SESSION

Continuous monitoring of the Frascati urban area is required to keep track of the loss of natural zones due to urban development and support urban planning activities. The present analysis (in progress) is conducted to examine the effects of the urbanization process occurred over the Frascati area of interest during the last years. This area has been particularly prone to the urban sprawl phenomenon, not only because of the concentration of population but also due to the interplay that exists between people, infrastructures and natural or man-made risks. Overlays of Sentinel OPTICAL/RADAR satellite images, collected over a three year period, were validated against very high resolution imagery; datsets were integrated into Geographic Information System (GIS), allowing an easier discrimination of the urban landscapes, through a particular procedure - the so called UAPI index analysis - applied to urban land/agricultural transformations.

Modelling of urban climate using coupled Global Climate Models (GCM) with urban canopy model had advanced significantly in last decade. For many operational weather forecast studies, urban area is treated as single entity.  However, improved urban parameterization to capture urban heterogeneity is found to be essential for representation of urban fluxes in Numerical Weather Prediction (NWP) models. In this study, the WRF model was integrated with four domains with a horizontal resolution of 9km, 3km and 1 km. Two study areas differing in size, shape, complexity and development characteristics, NCT Delhi and Chandigarh region were chosen for the study. Four simulations with different combinations 1. Default Land Surface Land Cover (LULC) and Land Surface Parameters (LSP), 2.  Default LSP with high-resolution urban LULC, 3. Default LULC with updated LSP and 4. High-resolution urban LULC with updated LSP) were carried out to assess the impact of improved urban parameterization on climatic variables by their integration in WRF-Urban model. The evaluation of simulated results with observed data showed significant improvement in RMSE of surface pressure, 2m temperature and wind speed obtained from high-resolution urban LULC and updated LSP combination. The RMSE of surface pressure, 2m temperature and wind speed obtained from the high-resolution urban LULC and updated LSP combination for the winter season winter was observed to be 0.94, 2.54, and 1.62 respectively for NCT Delhi and 1.96, 1.70, 4.93 respectively for Chandigarh area. as compared to the default’s simulation’s RMSE values of 1.80, 3.77, 4.66 for NCT Delhi and 3.53, 5.15, 8.70 respectively for Chandigarh region. Comparison of the model predicted surface temperature with MODIS 1km*1km LST revealed that simulations with urban LULC and updated  LSP produced better results for both the study area with a difference between the two ranging from -1 to 2 ^oC.

Urban geological hazards involving ground instability are costly, can be dangerous, and affect nearly everybody, yet information relating to type, extent or distribution within Europe’s towns is largely unknown. A reason for this is the impracticality of measuring ground instability associated with the many insidious processes that are often hidden beneath buildings and imperceptible to normal means of detection. Satellite radar interferometry, InSAR for short, offers a remote sensing technique able to map mm-scale ground deformation over wide areas from one image to the next given an archive of suitable multi-temporal radar data. The EC FP7 Space Project PanGeo (www.pangeoproject.eu) conducted between 2011 and 2014, incorporated the InSAR technique into work that mapped the area of unstable ground in 52 of Europe’s largest towns, representing ~15% of the total EU population. In partnership with all Europe’s national geological surveys, the project developed a standardised geohazard-mapping methodology and recorded 1286 instances of 19 types of geohazard covering nearly 17,000km².  Presented here is an analysis of the results of the downloading and further processing of the PanGeo-project output data to provide a first understanding of the distribution of geohazards by type across European towns, along with their frequency and probability of occurrence.  Intersection with Eurostat’s GeoStat data also provides a systematic estimate of population exposures.  Satellite radar interferometry is clearly shown as a valuable and unique tool in the detection and mapping of urban geohazard phenomena.

Managing urban development and growth patterns is a challenge for local administrations in charge of taking into account community’s needs, environmental protection, economic development and social well-being. Increased urbanisation emphasizes challenges for planners and decision makers because urban growth more often causes reduction of natural and agricultural areas, soil sealing, traffic congestions and mobility issues, increasing carbon emissions, worse quality of life. Actually, in 2015 cities and human settlements have been included in the UN Sustainable Development Goals (SDGs). Among the SDGs, SDG 11 (Make cities and human settlements inclusive, safe, resilient and sustainable) stands out as a goal that has placed explicit focus on the measurement of indicators requiring geospatial data for monitoring. Urban planning is a long journey from identification of strategic guidelines and requirements up to completion of plans, and it requires continual monitoring to check actual results and their impacts with respect to the original objectives. Thus, Earth Observation by Remote Sensing has a role to play.

