REAL Remote sensing identification and monitoring of abandoned land Tomáš Soukup, Lukáš Brodský, Václav Vobora GISAT Czech Republic
Abandoned land - intro Land abandonment is quite widespread process in all CEEC in last decades reflects substantial structural political and economical changes changes in ownerships changes in agriculture practices in unfavourable agriculture areas changes on agricultural commodity market leads to marginalization of certain agriculture regions represent a serious social, economical and ecological issue, even ecological impact assessment vary from habitat degradation, to positive wilderness
Abandoned land - policy framework policy targets related to land abandonment: sustainable use and management of abandoned areas prevention of abandonment land area enlargement backup of European policy documents: Agro-Environmental Regulation, Council Regulation No 2078/92, Rural Development Regulation [EEC, 1992], Council Regulation No 1257/99 [EEC, 1999]) backup of national policy documents: Biodiversity Strategy of Czech Republic (MŽP,2005), Post-accession Rural Policy 2004 2013 (MZE, 2005).
Abandoned land - information gap Nevertheless, the assessment of the policy strategies requires clear measures based on the figures on status and evolution of actual abandoned land area. Unfortunately, there are often currently no actual precise statistical data available on the extent of land abandonment Czech Republic - only rough estimates exist (e.g. total area of abandoned land about 300 000 ha, that is 7% percent of the total agricultural area (Ministry of Agriculture, 2001)
Abandoned land - project aim focus on agriculture abandoned land explore potential of the state-of-the-art remote sensing techniques for abandoned land identification and mapping methodology development in the general framework of Object Bases Image Analysis (OBIA) pilot areas for testing solution using object-based classification combining different strategies for abandoned land monitoring: spectral, textural, temporal and contextual ones. Implementation using the state-of-the-art DEFINIENS Enterprise software platform
Financial support The project was performed under the support of the EUPRO programme of Ministry of Education of the Czech Republic in the frame of the ERA-STAR GMES Pilot Project Brownfields.
Main activities Main activities in the REAL project can be summarizes as: creation of knowledge base for abandoned land - the preparation, processing and implementation of the knowledge base in a form useful for flexible exploration and assessment of various characteristics of abandoned land on EO data, hypothesis evaluation etc. deep understanding of individual signatures, their quantitative description and selection of signature subsets to be used in classification model making the classification model for abandoned land classification using OBIA approach (implementation in DEFINIENS software) producing classification test on a test areas including validation
Identification strategies General approach was based on different strategies for identification of abandoned land on EO data: Spectral NDVI index, NDGI index Temporal Textural Gradient index (Sobel Edge), Dissimilarity index (Haralick) Contextual and their combinations
Abandoned land - definition ambiguous definition of abandoned land most of the definitions are driven by legislative need for evidence of ownership, taxation purposes or the perspective of eligibility of subsidies examples: FAO defines abandoned land "... as unused for agricultural production or other agricultural purposes for at least the next 5 years" [FAO's World Census of Agriculture (FAO, 1986, 1995, FAO / UNEP, 1999)). CAP and agri-environmental measures support (Article 2.1 of Regulation (EEC) No 2078/92) refers to agricultural land, ".. that may be considered abandoned if it has not been the subject of an agricultural use or economic activity after a period of three years.
Abandoned land - definition From EO point of view abandoned land doesn t represent a single category of land cover and is more associated with changes in land use. These changes are then subsequently (and gradually) reflected in the land cover and its changes. From cultivated and cultural landscape over time it becomes a landscape of abandonment. From this point of view abandoned land represents more land use categories with heterogeneous spectral reflectance characteristics as it is composed by mixture of grassland/pastures and shrubs colonized by herbaceous and bushy vegetation in various stages of succession (by various land cover). More, particular species composition also reflects specific topographic, climatic and pedology conditions of the locality.
Abandoned land - definition technical definition of abandoned land from the remote sensing perspective abandoned land/potentially abandoned land = heterogeneous areas dominated by grasses, herbs and bushy communities in various stages of succession, with no evidence of agricultural use, or economic activity with a significantly different manifestations and evolution in the context of the surrounding cultural landscape
Typology used typology of abandoned land was prepared based on structure and composition of abandoned sites abandoned land decomposition to types based on field work and later on to base classes for classification purposes. 8 abandoned land sub-types has been defined in the REAL project based on various share and composition of 3 base classes
Test areas Field data collected for: Abandoned areas reference areas Clear classes reference areas to fill knowledge database
Knowledge DB technical implementation systematic DB of land abandonment indices technically implemented in the PostgreSQL relational database with the POSTGIS extension. can be accessed either directly through DB or GIS connectivity for a quick "exploring, the content of DB is also possible to access via web interface fast and flexible browsing capabilities, according to the entered search criteria, hypothesis testing, etc. open web client allows modification or extension to new locations, for new signatures, but also to a new functionality in terms of output definitions and user functions.
Knowledge DB Contents n-dimensional database of indices (~2000 combinations) spectral measures + composite indexes NDVI, NDWI, NDGI textural measures temporal measures contextual measures
Knowledge DB web front-end
DEFINIENS classification base Transferable classification model based on descriptive knowledge (description of objects) based on designed and evaluated customized features procedural knowledge base featuring semantic grouping and contextual aggregation with defined parameters (aggregating of primary classes to heterogeneous super-classes based on class specifications) which is in particular applicable for abandoned land areas.
Test classification - Example of results test area: Mladá Boleslav data used : VHR data (IKONOS) potential abandoned land share in test locality 4,97% Estimates of the Ministry of Environment [MoE, 2001] talk about 7% per cent of agricultural land area, This would mean for our locality share of around 4.5%.
Test classification - Example of results Okolí obce Kněžmost část Násedlnice (okr.mladá Boleslav)
Test classification - Example of results Okolí obce Dobrovice - Holé vrchy (okr.mladá Boleslav)
Test classification - Example of results Okolí obce Bakov nad Jizerou - část Studénka (okr.mladá Boleslav)
Test classification - Example of results Okolí obce Bakov nad Jizerou část Mráchov (okr.mladá Boleslav)
Abandoned land all types Test classification Overall accuracy
Test classification accuracy Abandoned land type 03 (OP3) *
Test classification accuracy Abandoned land type 04 (OP4) *
Conclusions potential abandoned land monitoring on EO data is technically feasible, but economically quite demanding as texture information is indispensable -> multitemporal VHR data twofold results: 1] identification of potential sites of abandoned land via elementary base classes of abandoned land subtypes 2] degree of association to the abandoned land class defined. combining these results can be used to identify hot-spots of potential abandoned land overall accuracy is relatively high (~80%, Kappa Index of Agreement (KIA) around 0.70), but considerable variability exist for different types of abandoned land from defined typology
Outlook total area is rather overestimated, than underestimated for hot-spots commissions can be better treated then omissions potential areas identification for biofuels hype, agriculture intensification improvements: better mask of urban (FTSP?) & SLUs better timing for multitemporal VHR data hyperspectral data Emap? potential abandoned land to be confirmed using additional information (ownership data)
REAL Remote sensing identification and monitoring of abandoned land Tomáš Soukup, Lukáš Brodský, Václav Vobora GISAT Charkovská 7 101 00 Praha 10 tel/fax: (+420) 27174 1935, 27174 1936 email: tomas.soukup@gisat.cz http://www.gisat.cz