Keywords

1 Global Urban Land

Product

LULC thematic

Dates

1980, 1990, 1995, 2000, 2005, 2010, 2015

Formats

Raster

Pixel size

30 m

Theme

Extent of artificial areas

Extent

Global

Updating

Not planned

Change detection

Yes

Overall accuracy

Expected to be > 80%

Website Language English

Website of reference

http://www.geosimulation.cn/GlobalUrbanLand.html

Download site

http://www.geosimulation.cn/GlobalUrbanLand.html

Availability

Format(s)

Open Access

.tiff

Technical documentation

Liu et al. (2018)

Other references of interest

–

Project

Global Urban Land, also referred to as Multi-temporal Global Impervious Surface (MGIS), is a project developed by researchers from different Chinese universities (Sun Yat-sen, East China Normal, Guangzhou and Jiangsu Normal) to create a high-resolution multi-temporal urban land dataset. They aimed to provide high-resolution data about urban areas at multiple dates, which could be useful for those studying urbanization and the impact of artificial surfaces and human activities on the environment.

In this dataset, urban land is understood as an impervious surface. It can therefore be assimilated to all the datasets mapping artificial or impervious surfaces, such as GAIA. Initially, the dataset was produced for the period 1990–2010, with maps every 5 years. However, it has since been updated, with new data for the years 1980 and 2015.

Production method

Global Urban Land is obtained through an index-based method that automatically predicts urban land: the Normalized Urban Areas Composite Index (NUACI). The index, implemented through the Google Earth Engine (GEE) platform, uses Landsat imagery and DMSP-OLS nighttime lights images as inputs.

To calibrate the index, the world was stratified into different urban ecoregion categories, according to the particular physical and socioeconomic characteristics of each urban region. Three indexes (NDWI, NDVI and NDBI) were extracted from Landsat imagery to calculate the NUACI. In addition, a binary mask was obtained by segmenting DMSP-OLS nighttime lights images into urban and non-urban by applying a specific threshold. On the basis of these data, the NUACI index was calculated, obtaining a raster showing the percentage of impervious surface area per pixel.

The final Global Urban Land dataset was obtained after applying region-specific segmentation thresholds to the NUACI images showing the degree of imperviousness. After this step, a binary urban/non-urban map was generated.

For the calibration of the NUACI index, as well as for the application of segmentation thresholds, cities were randomly assigned to three equal-sized groups: centroid sites, threshold sites and testing sites. Different criteria for index calibration and threshold segmentation were decided for each type of site.

Product description

The Global Urban Land dataset can be downloaded from three different servers: Baidu Drive, Google Drive and FTP. From them, users will be able to separately download the dataset for each of the available years of reference. For each year, there is a compressed folder (.zip) containing the whole dataset distributed in tiles.

An auxiliary vector file (.shp) is provided to help users identify the number of the files corresponding to their area of interest (field “grid_id”). The scientific paper presenting the dataset is also available for download, together with a text file with relevant technical information about the product and the reference data used to produce the dataset for the initial period 1990–2010.

Downloads

Global Urban Land 2010

– Raster files with the extent of the artificial surfaces for each tile into which the dataset was divided (.tiff)

Legend and codification

Code

Label

Code

Label

0

Non-urban land

1

Urban land

Practical considerations

The authors have identified several uncertainties and limitations in the dataset. The 1990 map has missing data areas due to the lack of Landsat imagery or reference data for these areas. The binary mask used to create the dataset may also introduce some uncertainties, as it was unable to detect some urban infrastructure. In addition, the accuracy of the dataset is relatively low in arid and tropical areas. The authors also described the limitations associated with a binary (urban/non-urban) mapping approach, which oversimplifies the real situation being mapped.

2 GHSL (Global Human Settlement Layer)—Built-up Area

Product

LULC thematic

Dates

1975, 1990, 2000, 2014

2016

2018

Formats

Raster

Pixel size

10 m (2018)

20 m (2016)

30 m, 250 m and 1 km (1975–2014)

Theme

Extent of built-up areas (1975–2014, 2016)

Built-up areas probability (2018)

Extent

Global

Updating

Expected

Change detection

Yes, except for the 2016 and 2018 layers

Overall accuracy

Expected to be > 89% (2014)

Website of reference

Website Language English

https://ghsl.jrc.ec.europa.eu/index.php

Download site

https://ghsl.jrc.ec.europa.eu/download.php

Availability

Format(s)

Open Access

.tiff

Technical documentation

Corbane et al. (2018), (2019a), (2019b), (2021), Pesaresi et al. (2016a)

Other references of interest

Joint Research Centre (2020), Melchiorri et al. (2018), (2019), Pesaresi et al. (2016b)

Project

The GHSL is a project supported by the European Commission through its Joint Research Centre (JRC) and the Directorate General for Regional and Urban Policy (DG REGIO) and for Internal Market, Industry, Entrepreneurship and SMEs (DG GROWTH). The project is part of the Human Planet Initiative of the Group on Earth Observations (GEO). It builds on the research activity carried out by the JRC since 2010.