Over the past few years, the ESA’s Copernicus Program and commercial data providers gave birth to a new era of Big Data in the field of Earth Observation (EO), easing the detection of changes in urban areas. Planetek Italia has great experience in urban monitoring and succeeded in creating Rheticus® Urban Dynamics, a vertical service devoted to provide continual monitoring of urban areas by integrating contents generated by its Rheticus® platform and based on VHR data. Especially, Rheticus® Urban Dynamics covers several continuous urban monitoring fields ranging from spatial planning to land use and land cover change monitoring, from soil consumption and imperviousness assessment, to urban heat island detection, urban sprawl, or infrastructural instabilities. Rheticus® Urban Dynamics simplifies the stream of data thanks to a cloud architecture and automatic processing chains, and it overcomes the difficulties and costs of field surveying. It provides a synoptic point of view over wide areas, through remotely measurements of indicators useful to assess different aspects of urban dynamics and to plan and achieve sustainable developments of urban areas. The service could also be available on EO Data and Information Access Services (DIAS) platforms, which are expected to be operational by the end of June 2018.

Among the innovations introduced by Rheticus® Urban Dynamics, shifting from a model of monitoring services on request to long-time information service subscriptions is the real disruptive innovation in the field of modelling of urban development and growth patterns. It helps national-to-local authorities, policy and decision makers to implement urban sustainable development and good practices in accordance with the UN SDGs, to support the quality of planning processes and their proper execution, and to anticipate environmental impacts of urban sprawl.

In the framework of technical-scientific activities supporting the Italian Civil Protection (CP) related to the earthquake occurred on the August 24, 2016 in Central Italy, we proposed a new EO-based application in order to characterize the urban rubble heaps deriving from buildings affected by partial or total collapse. The test area of Amatrice town (one of the most affected urban areas) was selected. The level of damage distribution for the buildings was initially provided by the Copernicus EMS operational service. The main goal of the present study was to estimate volumes and superficial percentage of the various types of materials forming the rubble heaps by using multiplatform remote sensing data.

Firstly, an expeditious procedure to assess rubble heaps volume was developed and applied, based on LIDAR data and aerial images (RGB, 0.15 m resolution), using aero-photogrammetric techniques, photo-interpretation and GIS methods (Cappucci et al., 2017).

Subsequently, a frame (~10X10 Km) acquired by the World View 3 (WV3) sensor, (1.35 m resolution for the 8 (VIS-NIR) multispectral channels; 0.33 m for panchromatic channel), was exploited for the surface characterization of rubble materials. The spectral signatures of various rubble materials were acquired on site through ASD FieldSpec Pro hyperspectral radiometer. Then, data were pre-processed and arranged in a spectral library of 23 different spectral signatures.

Given the spatial heterogeneity of the rubble, typically smaller than the WV3 pixel, the SMA (Spectral Mixture Analysis) approach was adopted. The spectral signatures collected from the field were resampled to make them compatible with the 8 channels of WV3 sensor. From the WV3 multispectral imagery, appropriately pre-processed to remove geometric distortions and atmospheric effects, a small number (10) of pure materials signatures (endmembers) was initially derived (SMACC algorithm) using sample heaps as reference for the main material classes, jointly with the support of the available RGB orthophotos (0.15 m resolution). SMA was then carried out on the heaps within the entire test area using the above mentioned endmembers.

Subsequently, percentages obtained in this way were recombined using the classification of the 10 materials extracted into the 23 field materials of the library on the basis of their resampled spectral signatures. To this end a machine learning SVM (Support Vector Machine) algorithm allowed to better discriminate the most representative categories of rubble heaps.

The final characterization of rubble within the test area was finally obtained through GIS processing (zonal analysis and normalization) executed on each polygon representative of the heaps layer previously identified. Thus, both the volume for each heap (range between 500 to 1000 m³) and its relative percentages of rubble materials considered were assessed. The most abundant rubble heap typology (macro class) identified is represented by concrete/debris (range: 40-50%), followed by brick/tiles (range: 25-30%) and local stones (range: 7-17%), with other materials (metal, plastic, roof-tiles, etc.) completing the list (range: 3-25 %).

Future activities will include a greater exploitation of high-resolution and hyperspectral satellite data, for assessing both volumes and typologies of earthquake rubble heaps, also investigating the possibility to detect hazardous materials (e.g. asbestos).  

