Urban 2

Dynamically changing urban agglomerations in developing and emerging countries suffer from social and environmental challenges, due to a growth population and socioeconomic developments. Within the RapidPlanning (funded by the Federal Ministry of Education and Research, Germany) research consortium, which strives for trans-sectoral infrastructure planning solutions, concerning water, waste-water, energy and solid waste disposal, quantified and objective information are required for the planning process. The project has a close cooperation with the planning administrations of the cities of Kigali (Rwanda), DaNang (Vietnam) and Asyut (Egypt). The outcomes of the project will be implemented in the local planning decision making processes. The spatial distribution of the population and the socioeconomic background serve as a basis for the planning of supply and disposal infrastructure. Hence, qualitative and quantitative information of the settlement structure is necessary. High-resolution multispectral imagery allows the identification of relevant built-up structures and adjacent landuse. Since the availability of aerial images, as well as of LIDAR-based digital surface models (DSM) often is limited in dynamically changing cities, satellite-based data are a cost-effective alternative to identify urban entities.

A Pléiades tri-stereoscopic scene (2015-08-09) was acquired for the city of Kigali, to conduct a case study of the proposed methodology. A DSM was processed from the data. The interpolating of the elevation of roads and open spaces, which were identified in a preliminary classification of the multispectral Pléiades scene, resulted in a bare earth model. This allowed the calculation of a normalized DSM. The validation of the computed building heights with ground reference measurements indicate a general underestimation. Detached buildings, as well as mid to high-rise building structures are relatively well distinguishable from surrounding landcover, while the differentiation by height of dwellings in unplanned settlements was only in a few cases possible. We applied an object-oriented approach, with the pan-sharpened multispectral information of the Pléiades scene. Ground reference information of 601 buildings were used for training and validation. As a first step, a landuse classification was conducted, utilizing spectral and geometrical values, as well as elevation and relational information. The classification was pursued with a Support Vector Machine (SVM) model, using an rbf-kernel (radial basis function). As second step, a rule-based decision process was applied to identify a basic building typology (e.g. unplanned single-family houses, single/two family detached houses, mid to high-rise apartment buildings, etc.).

The analysis led to a promising identification of most building-types. The discrimination of single dwellings in unplanned settlements showed inconsistent results. This is because of a combination of high density and low building height, as well as the spectral similarity of rusty corrugated metal roofs to adjacent dirt roads and bare soil. The next stages of processing will be comprised of including supplementary data like cadastral data for the city of Kigali and the application and evaluation of the approach for the cities of DaNang and Asyut.

The segmentation of Remote Sensing images of the visible spectral range into meaningful regions has a long tradition in the field of image processing, analysis and interpretation. It is the necessary first step to assign sets of pixels to certain imaged objects and is widely used in Earth observation tasks. Thus, image segmentation forms the base for many classification and change detection tasks.

In this paper, we propose a novel and generic approach, which abstracts from the pixel ordering of an image but starts at the level of superpixels. Following Ren and Malik, a superpixel is said to be a local, and coherent set of (sub) pixels, “which preserve most of the structure necessary for segmentation at the scale of interest.” [1]. Superpixels may be derived from an image using the Watershed Transform (WT) as well as the SLIC approach [2, 3]. We propose to store the superpixels in novel structure, which we refer to as a region shell graph (RS-graph). This graph does not only hold the geometry and adjacency of the superpixels, but of the boundaries (topological: shells) in between. We implement various morphological operations on the RS-graph, e.g. the merging of superpixels according to cost measures, based on statistical and geometrical properties of the superpixels or their shells. The successive application of these operations results in a fully navigable irregular image pyramid of RS-graphs. The top layer of this pyramid may represent the final segmentation. Unlike other approaches, the RS-graph can be constructed in a concurrently, which fits modern machines’ architectures and speeds up the process. Many of the operations on the RS-graph may also be performed in parallel. Moreover, the explicit inspection of partial results is supported by means of a free navigation though the irregular pyramid.

To demonstrate the use of the RS-graph and the irregular pyramid for monitoring of urban areas, we have selected the field of road segmentation and road tracking on publicly available Remote Sensing images. This also allows an indirect urbanization monitoring, since the expansion or development of a road network is a promising indicator. Compared to other pixel-based road detection approaches (like e.g. [4]), our approach combines statistical information about the superpixels and the boundaries following the Gestalt law of good continuation. Experiments show that our combined approach results in a higher overall accuracy and is invariant to the paving and the width of the roads. It also provides a partial invariance to occlusions of the road, e.g. by trees in rural areas. These properties strongly suggest the use of our approach for semi-automatic monitoring or change detection of urban areas. Three demos of our approach are shown in the PDF attached to this abstract submission.

