Monitoring Agricultural Activities Using Multi-temporal Asar Envisat Data

Agricultural activities affect strongly their environment. Therefore and because of their other economic, industrial and political effects, monitoring agricultural areas is increasingly demanded. Space borne remote sensing is a very cost effective and beneficial means for monitoring and mapping of earth’s surface. Due to independence from weather conditions, radar data are regularly available and facilitate a continuous monitoring of almost any area on the earth. In the framework of an ESA pilot project (AO335), methods are investigated aiming at reliable, cost efficient, and continuous monitoring of cultivation activities. A time series of doal polarization (VV/VH) of ENVISAT is used for the analysis. In addition, ground truth data are gathered by field acquisitions. The methodological approach for the monitoring task consists of supervised classification based on field data. For classification, some conventional methods and support vector machine (SVM) are applied and evaluated. In addition, the influence of some conventional despeckle filters on the classification accuracy is investigated. Results show that despeckling images by some filters before classification improves accuracy of multi temporal classification.. Matching of time series data to phenological period of crops is evaluated and compared with results of classification based on the entire data set. The results show that applying proper sets of data results in an exterior accuracy of over 80% on control fields. Overall accuracy of Maximum Likelihood and SVM are close to each other but their accuracy over control fields changes diversely. On the other hand, Maximum Likelihood is efficiently more accurate if data are matched to crop calendar. This is not the case for SVM.