Mining Multitemporal In Situ Heterogeneous Monitoring Information for the Assurance of Recorded Land Cover Changes

Abstract

We present a data mining methodology to filter and validate land cover change detections obtained from multitemporal in-situ surveys. As in-situ data we use the measurements from the European Land Use and Coverage Area frame Survey (LUCAS), which provides images with standardized metadata about land cover and land use within the whole territory of the European Union. Multitemporal LUCAS surveys present an anomaly in the amount of land cover changes that disagree with the estimated by experts. Therefore, our methodology analyses the available data in order to explain the existing irregularities in them. The initial step of our methodology is based on database query refinements. The data mining methodology continues with an image analysis process. This analysis calculates similarity measures of the multitemporal images which are used to identify the potential misclassifications. The final step involves a Geographic Information System (GIS) based on web technologies. By defining different color codes assigned by the similarity measures, the system represents the examined points on a digital Earth globe. There, a user can easily discriminate potentially misclassified points for subsequent detailed analysis or corrections. The final output of the methodology shows remarkable results for detecting misclassified land cover changes.