http://orcid.org/0000-0003-2427-1713
Figures
Abstract
Building a density map over large areas could provide essential information of land development intensity and settlement condition. It is crucial for supporting studies and planning of human settlement environment. The Global Human Settlement Layer (GHSL) is a comprehensive data set of mapping human settlement at a global scale, which was produced by the Joint Research Centre (JRC), European Commission. The built-up density is an important layer of GHSL data set. Currently, the validation of the GHSL built-up area products was preliminarily conducted over the United States and European countries. However, as a typical East Asian region, China is quite different from the United States, Europe, and other regions in terms of building forms and urban layouts. Therefore, it is necessary to perform an accuracy assessment of GHSL data set in Asian countries like China. With individual building footprint data of 20 typical cities in China, this paper presents our effort to validate the GHSL built-up area products. The aggregation mean and neighborhood search based algorithms are adopted for matching building footprint data and the GHSL products, through the regression analysis at per-pixel level, the building density map in raster format are generated as validation data. The accuracy index of GHSL built-up area was calculated for the study areas, and the validation methods were explored for GHSL built-up products at large scale. The results show that the built-up layer aggregated by the building footprint have the highest correlation with the coarse resolution GHSL built-up products, but GHSL tends to underestimate the building density of low-density areas and overestimate the areas with high density. This study suggests that GHSL built-up area products in 20 representative Chinese cities of China could provide quantitative information about built-up areas, but the product accuracy still need to be improved in the regions with heterogeneous formations of human settlements like China. There is a big picture of mapping high accuracy built-up density of China with the training data set acquired by the study.
Citation: Liu F, Wang S, Xu Y, Ying Q, Yang F, Qin Y (2020) Accuracy assessment of Global Human Settlement Layer (GHSL) built-up products over China. PLoS ONE 15(5): e0233164. https://doi.org/10.1371/journal.pone.0233164
Editor: Shihe Fu, Xiamen University, CHINA
Received: October 16, 2019; Accepted: April 29, 2020; Published: May 29, 2020
Copyright: © 2020 Liu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The validation data is obtained from the Baidu map (https://map.baidu.com/). The authors confirm they did not have any special access to this third-party data which others would not have.
Funding: This research was funded by the National Key R&D Program of China (Grant No. 2018YFC0506901) and the 100 Talents Program of the Chinese Academy of Sciences. The acquisition is Yuchu Qin who is the Correspondence of this manuscript. The author Yuchu Qin defined the research theme and supervised the research work. And checked the experimental results and revised the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Land cover is an important factor of environmental studies of the earth surfaces [1–3], since the land cover/land use change, environmental pollution, land degradation and loss of biodiversity have become increasingly serious. Thus, timely and reliable global land cover data has become an important data set for ecosystem assessment, environmental modeling, etc. [4]. Urbanization is one of the most significant factors that human influence the land cover of earth surfaces. Presently, more than 50% of the world’s population live in urban areas, compared with which is only 30% in 1950, and it is projected that the urban population will account for 66% of the world population by 2050 [5]. Although cities cover only a relatively small portion of the earth surfaces, studies in urban areas play a crucial role in human housing demand, climate change and response, disaster risk prevention, urban development and other sustainable development goals [6].
Currently, it is well addressed that remote sensing technology is a promising solution for large-area observation [7–9]. Satellite remote sensing has become an important means of obtaining information on land surfaces [10–12]. As the high-resolution remote sensing satellite data can provide detailed urban surfaces, Xu investigated the land cover information extraction using IKONOS panchromatic data with 1m resolution [13]. In [14], the principal component analysis is employed to fuse the texture and structure features derived from Landsat-7 ETM+ panchromatic data to extract the building information. Currently, the Global Urban Footprint (GUF) dataset is produced for urban mapping, it is based on the satellite SAR imagery acquired by the German satellites TerraSAR-X and TanDEM-X. With a fully automated processing system, global coverage of more than 180,000 very high-resolution SAR images with 3m ground resolution, mainly acquired between 2010 and 2013, were processed, the scattering amplitude is combined with the derived texture information to depict the human settlement. In addition, auxiliary data such as digital elevation models were fused with the SAR images to improve the classification accuracy [15]. The Global Human Built-up And Settlement Extent (HBASE) Dataset is a global scale product derived from the Global Land Survey (GLS) Landsat dataset for the year of 2010 [16], the product is only for the mapping and monitoring of urbanization. The Global Human Settlement Layer (GHSL) produced by the Joint Research Centre (JRC) provides much more detailed information on the growth of buildings and populations over the past 40 years (1975–2015) [17]. The products contain comprehensive data layer for urbanization assessment, land cover change, urban planning and management [6], species changes studies [18], but the accuracy of the product needs further validation. With urban building density information, many applications would be able to carry out, e.g. the economic development of the city and the expansion of urban space. Combining with urban lighting data and urban traffic data, the future urban development can also be predicted, which is of great significance to integrate urban management and improvement of the urban environment [12, 19–21]. However, there is no well validated global-scale human settlement mapping products for ecological environment studies [22]. Therefore, the development of large-scale building density remote sensing products, the formation of large-scale and long time series of remote sensing mapping products has become an urgent need for both research and applications.