This research study has been performed in cooperation with the Centre for Science and Peace Research (ZNF) in Hamburg/Germany to detect and monitor infrastructural changes in rural areas.

REFERENCES
[1] Xiaofeng Ren; Malik, J., "Learning a classification model for segmentation," in Computer Vision, 2003. Proceedings. Ninth IEEE International Conference on , vol., no., pp.10-17 vol.1, 13-16 Oct. 2003

[2] Jos B.T.M. Roerdink; Arnold Meijster. 2000. The Watershed Transform: Definitions, Algorithms and Parallelization Strategies. Fundam. Inf. 41, 1,2 (April 2000), 187-228.

[3] Achanta, R., Shaji, A., Smith, K., Lucchi, A., Fua, P., & Susstrunk, S. (2012). SLIC superpixels compared to state-of-the-art superpixel methods. Pattern Analysis and Machine Intelligence, IEEE Transactions on, 34(11), 2274-2282.

[4] Shukla, Vandana ; R.Chandrakanth ; R.Ramachandran: Semi-Automatic Road Extraction Algorithm for High Resolution Images Using Path following Approach. In: Proceedings of the Third Indian Conference on Computer Vision, Graphics and Image Processing, 2002

The Land Information System Austria (LISA) is an initiative for the implementation of a homogeneous land cover and land use solution throughout Austria. Due to the collaborative development efforts, LISA meets the needs of governmental institutions on state, provincial and municipality level. The conceptual basis consists of an object oriented data model that describes classes of land cover and land use together with a number of attributes. The data model implements the EAGLE (EiONET Action Group on Land Monitoring in Europe) data model on national scale and provides all necessary components to be transformed into an INSPIRE compliant dataset.

The land cover part of the data model defines 15 classes (e.g. “Buildings”, “Other constructed areas”, “Bare soil”, “Screes”), which describe the appearance on the earth’s surface. For each of the classes a minimum object size was defined, which starts for many classes at an area of 25 m². Objects below this size are merged with neighbouring objects in order to reduce the number of sliver polygons and to optimise the look-and-feel of the resulting map. Furthermore attributes were defined for all classes which provide additional information for each land cover object (e.g. area, height, biomass index).

The LISA data model and the production chain for operationally mapping land cover have been developed in the frame of two projects (2009-2012) as part of the Austrian Space Applications Programme (ASAP) of the Austrian Research Promotion Agency (FFG). These activities were continued within the ESA funded CadasterENV Austria project (2012-2015). While the first projects focused on creating the land cover maps with orthophotos and airborne laserscanning data, it was the aim of CadasterENV Austria to expand the production methodology for the ingestion of Pléiades imagery. While orthophotos are produced operationally each year for a third of Austria (resulting in a three-year-cycle for the entire country), acquisitions of the Pléiades constellation are possible in a more flexible and cost-efficient manner for ad-hoc requests and smaller areas. As a result up-to-date VHR land cover maps may be produced as flexible as Pléiades images can be acquired. It is important to mention that the LISA land cover maps are purely derived of remote sensing data and thus do not depend on additional thematic GIS data.

In the CadasterENV Austria project the LISA processing chain – originally developed based on orthophotos – was adapted for the use of Pléiades data. More than 5,800 km² of Pléiades data were acquired for the cities of Linz, Salzburg and Vienna and their urban hinterland. For the tasking of these satellite scenes it was crucial to specify an incidence angle near nadir in order to avoid tilting of buildings and trees in case of incidence angles larger than 10 degrees. This Pléiades coverage was complemented with orthophotos as well as WorldView-2 imagery and used for the production of VHR land cover maps for the main urban areas of Austria, totalling in more than 10,000 km², which is about an eighth of entire Austria and approximately one third of the permanent habitable area. The products were completed in October of 2015 and can be freely downloaded from the project’s website (www.landinformationsystem.at).

In recent decades, urban sprawl became a serious European-wide problem, not only due to total land taken, but also due to its spatial distribution patterns and quality of the land consumed. Land use efficiency is becoming a prime political objective at both European and city level, and the EU Land Communication aims to establish “zero net land take” across the EU by 2050. Land is a finite resource and therefore the sprawl has to be regulated. This can be realized through careful management of urban land, applying the concept of land recycling. Only such approach can assure sustainable development of European landscape in the long term perspective. The EU funded URBIS project (ICT PSP 2014–17) targets these issues and focuses on investigation of vacant land potential in urban areas, and the opportunities for previously developed land or brownfield to support urban regeneration safeguarding greenfield sites, as well as on investigation of urban green systems. URBIS delivers methodologies and tools to provide accurate up-to-date intelligence that is comparable across European cities to support the definition and implementation of sustainable planning and governance strategies in cities and city-regions throughout Europe.

