An Ontological System for Interoperable Spatial Generalisation in Biodiversity Monitoring

/ Matt Artz

cgComputers & Geosciences, Volume 84, November 2015, Pages 86–95

By Simon Nieland, Niklas Moran, Birgit Kleinschmit, and Michael Förster

“Semantic heterogeneity remains a barrier to data comparability and standardisation of results in different fields of spatial research. Because of its thematic complexity, differing acquisition methods and national nomenclatures, interoperability of biodiversity monitoring information is especially difficult. Since data collection methods and interpretation manuals broadly vary there is a need for automatised, objective methodologies for the generation of comparable data-sets. Ontology-based applications offer vast opportunities in data management and standardisation. This study examines two data-sets of protected heathlands in Germany and Belgium which are based on remote sensing image classification and semantically formalised in an OWL2 ontology.

Two step approach of the modified SPARK methodology. Part A shows the kernel reclassifier with an schematic 5 x 5 kernel. Input is a classification result, which illustrates the classified categories in different colours (see Part A and Part B(a)). The kernel includes three classes (sand (S), forest (F), and heath (H)). The kernel reclassifier performs a calculation of the class frequency in a frequency table and compares the outcomes to the associated rule. If the kernel corresponds to all formulated rules the centre pixel (grey) can be assigned. The interpolator eliminates patches that have not been assigned (b) and patches that are under a certain MMU (c) and interpolates gaps in a nearest neighbour interpolation procedure (d). (For interpretation of the references to colour in this figure caption, the reader is referred to the web version of this paper.)

Two step approach of the modified SPARK methodology. Part A shows the kernel reclassifier
with an schematic 5 x 5 kernel. Input is a classification result, which illustrates the classified
categories in different colours (see Part A and Part B(a)). The kernel includes three classes (sand
(S), forest (F), and heath (H)). The kernel reclassifier performs a calculation of the class
frequency in a frequency table and compares the outcomes to the associated rule. If the kernel
corresponds to all formulated rules the centre pixel (grey) can be assigned. The interpolator
eliminates patches that have not been assigned (b) and patches that are under a certain MMU (c) and
interpolates gaps in a nearest neighbour interpolation procedure (d). (For interpretation of the
references to colour in this figure caption, the reader is referred to the web version of this
paper.)

“The proposed methodology uses semantic relations of the two data-sets, which are (semi-)automatically derived from remote sensing imagery, to generate objective and comparable information about the status of protected areas by utilising kernel-based spatial reclassification. This automatised method suggests a generalisation approach, which is able to generate delineation of Special Areas of Conservation (SAC) of the European biodiversity Natura 2000 network. Furthermore, it is able to transfer generalisation rules between areas surveyed with varying acquisition methods in different countries by taking into account automated inference of the underlying semantics. The generalisation results were compared with the manual delineation of terrestrial monitoring. For the different habitats in the two sites an accuracy of above 70% was detected. However, it has to be highlighted that the delineation of the ground-truth data inherits a high degree of uncertainty, which is discussed in this study.”