The GHSL built-up area products are promising data collection set for characterizing the built-up area at large scale. However, GHSL is an experimental product, it is currently validated only in the United States and European countries [10]. Due to differences in population density and building structure between Asia and Europe, the validation results of GHSL built-up area products in the United States and Europe cannot assure the reliability of the accuracy in Asia. Therefore, this paper focuses on the validation and analysis of the accuracy of GHSL built-up area products in 20 representative Chinese cities in China. It is expected to provide reference accuracy information for applications of GHSL data in China. Based on the maps of building footprint derived from open geospatial web service in China, i.e., Baidu Map. The accuracy of the GHSL built-up area products at 250m resolution and 1000m resolution in 20 typical Chinese cities across different provinces were quantitatively assessed. The accuracy of GHSL built-up products in China is evaluated by aggregating the building footprint into building density products with the same resolution as GHSL built-up products. The results demonstrate that there is a certain misestimation in GHSL products over 20 representative Chinese cities. The results are expected to provide quantitative accuracy information of the GHSL built-up products application in China.
Study area and data sets
Study area
To better represent the different urban patterns and building forms, this paper selected 20 typical cities as study area located in different administrative regions of China with varieties of economic developments, population densities and physical environments (Fig 1, Table 1). The building density patterns from GHSL in 20 cities of China were validated in this paper.
GHSL built-up areas products
GHSL is a global scale human settlement map product extracted from Landsat images, it is developed with a classification method based on symbolic machine learning [23, 24]. To utilize long term remote sensing data record, GHSL adopts images acquired by the Multi-Spectral Scanner (MSS) and Thematic Mapper (TM), Enhanced Thematic Mapper (ETM+), Operational Land Imager (OIL) and Digital Elevation Model (DEM) for characterizing human settlement. The time series consisted mostly of four epochs, 1975, 1990, 2000, and 2014. The built-up area products in each epoch are provided with the resolution of 38m, 250m and 1000m. This paper selects the latest GHSL product data—built-up area products in China with the resolution of 250m and 1000m in 2014 for the validation. Fig 2 illustrates the GHSL built-up area products in China.
(a) The GHSL built-up areas with 250m resolution, (b) The GHSL built-up areas with 1000m resolution.
Building footprint data
The accurate building footprint data acquired for the validation is obtained from the Baidu map (https://map.baidu.com/), the online web map service provided very high resolution building footprint layer as shown in Fig 3. “TIANDITU Imagery” is a comprehensive geographic information service website provided by the China National Surveying and Mapping Geographic Information Bureau. It is loaded with geographic information data covering the whole world in three modes: vector, image and three-dimensional. Therefore, this map is used as a reference. In this study, the building footprint layer selected for validation is acquired in 2017 from Baidu map due to the lack of data in 2014.
(a): TIANDITU Imagery. (b): Binary image of building footprint intercepted by Baidu map: Beijing.
Methods
Data processing
The building footprint layer of Baidu map was download and converted to raster data with the same resolution as the GHSL built-up area products. One problem in the spatial alignment of Baidu map with GHSL data is the various local projections of Baidu maps compared to a consistent global projection system of GHSL. We proposed and implemented a specific procedure to address this issue by the following five steps:
- Vector data generation: Obtain the building footprint data of Baidu map, get the binary map of building outline based on the characteristics of houses, and then get the vector data of the building footprint by mosaicking, filtering, vectorization, and geo-registration.