URBIS services are built on data acquired in frame of the Copernicus programme. This generates large number of standard open-data, Earth Observation datasets as well as standard services (Urban Atlas, High Resolution Layers (HRLs)), which can be utilized for urban land recycling policy support. First, analysis based directly on Urban Atlas thematic layer are implemented, to obtain the overview about the amount and distribution of potential development sites. Second, advanced image analysis methods are employed, to gather additional information, which is not available in Urban Atlas thematic layer, in particular about internal structure of the sites. These techniques, utilizing both pixel and object based image analysis (OBIA) approaches, are dealing with multiple spatial, spectral and textural image characteristics. They are applied preferably on SPOT5 imageries, acquired for European Urban Atlas mapping. As a result, the inventory of potential development and green sites in urban areas, including their characteristics detectable from EO data, like level of sealing, amount and type of vegetation cover etc., has been obtained. Preparing detailed methodology, the robustness is an important factor, to assure the applicability of workflow for potential European-wide analysis in the future.

The implementation is demonstrated on three pilot sites - Greater Amiens (FR), Osnabrueck (DE) and the Moravian-Silesian Region (CZ). Baseline services for the year 2012 and backdating towards 2006 and cca 1996 will be implemented. Then, thematic services will be prepared, exploiting results of baseline and update services, to provide information about the urban re-development potential directly to the users. Character of URBIS services has been tailored to meet requirements of end-users, which were set in close cooperation with pilots and with an external Stakeholder Board. Both strategic users at different levels (local and regional authorities, European and national agencies in charge of urban planning) and operational users (such as industrial estates operators or private land developers which can be interested in information on suitable vacant sites) are targeted.

Monitoring of changes within urban and suburban environments is still an important research topic, primarily due to the requisite from local authorities to systematically observe the development of cities. The detection of such changes in such environments was achieved in the past commonly by visual interpretation or implementation of complex processing chains on limited number of aerial or optical satellite imagery, eventually hindering their use by non-experts.  
The ever increasing availability of satellite Synthetic Aperture Radar (SAR)data, the enhanced spatial and temporal resolution of modern systems as well as their advantages over the respective optical ones, regarding the independency from weather conditions, render them an essential mean for monitoring purposes.
The objective of this study is to introduce an innovative model for monitoring cities providing information on the location and time where changes took place, utilizing space-borne SAR data and tailored statistical time series analysis. The study utilises the case of Athens (Greece) as a paradigm for further urban change analysis.
The metropolitan area of Athens, as any other modern city, is continuously evolving as a result of its increasing population over the last 20 years, whereas significant/considerable changes in its infrastructure were introduced by the organization of the Olympic Games, which took place in 2004. Although the focus here is given on the Olympic Games of 2004, the period examined extends from 1992 to 2010, covering almost 18 years of observations, exploiting the entire SAR archive of ESA including ERS-1, -2 and ENVISAT missions.
Processing of SAR data was performed using the GAMMA software packages, while post-processing and statistical analysis of the results was held by developing appropriate code in IDL. Additionally QGIS, an open source Geographic Information System, was employed for further geographical analysis. State-of-the-art processing algorithms were applied to ensure the accurate retrieval of the SAR backscattering coefficient, accounting for the various factors, such as the changes of incidence angle within the scenes, the effect of topography and others. In addition, the time series of the backscattering coefficient was used to extract the statistical figure of temporal variability, a metric commonly employed in the literature to characterise changes in both time and space.
The results obtained indicate a significant correlation between the occurrences of human-induced changes with the evolution of temporal variability through time. Detection of activities related to the preparation phase prior to the Olympic Games, in terms of construction works and land use changes, were well identified, while the time span of each activity was also estimated. Nevertheless, the proposed monitoring approach do suffer from inadequate and non-systematic image acquisitions, often resulting into significant observation gaps. The above fact leads to a reduced capacity to precisely constraining the time span of the detected change. It is foreseen that operational missions with systematic acquisition plan, as the Copernicus Sentinel-1, will allow overcoming such limitations.
Finally, part of the study concerned the creation of a technical framework, to allow the systematic generation of EO-based change detection products, which are tailored for non-specialist users. Such products embed simple cartographic information and can potentially evolve into an operational service, particularly adapted to be used directly by local authorities and urban planners.