| Article ID | Journal | Published Year | Pages | File Type |
|---|---|---|---|---|
| 4375006 | Ecological Informatics | 2013 | 8 Pages |
The loss of species is an ongoing process threatening the services provided by ecosystems for humanity. In agricultural areas, which globally occupy the largest areas, species diversity is strongly dependent on the farmers' management. If management allows for a landscape structure with many remnants of natural or semi-natural vegetation, the diversity and amount of flora and fauna is higher compared to intensive agricultural landscapes with large, continuous field parcels and a low crop variety. Knowledge of agricultural management is crucial for the development and monitoring of political strategies that aim to enhance or conserve species in agricultural areas. Remote sensing is a cost effective way to acquire this knowledge for large spatial areas with high temporal resolution. Since 2008, modern satellite-based radar sensors deliver images of unprecedented high quality. Since the acquisition of radar images is not restricted by atmospheric conditions, it is very capable of multitemporal classifications. In the presented study, possibilities for supervised multitemporal classification of non crop areas are investigated based on TerraSAR-X images and two different classifiers (Maximum Likelihood and Random Forest). The best results were achieved for the classification of woody structures where producer's accuracies are above 80%. Despite lower values for the other classes (flower strips 75%, grasslands 75.8% and herbaceous 57%), these classes are easily recognized. This is illustrated by different map examples. The presented results can contribute essentially to the monitoring, investigation and increasing of habitat structures in agricultural areas.
► Semi-natural areas and grasslands were classified using images from satellite radar. ► The Random Forest Classifier performs better than the Maximum Likelihood Classifier. ► The results are valuable for the conservation of species in agricultural areas. ► The results are an important contribution to the mapping of ecosystem services.