- Grid generation: Converting the building footprint vector data into raster data of which the spatial resolution is consistent with that of the GHSL built-up areas product requires vector grid data with the consistent grid position of the GHSL built-up areas. Therefore, the grid data of the GHSL built-up areas products are used to generate the required grid data with 250m resolution and 1000m resolution.
- Intersection operation: The grid of vector data obtained from step 2 is intercepted with the building footprint vector data to overlay the attributes of the building footprint vector data into each grid.
- Built-up areas calculation: Built-up area is usually expressed as the occupied proportion of the building footprint area in a unit area. The formula for built-up area is:
(1) where Pi,j is the proportion of built-up area in each grid, Si, is the building footprint area in the (i, j)th grid, and S is the grid area.
- Rasterization: Convert the building footprint vector data with building density attributes into raster images, and the built-up areas validation data is produced for this study.
The above steps are applied to process building footprint data over 20 cities in China. A square area with size of 18×18km is selected as the study coverage for each city.
Validation
In this paper, statistical histogram and linear regression are used to verify the results. From the histogram, the distribution of built-up areas is intuitively illustrated, and the gap between the data can be analyzed. Linear regression is applied to quantitatively analyze the dependency relationship. It is expressed as:
(2)
where e is error and obeys the normal distribution with a mean value of 0 [25].
In this paper, linear regression is used to analyze the relationship between the GHSL built-up area and the built-up area of Baidu maps. And the correlation coefficient is calculated as:
(3)
where Cov(X, Y) is the covariance of X and Y, Var[X] is the variance of X, and Var[Y] is the variance of Y.
Results and discussion
Based on the 2014 GHSL built-up area products, this paper processes the 2017 building footprint data obtained from Baidu map and obtains built-up areas validation data of 250m resolution and 1000m resolution. Figs 4 and 5 show the GHSL built-up area maps at the 250m resolution and 1000m resolution of 20 study regions in China, respectively. It can be observed that the density of buildings in the urban center is generally higher than that in the suburban and rural areas, which is more apparent in the product with 250m resolution [26]. Figs 6 and 7 show the built-up area maps obtained from Baidu map building outlines at 250m resolution and 1000m resolution in 20 study areas in China, respectively. Similar to the result observed from Figs 4 and 5, the building density in the urban center is higher. Due to the acceleration of urbanization, the city’s building density can vary in different periods. The building density in most cities in China will relatively increase in pace with the development. Generally, the urban suburbs could become the urban core area and the suburbs would expand outwards. Density can change more greatly at the edge of the study area, while in the urban center area it will not change significantly due to limited construction land. However, by comparing Figs 4 and 6, Figs 5 and 7, what we can observe is that the intensities of pixels in GHSL products in 2014 are relatively higher than the values of corresponding pixels in the validation data in 2017 which do not meet the urban development trend. Therefore, it is particularly important to verify the accuracy of GHSL built-up areas data.
Each study area is: (a) Baotou, (b) Beijing, (c) Changchun, (d) Chengdu, (e) Guiyang, (f) Hangzhou, (g) Kunming, (h) Nanning, (i) Qingdao, (j) Shanghai, (k) Shenyang, (l) Shenzhen, (m) Taibei, (n) Wuhan, (o) Urumuqi, (p) Xiamen, (q) Xi’an, (r) Xining, (s) Yinchuan, (t) Zhengzhou.
To reflect the differences between built-up areas of different cities and the differences between GHSL built-up area products (noted as GHSL) and built-up areas of the validation data (acquired from Baidu map and noted as BD) more intuitively. The histograms of different cities at different resolutions are retrieved by performing inter-partition statistics on each pair of GHSL and BD images, as shown in Figs 8 and 9.
(a), (b) with 250m resolution; (c), (d) with 1000m resolution.
(a), (b) with 250m resolution; (c), (d) with 1000m resolution.
Under the same resolution, the bias is existed in the estimations of urban density between BD and GHSL products across different cities. On the one hand, the GHSL built-up area products has a larger number of pixels with an urban building density higher than 0.8 (Fig 8). And the results are similar at GHSL products with different resolutions. It means that the GHSL tends to overestimate large cities (high-density areas) as a whole and underestimate small cities (low-density areas) compared to BD (Figs 8 and 9). The reason is that the GHSL products with 250m and 1000m resolution are interpolated based on the product of 38m resolution which is affected by the mixed pixel impact in the generation of product. Therefore, the building density of regions with low-density buildings will be underestimated and the areas with high-density buildings will be overestimated.
