The interdisciplinary research project CRC TR32: ?Patterns in Soil-Vegetation-Atmosphere-Systems: Monitoring, Modelling and Data Assimilation? works on matter fluxes between soil, vegetation and atmosphere (SVA) on a regional scale. In this regard, high quality land use data is a strongly demanded spatial information. The study area of the CRC TR32, the catchment of the river Rur, is situated in western Germany, parts of the Netherlands and Belgium. Due to the huge amount of annually changing cultivated land in the area of investigation, detailed and up-to-date vegetation information is essential to accurately assess the impact of change in vegetation cover on SVA interactions. Moreover, for applications like modelling of (agro-)ecosystems the spatial information on agricultural crop types and crop rotations for every land use parcel need to be known. However, in case of the study area, the land use datasets, which are available from official sources, only contain rather generalised information concerning agricultural areas.
Thus, in this contribution, we present our work on the generation of multi annual disaggregated land use data for the years 2008-2010. Additionally, a crop rotation map of the study area is derived from the annual datasets. For this purpose the Multi-Data Approach (MDA) is applied, which integrates supervised classification results of multiple remote sensing scenes and additional spatial information by using GIS methods and expert knowledge based production rules.
For the annual land use analysis, multiple observation dates (about 3-5) during the growing season are usually necessary to capture the major crop types, due to the different crop phenology depending on the region and the crop type. For example, in the study area, crop information regarding winter wheat and winter barley can be ideally obtained in early June, while sugar beet and maize are differentiated well in observations of late July. To achieve sufficient spatial and temporal coverage, multispectral images from various sensors including ASTER, Landsat (TM, ETM+), IRS-P6 (LISS III) and RapidEye (level 3A) were purchased for the remote sensing analyses. Therefore, beside data specific pre-processing tasks (e.g. mosaicking, sensor calibration), several steps were necessary prior to the classifications. Due to differences in spatial resolution (ranging from 30 m to 5 m), all datasets were resampled to 15 m. This resolution was chosen as a trade-off between sensor capabilities and the amount of data to be processed. Subsequently, all datasets were carefully registered to a topographic base layer with high spatial accuracy, which is fundamental for multi data analysis. The highly accurate road network of the ?Authorative Topographic-Cartographic Information System? (ATKIS) was incorporated for this purpose. The ATKIS is the official Digital Landscape Model (DLM) for Germany, which is provided (as vector data) by the governmental survey and mapping agency. Subsequently, the land use for each time step was retrieved by using both Support Vector Machines (SVM) and the Maximum Likelihood Classifier (MLC). Our results indicate that both methods produce similar results, with SVM being more precise at class boundaries whereas MLC needing significantly less processing time.
In the GIS environment, the best classification results of each observation date are then stacked to a multi layer file for each year. The final annual land use classes were then determined with knowledge-based production rules, which extract the valuable information of each layer. Subsequently, the crop rotation map for 2008-2010 was obtained by combing the annual land use maps to a new dataset.
For the comprehensive mapping of crop types and crop rotations, the incorporation of multi sensor data is normally inevitable. Our results show, that the MDA is capable of integrating multiple data sources to derive enhanced land use information.

Keywords: remote sensing, GIS, land cover, land use, multitemporal, multisensor, classification, crop,