Urbanization as a global phenomenon is a multifaceted process, affecting the Sustainable Development Goals (SDGs) in and around urban areas. Here we do the first global attempt to characterize the complexity of urbanization from 1975 to 2015 in terms of population, built-up structure, and greenness, as well as monitoring urban sustainability indicators at the grid level covering all inhabited areas. We used Global Human Settlement Layer to assess built-up structure, population, and land-use efficiency (SDG 11.3.1), combined MODIS/Terra & GIMMS NDVI for long-term greenness, distributed statistical energy consumption by night lights and population for energy efficiency (SDG 7.3.1) and used near-surface PM2.5 dataset for air quality (SDG 11.6.2).

Our results emphasize that the multifaceted nature of urbanization as well as related sustainable challenges vary greatly across regions and times. (1) Increased population density and built-up patch density was dominant in Asia and Africa, while urbanization in Europe and North America took a rather steady pace, combined with widespread greening. (2) According to the urbanization types identified by a self-organizing map (SOM) algorithm, a large proportion of urban and suburban areas experienced two dynamic urbanization types - built-up extension/leapfrog and built-up infill with large population increase (Fig. 1). (3) During different historical periods (1975-1990, 1990-2000, and 2000-2015), annual increases in population and built-up density were slowing coinciding with an increasing greenness – signaling that urbanization processes are becoming less intense, more compact, and “greener” over the most recent period. (4) Land-energy-air SDGs have declined in over 30% of global inhabited grid cells from 2000 to 2015. (5) In land-energy-air sustainability trends, urban areas perform relatively better than rural areas in the Global South, while urban areas in the Global North tend to be less sustainable than their surrounding rural regions.

Our findings facilitate comprehensive understanding of global urbanization and relevant sustainability with many local variations and characteristics. Integrating Earth Observation data is crucial for tracking urbanization and sustainability, and can guide context-specific strategies towards a sustainable and livable future instead of a ‘one-size-fits-all’ policy for cities.


Fig. 1. Global map of urbanization types a, as well as spatial patterns in Eastern America b, Western Europe c, Central Europe d, Eastern China e, and South Africa f. This figure presents the spatial distribution of five urbanization types we identified based on changes in population density-PopD, built-up density-BuiD and structure (patch density-PatD and the largest patch index-LPI), and NDVI. The + and - signs represent whether the change is above or below global average change (+ is up to 0.25-0.5 s.d., ++ is 0.5–1 s.d., +++ is >1 s.d.). Dashed boxes within example maps (b-f) mark out the urban area of some large cities.

The Sustainable Development Goal (SDG) 11 of the United Nations (UN) aims at renewing and planning human settlements in a way that offers opportunities for all, including access to essential services, housing and energy, green public spaces, transportation while reducing the use of the resources and impact on the environment. In this context, accurate, reliable and frequent information is needed to comprehensively characterize human settlements. To this purpose, the increasing availability of Big Earth data (as from satellite observations) and related analytics tools has recently opened novel opportunities. However, in the last few years, this has led to the generation of several global layers, primarily focusing only on delineating the actual settlement extent, often with limited quality.

To overcome this limitation, the German Aerospace Center (DLR) in collaboration with the European Space Agency (ESA) and the Google Earth Engine team has been generating the World Settlement Footprint (WSF) suite, an unprecedented collection of open-and-free global datasets aimed at advancing the understanding of urbanization at the planetary scale. In this framework, the first layer to be released has been the WSF2015, a 10m resolution binary mask outlining the 2015 global settlement extent derived by jointly exploiting multitemporal optical Landsat-8 and radar Sentinel-1 (S1) imagery. The layer proved highly accurate and reliable, as quantitatively assessed by an extensive validation based on 900K ground-truth samples labelled by crow-sourcing photointerpretation of VHR satellite imagery. Starting from the WSF2015 two parallel activities have been then concurrently initiated.

On the one hand, for characterizing ongoing trends and foster the sustainable development of human settlements, a proper understanding of their past growth is essential. Accordingly, a novel iterative approach has been implemented that, starting backwards from 2015, effectively outlines on a yearly basis the settlement extent based on Landsat data alone (given the lack of systematically available archived high-resolution radar imagery). After an extensive test phase, this has been eventually employed in Google Earth Engine for generating the WSF-evolution, i.e. a dataset outlining the global settlement extent at 30m spatial resolution from 1985 to 2015. In particular, the WSF-evolution has proven to be the most accurate product of its type, enabling integrated analyses so far not yet possible. This has been assessed by means of an extensive campaign, similar to that carried out for the WSF2015, where overall ~1.2M samples have been labelled for the years 1990, 1995, 2000, 2005, 2010 and 2015.