The project aims to provide high-quality, detailed data that characterize human settlements at a global level over a period of time. The datasets obtained enable us to understand where people live and how human settlements have evolved over time. This provides a useful source of information in support of policy- and decision-making. In this regard, one of the purposes of this project is to contribute to the development of the indicators required to measure different policy objectives.

The project has delivered three main products, one of them referring to the urban footprint of human settlements (GHS-BUILT). This is the product described here, because of its assimilation to a Land Cover product. The other two products include a global grid of population density (GHS-POP) and a spatial layer of urban settlements classified according to their typology (GHS-SMOD). They have been produced for the same three time points and are based on the initial GHS-BUILT layers.

GHS-BUILT was initially produced for the years 1975, 1990, 2000 and 2014, providing a consistent time series of maps. They are available at three spatial resolutions, the finest one (30 m) providing information on the extent of the built-up areas. The aggregated maps (250 m, 1 km) give information on the percent of built-up areas per pixel.

New editions of the GHS-BUILT product have recently been released for the years 2016 and 2018. However, they are based on different imagery (Sentinel-1 and Sentinel-2) and were obtained using different methods. They are therefore not comparable to previous maps.

Production method

The GHS-BUILT maps for the period 1975–2014 were produced by classifying the historical archive of Landsat imagery through a Symbolic Machine Learning (SML) classifier. This is a supervised classifier that builds on a set of learning data. It includes previous information from older versions of the same product and other auxiliary datasets like the GLC30 or a global surface water product.

The classifier helped extract the following earth features from the imagery: clouds, water and built-up. After classifying the imagery mosaics for each of the periods under consideration (1975, 1990, 2000 and 2014), the classifications were then merged, thus ensuring the consistency over time of the historical series of maps.

The 2016 GHS-BUILT was also obtained using the SML classifier. However, the classification was carried out over Sentinel-1 backscatter imagery, so adapting the classifier to the potential and characteristics of this source of imagery. Certain differences can also be identified with regard to the learning data used in the image classification.

The 2018 GHS-BUILT was obtained by classifying a global cloud-free composite of Sentinel-2 imagery through a deep-learning-based framework, which is called the GHS-S2Net approach. A specific model was trained for each UTM grid zone of the global map, which allowed to account for local variability and computational model requirements. The model builds on a convolution neural networks architecture, which calculates the probability of built-up areas per pixel. Each model was trained with data from previous GHS-BUILT datasets, the European Settlement Map (see Sect. 6 in chapter “Supra-National Thematic Land Use Cover Datasets”), Facebook high-resolution settlement data and Microsoft building footprint data.

Product description

GHS-BUILT for the period 1975–2014 can be downloaded in small tiles or as a single global file. It is also provided at three different spatial resolutions (30 m, 250 m and 1 km) and in two different projections (Mollweide and Mercator).

The map at 30 m can only be downloaded as a multi-temporal product, providing information about the urban footprint for the whole period covered by the product (1975–2014). Maps at 250 m and 1 km can also be downloaded for specific years, without reference to built-up areas for other time points.

The dataset for 2016 obtained from Sentinel-1 imagery can only be downloaded for the whole world as a single zipped file. The dataset for 2018 is distributed in tiles corresponding with UTM grid zones. A vector layer representing the UTM grid zones in which the product is split can be downloaded as an auxiliary file, together with the product’s metadata.

Downloads

GHS—Built-up 2018 (10 m)

– Raster file with LUC information

GHS—Built-up 2016

– Raster files with LUC information for each of the tiles in which the product is divided (.tiff) (13_OTSU folder)

– Global mosaic of the product (.vrt) (V1-0)

– Vector file representing the tiles in which the product is distributed (.shp) (V1-0)

– PDF with the description of the product

GHS—Built-up 2014 (250 m)

– Raster file with LUC information

– PDF with the description of the product

GHS—Built-up multi-temporal (30 m)

– Raster file with LUC information

– PDF with the description of the product

Legend and codification

GHS—Built-up 2018 (10 m)

Code

Label

0–100

Probability of being built-up area (1–100)

255

No data

GHS—Built-up 2016 (20 m)

