more detailed reference data.