On the other hand, the availability of Sentinel-2 (S2) imagery has enabled unprecedented possibilities for monitoring urbanization thanks to its higher number of spectral bands and higher spatial resolution with respect to Landsat data, along with the 5-day revisit time since March 2018. A new approach has been then implemented, which jointly exploits multitemporal S1 and S2 data as input, together with training information extracted by means of the WSF2015. The method has been tested on a number of study sites throughout the different climate regions and, ultimately employed to generate the WSF2019, a novel 10m resolution mask outlining the global settlement extent for 2019 (i.e., the first year for which atmospherically corrected S2 data have been entirely available). Also in this case, the whole processing has been performed in the Google Earth Engine environment, while two independent crowd-sourcing-based validation exercises have been carried out with the support of Google and Mapswipe, respectively, where overall ~1M reference labels have been collected.

In the next few months, the WSF suite will be further enriched by 3 new global layers, namely:
i) The WSF3D, derived by means of the 12m resolution TanDEM-X digital elevation model and the WSF2019 layer, which – first of its kind - estimates globally the average height and volume of built-up areas at 90m resolution;
ii) The WSF2019-imperviousness, estimating for each settlement pixel of the WSF2019 the corresponding portion on the ground covered by paved surfaces, which is generally considered as a good proxy for the building density;
iii) The WSF2019-population, estimating at 10m resolution the number of inhabitants per pixel, which is derived by redistributing census figures available at the finest possible administrative level by exploiting the WSF2019-imperviousness, along with ancillary land-use data from OpenStreetMap.

Key stakeholders from different communities (World Bank, Asian Development Bank, UN Habitat, International Committee of the Red Cross, World Resource Institute, just to cite some) have already been extensively working with the different WSF products. In particular, as champion users they have been granted access for specific study regions even before the public release of the datasets. Here, different case studies clearly demonstrated their capabilities to effectively monitor urbanization and support evidence-based urban development policies, as well as the achievement of the New Urban Agenda and UN SDG 11.

The National Urban Planning process is comprised of several stages from Feasibility, to Diagnostics, Formulation, Implementation, Monitoring and Evaluation. In order to implement these steps evidence-based approaches for decision making are needed. The NUPs should address three main themes: urban legislation, urban economy and urban planning and design. Of the three themes, the urban planning and design is a key requirement at local and national level; without such plans urban development especially in developing countries is often uncontrolled resulting in urban sprawl, lack of adequate services and infrastructure and the development of slums which leads to a heavy financial burden on the governments who have to address these challenges. Geo-spatial data play a major role in evidence based urban planning. Unfortunately, many developing countries face challenges with the operational utility of Earth Observation (EO) and GIS for extraction of spatial data for urban planning. In this context the Multi-Lateral Development Banks (MDBs) who are supporting improved urban planning have an important role in promoting the utilisation of these technologies for improved urban planning in developing countries. The ESA EO for Sustainable Development (EO4SD) Urban project from 2016-2020, undertaken in collaboration with the International Financing Institutes (IFIs) supported urban planning and development programmes with a suite of geo-spatial data. The project provided 32 Cities with more than 500 products for which the overall accuracies ranged from 85-95%. Examples of analytical work included assessment of urban growth over time, LU/LC change over time, support to SDG Goal 11 and related Indicators and assessment of flood prone/flood risk areas.

A high demand product was the 3D building height data which can be used with cadastral data for modelling different property values and related tax rates. The production of building height datasets is a complex exercise which relies on the use of Very High Resolution EO data from different sensors for the production of digital surface models (DSMs) and digital terrain models (DTMs). The methodology uses both automatic and semi-automatic processes to generate high accuracy DSMs which are the basis for creating also the DTMs needed for the building height product. Innovations with the methodologies have led to a highly automated multi-source production line used to generate 0.3 m to 10 m resolution digital surface models and derived products for over 13 million km² and for applications at different scales, including 0.3 to 0.5 m elevation products for more than 280 cities in Latin America, Africa, Asia and Europe. Based on the multispectral images and on the elevation products, a deep learning-based processing framework has been developed to derive building footprints. In addition to the previously described datasets, 3D surface models with textured facades can be generated based on multi-stereo EO images, to further enhance the usability with more realistic 3D visualisation of buildings, which leads to increased application possibilities. This gain of efficiency and utility has allowed EO service industry to address better IFI tenders, which require these products for many cities over many countries in a very short time.

