Hendrik Laue*, Ralf Schmidt
Funding: Senator für Bildung und Wissenschaft; Programme "Arbeit und Umwelt" 

Objective of the project is the extraction of information from satellite images of high spatial resolution (i.e. better than 30m) for non-scientific users. The physical analysis of satellite images is strongly influenced by the needs and demands of the users which are predomninately interested in usefull information and not in the data base itself or the methods how the information is extraced from these data. In order to obtain realistic requirements for the data analysis and method development two pilot users participate in the project, which covers two areas of activity. The first  is the "Prielkantendetektion" (i.e. detection of waterlines in wadden flats) and the second is the  "Obsterntevorhersage" (i.e. prediction of the harvest yield for orchards).

The first area of activity is concerned with the detection of the waterlines (i.e. the borderline between water covered areas and dry wadden flats) of the the drainage channels (german "Priele") in the wadden sea. These data are needed by the Department of Planning of the City of Cuxhaven which is responsible for the area survey and mapping of Cuxhavens wadden sea. The mapping is produced in order to ensure the safety of visitors and tourists in this very complex and rapidly changing area.  Up to now these maps have been created from digitalized photographs taken by airborne conventional cameras. Each pixel of a waterline is marked by an operator by visual inspection with the mouse on the computer screen, which is a labourious and time consuming work. Thus the use of satellite images and the development of an almost automatic edge detection algorithm for waterlines would be a cost effective alternative. In this project edge detection algorithms for waterlines based on different image processing techniques have been developed for different types of satellite images, especially for panchromatic and synthetic aperture radar (SAR) images. An example for the results obtained by using an algorithm for panchromatic images is given in Fig. 1, which shows a drainage channel in the wadden sea of Cuxhaven. The automaticly detected edges are overlaid in yellow. Despite the fact that some gaps in the waterline remain and several "false" egdes on the wadden flats occur, about 80% of the waterline are detected.  These data can help to reduce the manuall portion of the work, which has still to be done, considerably (at least down to 20% of the original effort). 

The second area of activity is concerned with the extraction of information from multispectral satellite images which can be usefull for the support of  harvest yield prediction for orchards. In a region near the City of Hamburg/Germany, called "Altes Land", the farming is dominated by orchards which consist in more than 80% of the area of apple trees. A reliable prediction of the harvest yields for these fruits is very important for the trading organizations and for the farmers in order to optimize their stock capacities and get an estimation of the expected price developement. But all existing methods give predictions which can differ up to 100% from the finally obtained harvest yield.  Any information that could help to improve these predictions would be very valuable. Because of the extension of the area only a few spot-test are used which may not be representative for the whole region the uncertainty of the prediction is further increased. Thus the use of remote sensing data, especially satellite images can remedy the problem of the generalization of point-like information for a large area. In a first step in the project of the parameters which are usefull and extractable from remote sensing data are identified. The problem is that most of the commercially available multispectral satellites sensors povide only a few very broadbanded spectral channels in the visible (VIS) and near infrared (NIR). In order to achieve a better insight which parameters related to the apple trees can be derived from these data and which useful information can be extracted (e.g. strength of the blooming, health status of the trees, stress factors like drought or diseases, frost damages, etc.) ground based measurement with a hyperspectral scanner have been carried out. The results of these spectral and spatial very high resolved data will help to develop algorithms for the analsis of the satellite images. 
 
 

Fig. 1:  Subimage of an IRS-1D PAN-scene taken at approximately 1 hour bevor low tide. The automatically detected  edges are marked in yellow. (For abetter view please open the jpeg-image in a  separate window.) 
 
 

Fig. 2: Spectrum of sunlight reflected by an apple blossom. The bars illustrate the corresponding LANDSAT channels and their interpolated intensity.

Fig. 3: Example of data taken by an hyperspectral scanner. Clearly visible are the apple trees with white blossoms.

Teamwork

• Stadtplanungsamt Cuxhaven 
• Erzeugergemeinschaft Elbe-Obst r.V. 
• Obstbauversuchsring Jork e.V.
• airobotics GmbH 

• R. Schmidt and, Laue, H., NutzerORentierte SATellitenbildanalyse - NORSAT, Institute of Environmental Physics, University of Bremen, Final Report, Bremen, 2001 

(abstract of the report in german )