Code

Label

0

No built-up/no data

1

Built-up area

GHS—Built-up 2014 (250 m)

Code

Label

0–100

Built-up area density (1–100)

−200

No data

GHS—Built-up multi-temporal (30 m)

Code

Label

Code

Label

0

No data

4

Built-up from 1990 to 2000 epochs

1

Water surface

5

Built-up from 1975 to 1990 epochs

2

Land not built-up in any epoch

6

Built-up to 1975 epoch

3

Built-up from 2000 to 2014 epochs

  

Practical considerations

The maps for 2016 and 2018 are a test version of the product obtained with Sentinel-1 and Sentinel-2 imagery. They should not be therefore used together with the other GHS-BUILT maps, as if they were part of the same series of maps.

Users interested in the method used to produce this dataset can find the general workflow for built-up areas extraction in the MASADA (Massive Spatial Automatic Data Analytics) tool.Footnote 1

3 GAIA—Global Artificial Impervious Areas, GUB—Global Urban Boundaries

Product

LULC thematic

Dates

1985–2018 (GAIA)

1990, 1995, 2000, 2005, 2010, 2015, 2018 (GUB)

Formats

Raster (GAIA), Vector (GUB)

Pixel size

30 m

Theme

Extent of artificial areas

Urban boundaries

Extent

Global

Updating

Not planned

Change detection

Yes

Overall accuracy

Expected to be > 89% (GAIA)

Website of reference

Website Language English

http://data.ess.tsinghua.edu.cn/

Download site

http://data.ess.tsinghua.edu.cn/gaia.html (GAIA)

http://data.ess.tsinghua.edu.cn/gub.html (GUB)

Availability

Format(s)

Open Access

.tiff, .shp

Technical documentation

Gong et al. (2020), Li et al. (2020)

Other references of interest

Gong et al. (2019), Li et al. (2015), Li and Gong (2016)

Project

This project was led by researchers from Tsinghua University with the collaboration of colleagues from other Chinese and American universities, together with Google and the US Geological Survey. They have produced two different datasets: Global Artificial Impervious Areas (GAIA) and Global Urban Boundaries (GUB). The second was obtained from the first and both were produced to better understand global urbanization and other human socioeconomic activities and their impacts on the environment.

GAIA maps artificial surfaces across the world, whereas GUB maps urban areas. Unlike GAIA, GUB does not include small urban patches. In addition, in the GUB dataset non-artificial areas within cities, such as green areas or water bodies, are considered urban.

The project took advantage of the full Landsat data archive (1985–2018), providing a temporally consistent series of maps in which the only change possible was from non-artificial to artificial surfaces. The project is part of the global LUC mapping efforts carried out by Tsinghua University, such as FROM-GLC or GLC250, which are described in previous chapters of this book.

Production method

GAIA was first produced via the classification of the Landsat imagery archive (1985–2018). The dataset obtained in this way was then used to produce GUB. Google Earth Engine (GEE) was used to create both datasets.

Two different classification methodologies were followed to obtain GAIA: one for non-arid regions and the other for arid ones. This is due to the spectral confusion between impervious areas and bare lands. For classification purposes, the world was split into 583 tiles, of which 155 referred to arid environments.

The classification of non-arid areas was based on previous experiences of the production team in mapping artificial areas at local and national scales. Annual artificial areas were first obtained through an “ExclusionInclusion” algorithm, based on training data from earlier Landsat datasets and Google Earth imagery, and NVDI, MNDWI and SWIR data from Landsat imagery. The time series of maps obtained in this way was then further refined through a “temporal consistency check” approach.

For arid areas, a primary urban mask was first obtained for the year 2018 based on radar data from Sentinel-1 and VIIRS NTL. The classification of Sentinel-1 data was based on backscatter coefficients and NTL data was classified according to the quantile-based method. In both cases, different parameters were used for each arid biome. Once the two urban masks for 2018 had been obtained, they were mixed. Then, the time series of maps was created using the same “ExclusionInclusion” algorithm and “temporal consistency check” approach applied to the non-arid regions.

The GUB dataset was later obtained on the basis of a combination of two inputs: a kernel density map at a spatial resolution of 1 km obtained from GAIA based on a kernel density estimation (KDE) approach; and an initial urban boundary obtained from a Cellular Automata-based (CA) modelling exercise at 30 m. The results were improved through a morphological approach with dilation and erosion processing. This last step improved the mapped urban boundaries around fringe urban areas. Small holes inside urban areas were removed in a post-processing stage.

