Fire monitoring and management in Mediterranean countries such as Greece is of paramount importance. Almost every summer massive forest res break out, causing severe destruction and even human life losses. Thus, European initiatives in the area of Earth Observation (EO), such as GMES SAFER4, have supported the development of relevant operational infrastructures. In the context of the European project TELEIOS5, we aim at developing a re monitoring service, that goes beyond operational systems currently deployed in various EO data centers, by building on Semantic Web and Linked Data technologies. In this demonstration we present the re monitoring service that we have implemented using TELEIOS technologies focusing on its Semantic Web related functionality. The service implements a processing chain where raw satellite images are analyzed and hotspots (pixels of the image corresponding to geographic regions possibly on re) are detected. The products of this analysis are encoded in RDF, so they can be combined with auxiliary linked geospatial data (e.g., GeoNames, OpenStreetMap). By comparing detected hotspots with auxiliary data their accuracy can be determined. For example, hotspots lying in the sea are retrieved and marked as invalid. Additionally, we can combine diverse information sources and generate added-value thematic maps which are very useful to civil protection agencies and re ghting teams during emergency situations. In the rest of this demo paper we rst describe in short the contributions of project TELEIOS. Then, we present the developed re monitoring service and its advances compared to relevant deployed services. Finally, we describe how we plan to present this service through a live demonstration.
? This work has been funded by the FP7 project TELEIOS (257662). 4 http://www.emergencyresponse.eu/ 5 http://www.earthobservatory.eu/
TELEIOS is a recent European project that addresses the need for scalable access to PBs of EO data and the e ective discovery of knowledge hidden in them. TELEIOS started on September 2010 and it will last for 3 years. In the rst 18 months of the project, we have made signi cant progress in the development of state-of-the-art techniques in Scienti c Databases, Semantic Web and Image Mining and have applied them to the management of EO data.
We have developed SciQL [
We have also developed the model stRDF, an extension of the W3C
standard RDF for representing time-varying geospatial data [
The National Observatory of Athens (NOA) operates an MSG/SEVIRI satellite acquisition station, and has developed a real-time re hotspot detection service for e ectively monitoring a re-front. We present this service graphically in Figure 1 and explain below in some detail the improvements that we have achieved by using TELEIOS technologies.
On a regular basis (5 or 15 minutes) satellite images arrive at the acquisition
station and are stored as arrays in MonetDB. The arrays are processed with a
series of SciQL queries (for cropping, georeferencing, and hotspot detection) and
shape les describing the detected hotspots are generated for each acquisition.
Because of the low spatial resolution of the SEVIRI instrument, possible errors
in image georeferencing, and potential weaknesses of the algorithms in [
Semantic Annotation Hotspots Detection
SQL SciQL stSPARQL NOAOntology
Linked Geospatial
Data
The main problem with the product accuracy is the existence of false alarms in the re detection technique. For example, hotspots shown to be occurring in the sea or in locations with inconsistent land use (e.g., urban areas) should be considered false alarms instead of forest re emergency situations. To query generated data using stSPARQL and combine it with linked data, we derive stRDF triples from the generated shape les. The derived triples mainly hold information about the coordinates of detected re location, the date and time, and the con dence level of the detection for each hotspot. We execute stSPARQL updates which compare the hotspots with two RDF datasets and mark as false positives the hotspots that lie in the sea or in locations with inconsistent land use. The datasets that we use are: (i) a dataset describing the coastline of Greece8, and (ii) a dataset describing the Greek environmental landscape9.
Another problem is spatial and temporal inconsistencies in hotspots generated by the processing chain due to using a single image acquisition and not using information from previous acquisitions. A simple heuristic we use is retrieving hotspots that were detected at least once during a speci c time period (e.g., half hour) but they were not detected in the last acquisition. In this case we add a virtual hotspot for the last acquisition with a con dence level equals to the average con dence level of the real detections during the last half hour.
Finally, the need to generate added-value thematic maps is addressed. The Linked Open Data Cloud supplies an abundance of datasets, in addition to internal EO data, that cover a large variety of geospatial entities, ranging from negrained geometric objects like re stations, to coarser ones like countries. So, instead of manually combining heterogeneous data, a user can pose an stSPARQL query for each layer that she wants to depict in a map and overlay the retrieved data using the ability of Strabon to expose data in KML or GeoJSON. Although this service has been designed for Greece, it can be applied to any geographic area due to the generality of the used technologies(e.g., RDF, linked data, KML). 8 This dataset has been compiled in the context of TELEIOS and is available from http://geo.linkedopendata.gr/coastline_gr/. 9 http://geo.linkedopendata.gr/corine/
The demonstration consists of three parts. First, the user will start an instance of the processing chain described above and browse its results in the GUI of the application (shown in Figure 2). The user can also use the search functionality or pose stSPARQL queries to retrieve re products of previously executed instances of the processing chain. Second, the demonstration focuses on the improvement of the accuracy of the re products. We will demonstrate how stSPARQL update statements and linked geospatial data are used in order to increase the accuracy of derived re products. Finally, the creation of added-value thematic maps by combining information from di erent data sources will be demonstrated.