Methodische Probleme bei der Erstellung von Unkrautverteilungskarten mit Geoinformationssystemen (GIS)

This study discusses the methodical problems which occur during the process of generating weed distribution maps in geoinformation systems (GIS). Particular attention will be paid to the question of whether or not the weed sampling data and the weed distribution maps for site-specific weed control are sufficiently representative. Manual or image-processing sampling procedures are applied in order to generate weed distribution maps. In these procedures weeds are counted in counting areas of a defined size distinguishing between different weed species. The samples gathered in this way are treated as being sufficiently representative for the specific site in question. This does not take into account the fact that the sampling procedure with a counting area or quadrate is transferred to site-specific weed control from plant sociology and integrated pest management which use this strategy for 100 years in order to estimate the mean frequencies of a species in a certain area. The number of samples in these particular strategies is chosen high enough to estimate the mean frequencies adequately. In site-specific weed control there is only one single sample taken for a reference area. This procedure can lead to non-representative, ambiguous weed distribution maps, which will be demonstrated by applying a GIS-based analysis. Spatial interpolation processes in a GIS are applied in order to be able to produce weed distribution maps from the sampling data. Whilst this transformation from discrete sampling data to a continuous surface takes place, several problems occur during the choice of an interpolation method depending on its requirements towards the sampling data. In this study the interpolation methods recently used in weed research will be analysed and described in the context of site-specific weed control. Recent methods of data acquisition and weed mapping so far can only be assessed by uncertain knowledge applying sampling data. Hence, in this study a new method was implemented which offers a detailed picture of the precision of recently applied sampling procedures for site-specific weed control. A prototype was constructed, (WeedScanner) in order to detect weeds at a high resolution for a detailed analysis of some arable fields applying digital camcorders. Based on the resulting high resolution weed distribution maps several detailed studies have been conducted in this work. A custom GIS-software (Sampling Simulator) was developed in this study in order to be capable of calculating the errors of a sampling procedure based on the above described high resolution datasets or simulated populations. It will be demonstrated that manual weed sampling methods cannot lead to weed distribution maps which adequately represent the true weed distribution in the arable field. Nevertheless, it will also be shown that image-processing methods of sampling prove reliable results. In order to minimize methodical problems during the working process with weed survey datasets a metadata profile conforming to ISO 19115 metadata standard for geospatial data was drafted. This profile will improve the sustainability of data and their administration. With regard to the outsourcing of production processes to agricultural service providers and the stipulated documentation of the production process in agriculture in the next year it will be vital to implement metadata management. Objective evidence for the insufficiency of recently applied offline-methods of weed sampling for site-specific weed control is provided in this study. Besides, with the WeedScanner a tool for the semi-automatic precise detection of the basic population of weeds in an arable field is introduced.