Product description

GAIA is distributed in 3.5º × 3.5° tiles, named according to the latitude and longitude of their upper-left coordinates. Users can download a vector file (.shp) drawing all the tiles and providing their names (field “FName_ID”).Footnote 2 GUB is distributed as a single global file for each of the 7 years available.

Downloads

GAIA

– A raster file with the extent of artificial areas (.tiff)

GUB

– A vector file with urban boundaries (.shp)

Legend and codification

GAIA

Code

Labela

Code

Label

Code

Label

Code

Label

Code

Label

1

2018

8

2011

15

2004

22

1997

29

1990

2

2017

9

2010

16

2003

23

1996

30

1989

3

2016

10

2009

17

2002

24

1995

31

1988

4

2015

11

2008

18

2001

25

1994

32

1987

5

2014

12

2007

19

2000

26

1993

33

1986

6

2013

13

2006

20

1999

27

1992

34

1985

7

2012

14

2005

21

1998

28

1991

  
  1. aThe label refers to the time when the pixel was sealed

Database

GUB

– Orig_FID: Unique identifier for each polygon

– UrbanArea: area of the delimited urban area

4 Global Urban Expansion 1992–2016

Product

LULC thematic

Dates

1992, 1996, 2000, 2006, 2010, 2016

Formats

Raster

Pixel size

1 km

Theme

Extent of Urban areas

Extent

Global

Updating

Not expected

Change detection

Yes

Overall accuracy

Expected to be > 90%

Website of reference

Website Language English

https://doi.pangaea.de/10.1594/PANGAEA.892684

Download site

https://doi.pangaea.de/10.1594/PANGAEA.892684

Availability

Format(s)

Open Access

.tiff

Technical documentation

He et al. (2019)

Other references of interest

–

Project

The dataset on Global Urban Expansion is the result of the work carried out by a group of researchers from the Beijing Normal University, the China University of Geosciences and Murray State University in the USA. Their aim was to create a new dataset on urban expansion using fully convolutional network (FCN)-based methods, which would be able to overcome some of the limitations of previous datasets on the same topic: outdated datasets, low spatial resolutions and low levels of accuracy.

The dataset provides useful information for studies addressing global urbanization and its impacts on the environment. It considers as urban all those built-up areas where human-constructed or artificial elements cover more than half of the area or pixel.

Production method

A specific fully convolutional network (FCN) was developed to map the urban areas in the Global Urban Expansion dataset. FCN are deep learning structures based on convolutional neural networks (CNN) that employ pixel-to-pixel image recognition.

The FCN was fed with different sources of input data: Nighttime Light (NTL) imagery from NOAA and NPP-VIIRS, as well as Normalized Difference Vegetation Index (NDVI) and Land Surface Temperature (LST) data from MODIS. Other auxiliary data sources were also employed to obtain the Global Urban Expansion dataset: urban population statistics, Landsat imagery and the GHS and LC-CCI LUC datasets. LST data is only available for the period 2000–2016 and was not used to map the urban areas in 1992 and 1996.

The FCN was calibrated with data from MODIS Land Cover, differentiating urban from non-urban areas. The calibration provided the weights of the FCN, which were then used to obtain the final Global Urban Expansion dataset.

A post-classification stage using population density data was carried out to ensure the consistency over time of the maps obtained.

Product description

The dataset can be downloaded as a single compressed file (.zip), including the raster files showing the urban expansion for each available year. No auxiliary information is provided with the dataset.

Downloads

Global Urban Expansion

– Raster files with urban expansion data for each mapped year (.tiff)

Legend and codification

Code

Label

Code

Label

0

Non-urban area

1

Urban area

5 ISA—Global Inventory of the Spatial Distribution and Density of Constructed Impervious Surface Area

Product

LULC thematic

Dates

2000 / 01, 2010

Formats

Raster

Pixel size

1 km

Theme

Impervious area density (0–100%)

Extent

Global

Updating

Not expected

Change detection

Unknown

Overall accuracy

Not specified

Website of reference

Website Language English

https://www.ngdc.noaa.gov/eog/dmsp/download_global_isa.html

Download site

https://www.ngdc.noaa.gov/eog/dmsp/download_global_isa.html

Availability

Format(s)

Open Access

.tiff

Technical documentation

Elvidge et al. (2007)

Other references of interest

Elvidge et al. (2004)

Project

ISA is the result of a project partially funded by NASA’s Carbon Cycle research program and is made up of researchers from different American institutions and universities. It builds on a previous attempt to map Impervious Surface Area (ISA) for the USA led by the NOAA (National Oceanic and Atmospheric Administration).