On the other hand, the comparison results of GHSL and BD are different between big cities and small cities under the same resolution. Large cities is generally estimated by GHSL as a high-density area, while small cities with relatively backward economy and population are estimated to be relatively uniform at different building density levels (Fig 8). However, BD generally recognizes the building density of big cities and small cities between 0–0.5, and the building density level of big cities is generally higher than that of small cities in different histogram intervals (Fig 9). This is because the number of pixels of high-density land and low-density land in large cities with higher development level is correspondingly more than that in small cities. Meanwhile, the BD employed for validation (Fig 9) shows that the regions with high building density (above 0.6) correspond to a small number of pixels. Due to the need for part of the land for lighting, urban greening and transportation between buildings, a certain proportion of land will be occupied, thus the histogram distribution of validation data, i.e. BD, is more in line with that of the urban built-up areas in China.
In terms of the comparison between GHSL and BD at different resolutions, both GHSL and BD achieve smoothing effect with the decrease of resolution, reducing the contrast between high-density areas and low-density areas (Figs 8 and 9). For GHSL, the extreme trend of both ends of 250m product is more obvious, while 38m high value data will be smoothed out on 1000m scale due to the change of interpolation unit range, thus reducing the proportion of high value overestimation (Fig 8). The smoothing effect is also applicable to BD products for its data is obtained by converting vector data to grid, thus the high-density area greater than 0.6 is smoothed out with the decrease of resolution to 1000m, and the overall pixel value is concentrated between 0–0.5 (Fig 9). In a conclusion, (1) BD data is more in line with China’s reality in terms of histogram distribution and comparison with the number of pixels across different cities. (2) There is a certain overestimation of the building density of GHSL products in large cities, compared with the estimated density of medium-sized cities, which has a higher consistency with BD. Therefore, BD products are of great value and significance in correcting the overestimation bias of GHSL to big cities.
The pixel values of GHSL built-up area products are higher at different resolutions (250m and 1000m), the data with a saturation value above 0.99 with the resolution of 250m and the data with a saturation value above 0.9 with the resolution of 1000m in GHSL products are considered for quality control according to the distribution of the statistics of GHSL products in all 20 cities. Then the value of GHSL built-up areas and validation data from the cell by pixel with the saturation value removed and the saturation value not removed is compared, as shown in Fig 10 and Fig 11, and the slopes and R2 are shown in Tables 2 and 3.
(a) Regression of products with 250m resolution; (b) Regression of products with 250m resolution and saturation values removed; (c) Regression of products with 1000m resolution; (d) Regression of products with 1000m resolution and saturation values removed.
(a) Regression of products with 250m resolution; (b) Regression of products with 250m resolution and saturation values removed; (c) Regression of products with 1000m resolution; (d) Regression of products with 1000m resolution and saturation values removed.
The results of the regression analysis (Figs 10 and 11) suggest that there is no significant difference in the regression slope between the saturated value and the unsaturated value at 250m resolution and 1000m resolution. However, the R2 of data with saturation values are higher than the R2 of data without saturation values, and the correlation of data set with 1000m resolution is higher than that with 250m resolution. The comparison between the results of 1000m resolution and 250m resolution products shows that the slope of 1000m resolution product is relatively smaller. In the regression analysis of this study, p < 0.001. From Table 2, we can find the regression parameters of cities with better economic development are lower, such as Beijing, Shanghai, Shenzhen and etc., which may result from the better development of economics, the rapid expansion of construction, the opposite economic level of the city and that the population is not floating in a small number of cities and the speed of construction expansion is relatively slow. Moreover, the trend line between GDP of each city and corresponding regression slope illustrates that the regression slopes of cities with higher GDP are lower, as shown in Fig 12.
