M.Vogiatzis, C.G.Karydas, T.Alexandridis, G.Zalidis, and N.Silleos, 2008. ‘Αn object-based approach for wetland habitats inventory and assessment using ALOS AVNIR-2 and field data’, ALOS PI Symposium 2008, Rhodes, 3-7/11/08 AN OBJECT-BASED APPROACH FOR WETLAND HABITATS INVENTORY AND ASSESSMENT USING ALOS AVNIR-2 AND FIELD DATA Moschos Vogiatzis (1) , Christos Karydas (1) , Thomas Alexandridis (1) , George Zalidis (2) and Nikolaos Silleos (1) (1) Lab of Remote Sensing and GIS, Faculty of Agronomy, Aristotle University of Thessaloniki, Thessaloniki 54124, Greece, Email: mvogiatz@ktimatologio.gr (2) Lab of Applied Soil Science, Faculty of Agronomy, Aristotle University of Thessaloniki, Thessaloniki 54124, Greece, Email: zalidis@agro.auth.gr ABSTRACT A hierarchical process for a Mediterranean wetland habitats inventory and assessment using object-based image analysis applied to medium spatial resolution satellite is discussed. The method is based on spectral, contextual and spatial criteria, coupled with knowledge of ecological conditions of the wetland. The main objective of this research was to evaluate object-based methodology performance to assess wetland condition and discriminate wetland vegetation species and habitats as defined by Natura 2000 framework. A medium spatial resolution ALOS AVNIR-2 image, captured in July 2007, and coincident field data covering the wetland area of lakes Koronia-Volvi in Greece, was used in this analysis. An innovative sampling design is presented and a 4-level hierarchical structure classification scheme is introduced. Issues regarding shape and size of sampling units, configuration and their spatial random distribution by stratum are addressed. The preliminary results show excellent discrimination among most vegetation species, while habitat mapping has proved a time-consuming task. Thematic accuracy of habitats mapping will be assessed using field data acquired during intensive field survey. We believe that integration between hierarchical levels will further improve classification performance. The work presented was supported by the WETMUST project (Ιntegrated multiple level wetlands monitoring system using innovative technologies, INTERREG IIIB ARCHIMED) and European Space Agency (ESA). 1. INTRODUCTION In response of European Union (EU) Habitats Directive 92/43/EEC [1], Greece proceeded to the identification and mapping of habitats within the sites proposed to be included in Natura 2000 network [2]. The project employed standard aerial photointerpretation techniques to identify homogeneous areas assisted by field verification. The resulting maps displayed the spatial distribution of habitat types. However, continuous monitoring of habitats is required in Natura 2000 sites. A sound statistical strategy to collect basic information and measure specific variables is fundamental to the design of scientifically credible inventory [3]. The lakes-wetland ecosystem was conceived as an ideal application area due its complexity and heterogeneity, for developing an integrated geospatial framework at the site level to provide detailed information on the current condition of the wetland’s species and habitats and establish a mechanism to monitor the condition of species and habitats. Growing demands for geospatially explicit information over time are emerging as a result of complex sustainability challenges. Earth Observation (EO) has enormous potential as a source of relevant information. Several studies can be found in literature that use remote sensing data for mapping and monitoring of wetlands and their habitats ([4], [5], [6] and [7]). Further to the type of EO data employed, methodology is equally important especially if a complex environment has to be mapped; in this direction, various methods and measures have been developed ([8], [9]). Traditional pixel based supervised classification methods (e.g. maximum likelihood, etc.) assign a particular pixel to a class without taking into account the neighbourhood of the pixel. Recently, the object- based approach opens new opportunities for describing the complexity of wetland ecosystem resource attributes at multiple resolution levels and for advancing the design of current inventory and monitoring programs ([10], [11] and [12]). Object-based analysis is a relative new methodological approach in digital image analysis. This approach is closer to human cognitive process than the analysis based on individual pixels; the Cognitive Science describes information treatment in human brain as a conceptual and ablative process directed to objects. With regard to the perception of the environment, objects with meaning and importance resulting from analysis of image data represent very efficiently structures of the real world [13]. Image classification based on object-based analysis is a form of supervised classification, as it allows the user to educate the system either by sampling objects or by setting and combining rules. Object-based classification is particularly suitable