ISA was initially produced for the reference year 2000/01. A new version of the dataset is available for 2010. The dataset is useful for understanding the global distribution of impervious areas and for studies analysing the impact of these covers and their associated uses on the environment.

In addition to the production of an ISA density grid, the project’s outputs also include spreadsheets with information about the quantity of ISA per person at a country level and the ISA density per watershed areas. These are classified according to the proportion of ISA in three groups: stressed (1–10% ISA), impacted (10–25%) and degraded (>25%).

Production method

The ISA density grid for the reference year 2000/01 was obtained through a model making use of night-time lights imagery (DMSP OLS) and a population count grid (LandScan). Night lights imagery were captured in 2000–01, whereas the population count grid dates from 2004. A linear regression was defined to estimate the ISA density based on those two inputs. Only cells with a population count of at least 3 were considered in the regression. The model was calibrated with the ISA dataset produced for the USA at 30 m.

There is no accompanying information about the production process of the 2010 map. Therefore, we cannot know if it followed the same method as the previous map or some changes were introduced in the production process.

Product description

The ISA dataset for the reference year 2010 is distributed as a single compressed file (.gz). For the reference year 2000/01, the dataset is distributed in two different projections (GCS, Mollweide) and formats (ENVI, GeoTiff).

Spreadsheets containing ISA information per country and watershed are also available on the project website. This data is distributed together with a text file offering a technical explanation of these results.

Downloads

ISA (GeoTiff 2000–2001)

– Raster file with ISA proportion (.tiff)

Legend and codification

Code

Label

0–100

Impervious area density

Practical considerations

Although there is an ISA map for 2010, no information is available about the way it was produced. If there were important differences between the production methods used in 2000/01 and 2010 editions of ISA, they could not be used for comparison purposes or land change studies.

ISA was obtained from a calibration based on data for the USA. This may make the final result less accurate for countries with different night lights conditions, such as African countries. It is therefore likely that this dataset underestimates ISA densities in many different parts of the world.

6 HBASE and GMIS (Global High Resolution Urban Data from Landsat)

Product

LULC thematic

Dates

2010

Formats

Raster

Pixel size

30 m, 250 m, 1 km

Theme

Extent of urban areas

Percentage of impervious areas

Extent

Global

Updating

Not expected

Change detection

No (only one date)

Overall accuracy

Not specified

Website of reference

Website Language English

https://sedac.ciesin.columbia.edu/data/collection/ulandsat

Download site

https://sedac.ciesin.columbia.edu/data/set/ulandsat-hbase-v1/data-download

https://sedac.ciesin.columbia.edu/data/set/ulandsat-gmis-v1/data-download

Availability

Format(s)

Open Access after registration

.tiff

Technical documentation

De Colstoun et al. (2017), Wang et al. (2017)

Other references of interest

–

Project

Researchers from NASA, in collaboration with the University of Maryland and other American institutions, created two datasets to globally map artificial areas across the world: Global Human Built-up and Settlement Extent (HBASE) and Global Man-made Impervious Surface (GMIS). These were created within the context of NASA’s Land Cover and Land Use Change (LCLUC) program.

Both datasets used Landsat imagery available through the Landsat Global Land Survey (GLS) archive to consistently map impervious surfaces across the globe at high spatial resolution for the reference year 2010. These datasets aimed to overcome the resolution-related limitations of previous datasets. They can be useful for anyone studying impervious surfaces, their impact on the environment or their relation with other land dynamics. Because of the detail they provide, they can be used for studies and applications at global, supra-national, national and local scales.

HBASE and GMIS are complementary datasets, jointly produced to address the spectral confusion arising from the fact that many impervious areas are sealed with soil, sand, rocks, etc. and can therefore be confused with bare land. The HBASE dataset provides a mask to remove such areas from the GMIS dataset.

Production method

HBASE and GMIS were produced separately, although the first was used as a mask in the production of the second. In both cases, the GLS 2010 Surface Reflectance Dataset from Landsat was the input imagery.

For the production of HBASE, the first stage was to segment the GLC imagery using a Recursive Hierarchical Image Segmentation (RHSeg) software package. This produced a series of objects, from which different textures and other variables were extracted. On the basis of these variables, a random forest (RF) classification was carried out to classify the segmented objects in HBASE/non-HBASE categories. Training data for the classification was obtained from Landsat and Google Earth imagery. In addition, OpenStreetMap was used as an auxiliary dataset in the post-classification process to improve the mapping of the roads, which had not been correctly classified in the previous stages.