In GHSL, the built-up area class is defined as the union of all the spatial units collected by the specific sensor and containing a building or part of it [10]. For the saturation values in the results, it can be explained by the confusion between bare soils in agricultural fields and built-up areas, the appearance similarity between the ridgeline and the building footprints, and that highway (especially asphalt concrete roads) can be misclassified as built-up areas when classifying built-up areas. GHSL built-up area products tend to overestimate the building density over areas with high density since the Landsat image used is of 30m resolution, under which condition single buildings and small settlement patterns surrounded by vegetation may be difficult to identify. The use of the data with high resolution may help in alleviating the problem of confusion between built-up areas and other types of land cover such as artificial open spaces, river gravel and sand dunes [10]. Therefore, the product at 1000m resolution has the best validation result. The validating data used in this paper contains only the building footprints, with no other non-building data existing. In summary, although the R2 of regression results are not very good, GHSL products have a certain effect on China’s instructions for built-up areas. In terms of the applications of this product in China, it may not be able to meet the high accuracy requirements of building density, but it is suitable for the wide-range study of low resolution.
Conclusions
Aims to assess the accuracy of GHSL built-up products in China, 20 typical cities across entire China were selected as study sites. The quantitative assessment of the GHSL built-up products was carried out over different types of cities, it is expected to provide a reference of applications of the GHSL products in the dense urban area over East Asia, especially in China. With the report of the assessment, we can conclude that the GHSL built-up products are a promising product for characterizing building density, and the pixel values have a good correlation with the ground truth. However, the comparison with the assessment in the United States and European countries suggest that the significant difference between the regression slopes. Built-uparea per tiles from European and United States as reference data and the GHSL layer has been compared with a regression slope of 0.2164 [17], which is lower than in China (Table 3). The reason for this phenomenon is that the cities in China have much higher building density and height, the shadow effect would be much significant, which will affect the estimation of building density with remotely sensed images [27]. In addition, the Landsat images with 30m spatial resolution were used by JRC in generating the GHSL built-up product. It means individual building and building clusters surrounded by dense vegetation cannot be accurately detected, and the building density was underestimated over where the areas with low buildings. The existence of the mixture pixels which contain signals of roads, bare grounds and buildings would also affect the accuracy of estimation.
In summary, the quantitative assessment over 20 cities in China suggests that the GHSL built-up products have a good correlation with ground truth, however, it is also observed that the products need to be further validated and improved in dense urban areas, especially in East Asia like China. The future work would be focused upon the investigation of estimation models of building density in both sparse and dense urban environment with time series and multi-source data, and eventually develop more generative models for generating building density products over large areas with an operational manner.
Acknowledgments
The authors gratefully acknowledge the Joint Research Centre (JRC) for the GHSL data. The authors would like to thank the Baidu map for the Building footprint data.
References
- 1. Yang Y.K., et al., Accuracy assessment of seven global land cover datasets over China. Isprs Journal of Photogrammetry and Remote Sensing, 2017. 125: p. 156–173.
- 2. Qin Y.C., et al., Synergistic application of geometric and radiometric features of LiDAR data for urban land cover mapping. Optics Express, 2015. 23(11): p. 13761–13775.
- 3. Lin Z.Y., et al., An active fire detection algorithm based on multi-temporal FengYun-3C VIRR data. Remote Sensing of Environment, 2018. 211: p. 376–387.
- 4. Ban Y.F., Gong P., and Gini C., Global land cover mapping using Earth observation satellite data: Recent progresses and challenges. Isprs Journal of Photogrammetry and Remote Sensing, 2015. 103: p. 1–6.
- 5.
United Nations, D.o.E.a.S.A., Population Division in World Urbanization Prospects: The 2014 Revision. 2015.
- 6. Chrysoulakis N., et al., A Conceptual List of Indicators for Urban Planning and Management Based on Earth Observation. Isprs International Journal of Geo-Information, 2014. 3(3): p. 980–1002.
- 7. DeFries R., Hansen M., and Townshend J., Global discrimination of land cover types from metrics derived from AVHRR pathfinder data. Remote Sensing of Environment, 1995. 54(3): p. 209–222.
- 8. McCallum I., et al., A spatial comparison of four satellite derived 1 km global land cover datasets. International Journal of Applied Earth Observation and Geoinformation, 2006. 8(4): p. 246–255.
- 9. Sun J. and Southworth J., Remote Sensing-Based Fractal Analysis and Scale Dependence Associated with Forest Fragmentation in an Amazon Tri-National Frontier. Remote Sensing, 2013. 5(2): p. 454–472.
