Environmental Monitoring of Agricultural Activities Using ASAR ENVISAT Data (2010)

Environmental Monitoring of Agricultural Activities Using Satellite SAR Data

Increasing demands for sustainable and environmentally conscious use of natural resources, fertilizers and pesticides, require the application of new technologies in agriculture. In this research and development project, the use of multi-temporal ENVISAT dual-polarization SAR data is investigated for monitoring agricultural land use and its change. The research is carried out within a water protection area, which supplies about 90% of the drinking water for the region of <st1:place>Hannover</st1:place> (“Fuhrberger Feld”). The satellite data are classified using different techniques and the results are compiled as thematic maps within a GIS. In-situ ground truth for analysis result evaluation is acquired through field inspections parallel to data takes of the satellite. Time series are collected since 2003.

The overall aim of the study is to maximize the classification accuracy. For this purpose the influence of various parameters and methods is investigated systematically. For example, the choice of a discriminative and as small as possible set of acquisition dates of the SAR imagery is important, which can be supported by exploiting context knowledge, such as crop calendars. Furthermore, the impact of various image pre-processing techniques, such as speckle-filtering, is analysed. In addition, the benefit of modern classification methods, such as Support Vector Machine, compared to standard techniques is evaluated. The results could be used by the farmers to proof good technical practise which is legally required for the protection of the environment.

Multitemporal image (July/June/April)	Land use classifikation

Recent works, make use of the new high-resolution TerraSAR-X images. A total of 10 SAR data-takes (HH and polarisation VV) forming a time series have been used in a pixel based Maximum-Likelihood classification including information of a regional crop calendar. It could be shown that by the combined use of these informations a classification accuracy of more than 90% becomes possible. Further the multi-temporal data was evaluated by a factor-analysis (variance / Covariance-Analysis) using SPSS statistic package. Besides an overall correspondence of the factors of high loadings to important acquisition dates, this enables a high amount of data and time reduction, while maintaining the overall accuracy.

Cultivation Calendar and Result of Factor-Analysis (FA)