GMIS was obtained in two steps, with classifications carried out at the scene level. First, an object-based classification of GLC imagery was performed using the HSeg (Hierarchical Image Segmentation) Learn software to classify all the areas as either impervious or non-impervious. Only pixels effectively classified as HBASE in the previous dataset were considered and pixels with a low-quality classification were discarded. Later, the percentage of impervious area per pixel was calculated for all pixels classified as impervious through a regression-tree algorithm (Cubist). The algorithm was run with reference data from the National Geospatial-Intelligence Agency (NGA) at a spatial resolution of 30 m.

Product description

An online viewer allows users to download HBASE and GMIS: (i) for a specific country, (ii) for the tiles into which the datasets are splitFootnote 3 or (iii) for user-defined areas of interest (by drawing a polygon or shape or uploading a shapefile file that defines the area). The files can be downloaded at the original resolution (30 m) and resampled at 250 m and 1 km. Users can also choose between two projections: geographic or UTM.

Other complementary products are also available for download: a layer of standard error for the production of the GMIS dataset and an HBASE probability layer.

Downloads

GMIS

– Raster files with information on the percentage of impervious surface area (.tiff)

– A text document with technical information about the product (.txt)

HBASE

– Raster files with information on the urban extent (.tiff)

– A text document with technical information about the product (.txt)

Legend and codification

Global Man-made Impervious Surface (GMIS)—Percentage

Code

Label

0–100

Percentage of impervious surface area (0–100%)

200

Non-HBASE

255

No data, clouds, shadows

Global Human Built-up and Settlement Extent (HBASE)

Code

Label

Code

Label

200

Non-HBASE

202

Road

201

HBASE

255

No data, clouds, shadows

Practical considerations

Users can explore the different datasets available online,Footnote 4 including the complementary layer about the standard error of the Impervious Surface Percentage raster and the HBASE probability layer. Full metadata for GMIS and HBASE is also available online.Footnote 5

GMIS and HBASE have some limitations associated with their production methodology. For example, they may present areas of missing information due to cloud cover or other factors. The technical documents for the product (cited below) provide a detailed description of all these limitations.

As part of the same project, Landsat imagery composites for 66 urban areas are also available for download.Footnote 6

7 GUF—Global Urban Footprint

Product

LULC thematic

Dates

2011

Formats

Raster

Pixel size

0.4 arc seconds (~12 m near the Equator)

2.8 arc seconds (~84 m near the Equator)

Theme

Extent of built-up areas

Extent

Global

Updating

Not expected

Change detection

No (only one date)

Overall accuracy

Not specified

Website of reference

Website Language English

https://www.dlr.de/eoc/en/desktopdefault.aspx/tabid-9628/16557_read-40454/

Download site

https://www.dlr.de/eoc/en/desktopdefault.aspx/tabid-11725/20508_read-47944/

Availability

Format(s)

Open Access on request after filling in a request form

.tiff

Technical documentation

Esch et al. (2010), (2012), (2013), (2017)

Other references of interest

Esch et al. (2011), (2014), (2018a), (2018b), (2020), Marconcini et al. (2014)

Project

Global Urban Footprint (GUF) is a dataset produced by the German Aerospace Center (DLR) from radar imagery at very high spatial resolution: 0.4 arc seconds, which is equivalent to about 12 m at the Equator. The dataset at the highest resolution is envisaged for scientific uses, whereas a coarser resolution of the dataset at 2.8 arc seconds (~84 m near the Equator) has also been produced for non-commercial use by the general public.

The dataset aims to facilitate the quantitative and qualitative characterization of urban surfaces (size, form, spatial distribution) at different scales, from local to continental and global. Because of its high resolution, it allows all artificial surfaces to be analysed, in both urban and rural landscapes. This information is useful for researchers investigating the different impacts of the urbanization process, be they environmental, economic, political, societal or cultural.

The dataset was produced to overcome some of the limitations associated with previous global datasets on impervious surfaces, usually produced from demographic data. In this regard, by the time it was produced, high spatial resolution datasets were only available for specific regions, such as North America and Europe.

The project is part of the Urban Thematic Exploitation Platform (U-TEP) of the European Space Agency (ESA), which explores new methods and techniques to understand urban patterns and dynamics across the world. U-TEP is one of the seven Thematic Exploitation Platforms developed by the ESA to help data user communities.

In the context of U-TEP, DLR has also developed the WSF dataset, which outperforms GUF and resolves some of the limitations associated with it. WSF, which is described later on in this chapter, is a natural progression from the work undertaken to produce GUF. The two datasets are closely linked.