The EO4SD Urban project thus provided valuable evidence for the potential to harness and mainstream EO data and applications for Urban Development programmes; the successful implementation of the cases are envisaged to be carried forward in 2022 via the ESA GDA Urban Programme and further collaboration with the IFIs.

In a context where the dynamic trend of urbanization is becoming more and more rapid, urban planning is a challenge as it involves limiting the amount of land being taken up while providing new living spaces. It is therefore important for local authorities to undertake effective measures in urban planning and development, in order to control the urban expansion, enhance the resilience of cities and preserve green spaces. In former industrial regions, such as Wallonia, there is a large number of brownfields, called "Redevelopment Sites" (RDS), which offer an opportunity for sustainable urban planning due to their redevelopment potential. These are mainly urban sites previously used for industrial activities and now abandoned. Currently, in Wallonia slightly more than 2200 RDS are listed in an inventory managed by the Walloon authorities, which required considerable time and resources to maintain up-to-date.

Within this perspective, the Sentinel satellites of the European Copernicus program are a real opportunity. Thanks to their high temporal and spatial resolution, their open access and the possibility of complementary use of different sensors, combined with the RDS inventory, they allow the implementation of an operational tool for RDS monitoring, in near-real time and over the long term.

What we propose in this study is an operational and automatic solution for the processing of Sentinel-1 and Sentinel-2 data where a combination of change detection and change classification methodologies is used to generate a final report that is directly usable by public authorities. The complete processing chain is implemented in Terrascope, the Belgian Copernicus Collaborative Ground Segment, which offers, via virtual machines, pre-processed Sentinel data and computational capacity. This enabled the automation of the process while processing and analyzing large volumes of data and images.

As far as the methodology is concerned, first, a suitable set of features (backscatter from the Sentinel-1 VH band and a selection of Sentinel-2 indices) is extracted from the data and used to create average temporal profiles for each polygon contained in the RDS vector file. The latter allows object based methodologies, one object being one RDS. Next, the PELT (Pruned Exact Linear Time) change detection method is applied to the Sentinel-1 VH and Sentinel2 NDWI2 features to determine if a change has occurred and estimate the date of the change. Finally, a classification of the changes, which is exploits the Sentinel-1 VH and Sentinel2 (NDVI, BI, BI2, SBI & BAI) features is performed to provide information on the type of change (vegetation, building and soil), the direction of the change (increase/decrease), if any, and the amplitude. This last part is composed of two distinct processes: (1) the "summer classification", which is best suited for detecting and classifying gradual changes, and (2) the "change point classification", which provides information on the type of change each time a change point is detected and an estimated date is indicated. In general, the multi-temporal approach allows to: (1) select and pre-process Sentinel data while removing outliers, (2) estimate the dates of changes, (3) characterize the changes according to the estimated dates, and (4) automatically provide results at regular intervals and/or on demand.

The results were validated based on 2 sets of ground truth created by visual analysis. The first one was obtained from orthophotos (25cm resolution) taken once a year between 2016 and 2018. The second one was created from Pleiades images (4-band pan-sharpened products at 0.5 m resolution) taken once a month over two years (2019 and 2020). The validation highlighted the relevance of the processing chain for the change identification, with a satisfactory accuracy. This applies both for the change detection (overall accuracy 62% to 85%), which also provides an estimate of the change date, and for the change classification (overall accuracy 69% to 90%), which indicates the type of change. For the change classification, the two processes "summer classification" (overall accuracy from 79% to 90%) and "change point classification" (overall accuracy from 69% to 85%) performed sufficiently well. The results showed the suitability of the combination of the two different methods, which allows us, on the one hand, to classify the type of change when a change point is detected via the PELT methodology and, on the other hand, to identify the gradual changes. Therefore, the processing chain presented in this project enables the creation, in near-real time and automatically at regular intervals, of a priority order list that highlights the RDS with the most changes and thus guides the work of the field operators, allowing them to focus on the sites that need their attention the most. This helps the Walloon authorities to manage the RDS inventory in a more efficient and reactive way, which is an important contribution to the improvement of urban planning and development measures.

As a conclusion, even if earth observation data present some limitations and challenges, such as the spatial resolution of Sentinel images, the project results show that the use of these types of data represents an opportunity to improve urban development policies. In particular, it shows an application that can be directly used by the authorities to monitor the evolution of sites that present a high potential for redevelopment. As a further use, the proposed processing chain could be used to monitor other types of sites in the field of urban planning, but also in agriculture, forestry or in disaster response.