- 10. Pesaresi M., et al., Assessment of the Added-Value of Sentinel-2 for Detecting Built-up Areas. Remote Sensing, 2016. 8(4).
- 11. Sharma R.C., et al., Global mapping of urban built-up areas of year 2014 by combining MODIS multispectral data with VIIRS nighttime light data. International Journal of Digital Earth, 2016. 9(10): p. 1004–1020.
- 12. Godefroid S. and Koedam N., Urban plant species patterns are highly driven by density and function of built-up areas. Landscape Ecology, 2007. 22(8): p. 1227–1239.
- 13. Xu B., et al., Comparison of gray-level reduction and different texture spectrum encoding methods for land-use classification using a panchromatic ikonos image. Photogrammetric Engineering and Remote Sensing, 2003. 69(5): p. 529–536.
- 14. He C., et al., Incorporation of Texture and Structure Information for Urban Building Detection by Using Landsat7 ETM~+ Panchromatic Image. Geomatics and Information Science of Wuhan University, 2004. 29(9): p. 800–804.
- 15. Esch T., et al., Urban Footprint Processor-Fully Automated Processing Chain Generating Settlement Masks From Global Data of the TanDEM-X Mission. Ieee Geoscience and Remote Sensing Letters, 2013. 10(6): p. 1617–1621.
- 16.
Wang, P., et al., Global Human Built-up And Settlement Extent (HBASE) Dataset From Landsat. NASA Socioeconomic Data and Applications Center (SEDAC): Palisades, NY., 2017.
- 17.
Pesaresi, M., et al., Operating procedure for the production of the Global Human Settlement Layer from Landsat data of the epochs 1975, 1990, 2000, and 2014. JRC Technical Report; European Commission, Joint Research Centre, Institute for the Protection and Security of the Citizen: Ispra, Italy., 2016.
- 18. Ye X.P., et al., Impacts of future climate and land cover changes on threatened mammals in the semi-arid Chinese Altai Mountains. Science of the Total Environment, 2018. 612: p. 775–787.
- 19. Shi L., et al., Spatial expansion remote sensing monitoring of special economic zones from 1973 to 2013. Journal of Remote Sensing, 2015.
- 20. Chen J., et al., The Study on Urbanization Process in China Based on DMSP/OLS Data: Development of a Light Index for Urbanization Level Estimation. Journal of Remote Sensing, 2003. 7(3): p. 168–175,T001.
- 21. Kubota T., et al., Wind tunnel tests on the relationship between building density and pedestrian-level wind velocity: Development of guidelines for realizing acceptable wind environment in residential neighborhoods. Building and Environment, 2008. 43(10): p. 1699–1708.
- 22.
Pesaresi, M., et al., GLOBAL HUMAN SETTLEMENT ANALYSIS FOR DISASTER RISK REDUCTION, in 36th International Symposium on Remote Sensing of Environment, G. Schreier, P.E. Skrovseth, and H. Staudenrausch, Editors. 2015. p. 837–843.
- 23.
Martino, P., et al., Atlas of the Human Planet 2016. Mapping Human Presence on Earth with the Global Human Settlement Layer.. EUR 28116 EN, 2016.
- 24.
Martino, P., S. VASILEIOS, and J.A. MARIA, Benchmarking of the Symbolic Machine Learning classifier with state of the art image classification methods—application to remote sensing imagery. 2015.
- 25. Rutherford A., Applied multiple regression/correlation analysis for the behavioral sciences. British Journal of Mathematical & Statistical Psychology, 2003. 56: p. 185–186.
- 26.
Bechtel, B., et al., TOWARDS CONSISTENT MAPPING OF URBAN STRUCTURES—GLOBAL HUMAN SETTLEMENT LAYER AND LOCAL CLIMATE ZONES, in Xxiii Isprs Congress, Commission Viii, L. Halounova, et al., Editors. 2016. p. 1371–1378.
- 27.
Nex, F., E. Rupnik, and F. Remondino, BUILDING FOOTPRINTS EXTRACTION FROM OBLIQUE IMAGERY, in Cmrt13—City Models, Roads and Traffic 2013, F. Rottensteiner, U. Stilla, and S. Hinz, Editors. 2013. p. 61–66.