Based on GUF, a new layer on global built-up density was produced at a spatial resolution of 30 m for the reference year 2012 (GUF-DenS 2012). It provides information about the percentage of sealed surface or greenness per cell. Other complementary products based on GUF have been also produced, although they have not been made available to the public, namely a layer characterizing settlement properties and patterns (GUF-NetS) and a layer defining the average building height (GUF-3D).

Production method

GUF was produced from radar imagery from the TerraSAR-X/TanDEM-X satellites at a spatial resolution of 3 m. The imagery was captured between 2011 and 2012, except for a few images from the years 2013 to 2014.

The first stage of the production process was to extract a texture feature (speckle divergence) from the input imagery. Then, based on those features, a binary settlement layer differentiating between built-up and non-built-up areas was generated through an automatic unsupervised classifier: Support Vector Data Description (SVDD) one-class classification. The classification was carried out in 5° × 5° tiles. Once all the tiles had been processed, the obtained layers were mosaicked.

In a post-classification stage, the dataset was assessed against reference data, which confirmed or excluded the presence of built-up surfaces: Open Street Map, GLC30, NLCD, Imperviousness HRL and SRTM DEM.

Seven different layers were finally obtained on the basis of different classification settings: from very conservative settings (version 1) to very relaxed settings (version 7). Version 1 followed very strict criteria for classifying areas as built-up, whereas Version 7 followed much more relaxed, more inclusive criteria.

Product description

Interested users should request the product for their area of interest from the map’s producers. Before accessing the dataset, they have to sign a license agreement. Depending on the use they intend to make of the dataset, they can access the fine resolution version of the dataset (0.4 arcsec), which is only available for scientific purposes, or the coarser version (2.8 arc seconds). In both cases, the download only includes the raster file with the LUC information.

Downloads

GUF

– Raster file with built-up areas for the requested area of interest (.tiff)

Legend and codification

Code

Label

Code

Label

0

Non-built-up areas

128

No data

255

Built-up areas

  

Practical considerations

The dataset can be consulted online at the two spatial resolutions available.Footnote 7 A short document summarizing the technical characteristics of the product and its methodology is also available online.Footnote 8

Many other interesting data sources for characterizing urban areas can be found at the U-TEP Visualisation and Analytics Toolbox.Footnote 9 Users can also visualize the GUF-DenS 2012, which is not available for download. This dataset is complementary to GUF and provides information on the percentage of sealed surface for all the areas classified as built-up in GUF.

8 WSF—World Settlement Footprint

Product

LULC thematic

Dates

1985–2015, 2014 / 15, 2019

Formats

Raster

Pixel size

10 m, 30 m, 100 m, 250 m, 500 m, 1 km, 10 km

Theme

Extent of settlement areas (10 m)

Percentage of settlement areas (100 m, 250 m, 500 m, 1 km, 10 km)

Extent

Global

Updating

Expected

Change detection

Not yet (will be available with updates)

Overall accuracy

Expected to be > 86%

Website of reference

Website Language English

https://www.esa.int/Applications/Observing_the_Earth/Mapping_our_global_human_footprint

Download site

https://springernature.figshare.com/collections/Outlining_where_humans_live_-_The_World_Settlement_Footprint_2015/4712852/1

Availability

Format(s)

Open Access

.tiff

Technical documentation

Marconcini et al. (2020)

Other references of interest

Esch et al. (2018a), Esch et al. (2020)

Project

The World Settlement Footprint (WSF) is a dataset produced by the German Aerospace Center (DLR) within the context of a project (SAR4URBAN) funded by the European Space Agency (ESA) in which Synthetic Aperture Radar (SAR) is used to monitor urbanization. The project aimed to develop a new method to automatically map built-up areas via the joint use of radar and optical data.

The dataset obtained is useful for the characterization and analysis of urban patterns across the world. It overcomes the limitations of previous high spatial resolution datasets mapping impervious surfaces by making use of both radar and optical imagery at the same time. This allows WSF to avoid the misclassifications that can result from using only one of the two types of sensors: optical imagery misclassifies sand and bare soil, whereas radar imagery misclassifies complex topography areas and forested regions.

WSF is produced by the same institution as the Global Urban Footprint (GUF) described earlier in this chapter. In spite of this, it overcomes some of the limitations associated with GUF, such as the misclassifications arising from the use of single-date scenes and the use of commercial imagery, which makes updating more difficult due to the associated costs. Like GUF, WSF was also developed within the framework of the Urban Thematic Exploitation Platform (U-TEP) of the ESA.

The dataset was originally produced at a spatial resolution of 10 m, although resampled versions at 100 m, 250 m, 500 m, 1 km and 10 km are also available for download. The resampled versions show the percent of settlement area in each pixel instead of a binary classification differentiating between settlement and non-settlement areas.