Unprecedented rates of urbanization are taking place around small- and medium-sized cities in biologically critical zones. Characterizing urban land cover with high spatiotemporal resolution will improve our understanding of how the Earth system functions today, how it supports life, and how conditions might change to alter the climate and human well-being. However, small-scale urban land changes are often invisible in global urban land cover datasets. Here, we present the usefulness of using convolutional neural networks compared to conventional machine learning methods (random forest regression and maximum likelihood estimation) to characterize urban land cover and change at the sub-pixel level. Also, we compare the two popular input datasets from Landsat time-series data: 1) multispectral imagery and 2) coefficients of Continuous Change Detection and Classification (CCDC) algorithm for training the models. In total, six approaches (three models with two input data) were compared and applied to challenging landscapes on the Himalayas mountain ranges: (1) maximum likelihood estimate + multispectral imagery (MS), (2) maximum likelihood estimate + CCDC, (3) random forest + MS, (4) random forest + CCDC, (5) convolutional neural network (CNN) + MS, and (6) CNN + CCDC. Nepal’s dataset was used for training and India and Bhutan’s were used for an independent test. We also evaluated temporal accuracy using a held-out region with annual time series data over 2010-2020. The results show that CNN + MS has outperformed the other methods in the following aspect: First, the use of CNN has substantially improved the accuracy for small urban settlements. Compared to RF’s average magnitude of the errors in small settlements at 5% (urban fraction), CNN + MS’s average magnitude of the errors is < 1%. Second, CNN + MS shows the best ability in characterizing urban land changes at a range of temporal resolutions from one to ten years. CNN + MS’s errors in predicting change magnitude range from 1% (one-year interval) to 8% (ten-year interval), while RF’s errors in change magnitude range from 1% (1y interval) to 17% (10y interval). CNN + MS explained 50% of the variation in change magnitude at the ten-year interval, while RF + CCDC (the second-best method) explained only 18% of the variation. Third, CNN + MS reduced false positives from spectrally similar landscapes (e.g., tilled agricultural fields, landslides at hilly areas, river silt, and snow cover, see Fig. 1). Our results show quality multitemporal urban patterns compared to other global products. Overall, using multispectral imagery as input for CNN models results in significantly higher accuracy than using CCDC as input, indicating noisy raw satellite data is better for deep learning than regressed or indexed data. Our analysis reveals that a simple method can be valuable in monitoring small-scale urbanization at a subpixel level with 30m resolution, which may help to inform resilience-oriented policies in biologically critical and hazard-prone areas.

The world is rapidly changing - as humans expand not only established cities, villages and towns, but also into the most remote areas of the world, there is a great need for broad area monitoring to find the areas growing fastest and having the greatest impact on its surroundings. Planet’s near-daily global imagery at ~4 meter resolution makes it the perfect dataset to track global settlements as they change. This rapid revisit enables Planet to create mostly cloud-free monthly mosaics that serve as a clean canvas for a supervised semantic segmentation model to run upon. In this way we can make the most efficient use of human eyes and let the machines analyze every pixel we publish. Our globally trained model classifies each pixel as building, road or other as each monthly mosaic is published and the resulting classified layers are made available for our users to stream and download for further analysis. The presence of roads and especially buildings derived from Earth observation data serves as a proxy for population which is key information for monitoring Sustainable Development Goals and for disaster risk management. Tracking settlement growth regularly enables NGO’s to prioritize where they send aid to improve people’s life or provide help in case of disasters. Having access to such information at high spatial resolution allows population modellers to estimate with higher precision how many people live where.
Once these areas of rapid expansion have been found through the Planetscope broad area monitoring mission, Planet’s Skysat constellation can focus its attention to capture sub-meter imagery of the area to get an even greater understanding of what development is taking place. With the 50cm Skysat data, true building footprints and road networks can be seen or even extracted through downstream analysis, leading to even greater precision for population modellers and disaster response teams. By combining broad area monitoring to find settlement expansion with targeted submeter captures to get a clearer view of what development is taking place, Planet can provide a window into how our world is changing.
We’ve begun to unlock the power of Planet’s daily global dataset and the applications that can be built on top of it. In this presentation, we will review the global road and building layer and explore the growth of some key settlements over time to showcase just how useful this data can be, especially when combined with 50cm data provided through Planet’s Skysat constellation. We will show how development within cities, rural areas and even the most remote locations are changing and how Planet is uniquely positioned to highlight their development.