The DLR has recently worked with the Google Earth Engine Team on the update of the product, creating a WSF-Evolution dataset that maps the global evolution of built-up surfaces yearly from 1985 to 2015 with a spatial resolution of 30 m.

Production method

The WSF production methodology was first tested at a range of selected sites and, once validated, was applied to generate the global dataset. It used Sentinel-1 and Landsat 8 data for the reference years 2014 and 2015 as input.

From Sentinel-1 data, key temporal statistics were extracted from the original backscattering value. From Landsat 8 imagery, different spectral indices were extracted: vegetation index, built-up index etc. Based on the extracted information, a binary classification (settlement/non-settlement) was computed through an ensemble of Support Vector Machines (SVM) classifiers for each type of input data: radar and optical. The two results were then combined.

In a post-classification stage, the obtained result was assessed against reference information, following the post-editing object-based approach applied in the production of GUF. The auxiliary datasets were: Open Street Maps, GLC30, SRTM DEM, ASTER DEM, NLCD and the High-Resolution Layer on imperviousness.

Product description

WSF can be downloaded at multiple spatial resolutions. For the original resolution (10 m), the users will download a compressed file (.zip) that includes all the raster files into which the dataset is split (306.tiff files). The download also includes a virtual raster that merges all the tiles in a single mosaic. For all other available resolutions (100 m, 250 m, 500 m, 1 km and 10 km), users can only download a .tiff file with data on the settlement percentage per pixel. No auxiliary information is provided in either of the two cases.

Downloads

WSF 10 m

– Raster files with the settlement extent for the 306 tiles into which the product is divided (.tiff)

– Raster file with a mosaic of the WSF tiles (.vrt)

WSF 100 m, 250 m, 500 m, 1 km, 10 km

– Raster files with the settlement percentage (.tiff)

Legend and codification

WSF 10 m

Code

Label

Code

Label

0

Non-settlement

255

Settlement

WSF 100 m, 250 m, 500 m, 1 km, 10 km

Code

Label

Code

Label

0–100

Settlement percent (0–100%)

255

Settlement

Practical considerations

WSF is considered by the authors to be the most accurate dataset of its type. It is part of the U-TEP tool, which also distributes many other datasets for characterizing urban areas that may be of interest to users. Users can access an online visualization of the dataset on the U-TEP tool website.Footnote 10

For more detailed information about the characteristics of the dataset, we recommend interested users to read the scientific paper in which it was presented.

9 GISM—Global Impervious Surface Map

Product

LULC thematic

Dates

2015

Formats

Raster

Pixel size

30 m

Theme

Extent of impervious areas

Extent

Global

Updating

Not expected

Change detection

No (only one date)

Overall accuracy

Expected to be >95%

Website of reference

Website Language English

Not available

Download site

https://zenodo.org/record/3505079#.YFoEI2hKiUk

Availability

Format(s)

Open Access

.tiff

Technical documentation

Zhang et al. (2020)

Other references of interest

–

Project

A group of researchers from Chinese institutions (Chinese Academy of Sciences, University of Science and Technology) and the University of Wisconsin-Milwaukee produced a Global Impervious Surface Map, which aimed to overcome some of the limitations of previous datasets.

GISM is part of recent efforts to produce a detailed global mapping of artificial or impervious surfaces with a high level of accuracy to provide useful data that can help characterize artificial areas and their associated environmental and socioeconomic impacts. The dataset was produced with that aim, without any further updates being planned.

Production method

GISM was obtained by classifying Landsat and Sentinel-1 data in the Google Earth Engine (GEE) platform, using the MSMT_RF method. First, temporal–spectral–textural features were extracted from Landsat imagery. Then, temporal-SAR features were extracted from Sentinel-1 imagery. On the basis of all these features, a classification was carried out with a random forest classifier in 5° × 5° tiles. Training data for the classification were obtained from GLC30, VIIRS NTL and MODIS EVI imagery. SRTM DEM was used as an auxiliary dataset in the classification process.

Product description

GISM is distributed as a single compressed file (.zip) containing all the raster files into which the product is distributed: 954 5 × 5 degree tiles. No auxiliary information is provided.

Downloads

GISM

– Raster files mapping impervious areas for each of the tiles into which the dataset was divided (.tif)

Legend and codification

Code

Label

Code

Label

1

Non-impervious

2

Impervious

Practical considerations

The only other relevant information on the dataset can be found in the scientific paper in which it was presented. Users wishing to find out more about the characteristics of this product should consult this paper.