The metropolitan area of Bolivian capital city La Paz, together with its satellite cities El Alto and Viacha faced during the last 4 decades a very dynamic urban sprawl. In less than 30 years, the cities almost doubled their size, with growth rates up to 2.7%/year. Concomitantly, the population count matched the same trend with growth rates up to +20.000 inhabitants per year. This dynamics put tremendous pressure on the local environment and led to an increasing necessity to model the conurbation of La Paz – El Alto – Viacha.

Earth Observation data (EO) represent key information when assessing city expansion over longer time spans. Nevertheless, the extraction of urban masks from these data induces multiple challenges related to spatial resolution, imagery co-registration, slope and cloud coverage.

This research aimed to derive urban masks at regular time intervals (at least every 5 years) starting from 1960 in order to be able identify the breakpoints and the directions of the local urban development. The EO database included a significant number of satellite images, in accordance with the time interval under analysis:

Corona, Landsat MSS and TM: 1960 – 1985;

SPOT 5, Landsat TM and ETM+: mid-90s;

Landsat OLI, Sentinel-2, Rapid-Eye, SPOT 6/7, Planet: the modern period.

The urban masks were extracted using a mixture of supervised classifications and segmentation techniques. The results enabled the calculation of the total built-up areas within the area of interest, as well as several additional indicators:

the yearly growth rate of the built-up areas,

the number of new buildings,

the average footprint of the new households,

the length and area of the newly built roads and their yearly growth rate,

the future directions of expansion of the urban areas and transportation network.

The superior resolution of VHRO revealed regional differences and different patterns of the urban expansion process that were investigated through a series of case-studies:

the development of new residential areas towards the plateaus or the flat areas around (e.g.: the northern part of the Santisima Trinidad Urbanization),

the city expansion towards the valleys in the eastern part of the city (e.g.: Lomas de Irpavi – a highly urbanized area in the north-eastern part of La Paz, Lomas de Achumani Urbanization – in the south-eastern part of the city of La Paz)

The results were also compared against population and land-cover dynamics derived from lower resolution datasets such as ESA CCI and HYDE 3.1 or from partial excerpts from Bolivian authorities.

Singapore is a Party to the United Nations Framework Convention on Climate Change (UNFCCC) and reports regularly statistics of its greenhouse gas (GHG) emissions and removals from various sectors including for biennial update reporting. The National Parks Board (NParks) of Singapore provides statistics of the Land Use, Land Use Change and Forestry Sector (LULUCF), and has engaged the Austrian Natural Resources Management and International Cooperation Agency (ANRICA) to establish a national GHG Monitoring, Reporting and Verification (MRV) System. The expert institutions involved in the consortium are the Environmental Agency Austria, the Federal Research Centre for Forests, and Joanneum Research.  They cover all aspects from remote sensing, biomass and soil inventories to modelling and IPCC reporting. The presentation deals with all the aspects of remote sensing, image analysis and geo-informatics. 

NParks has decided to implement the collection of precise and large scale activity data on wall to wall basis by means of remote sensing technology. Annual monitoring of land use (LU) and land use change (LUC) is based on different satellite data sources. The historic development, which is needed to analyse trends, is based on annual Spot image mosaics for the years 1989 to 2014 with a ground resolution of 5 to 20m.  Recent years are mapped based on very high resolution (VHR) Pléiades data which meets the country’s requirement characterized by heterogeneous spatial LU distribution and the high dynamic of LUC. 

About a dozen IPCC-compatible land-use categories out of four land-cover classes are distinguished based on rules jointly defined by the consortium. The baseline data collection carried out for the year 2013 served as reference for the mapping of the time series. Ground truth collection was performed by means of VHR data interpretation, field work, and additional data sources provided by NParks. Classification of LUC-categories within each year was carried out based on this ground truth. We chose segmentation of radiometrically calibrated image mosaics and subsequent classification of land cover, followed by rule-based assignment of LU and LUC as a well proven and traceable approach.

The LU maps of the years 2015-2017 could be substantially improved by the use of Pléiades data with an enhanced geometric resolution of 0.5 m.  Whilst the LU category definitions remain consistent over the entire time series, LU area borders are much more accurate for the current and future LUC assessments. Furthermore, as innovative aspect 3D information derived from stereo image pairs allows for significantly improved distinction between high and low vegetation categories. The challenges of different resolution and accuracy between historic and contemporary data were tackled by inventive techniques supported by redundant mapping of both Spot and Pleiades data within the years 2014 and 2015. The results are a LU map for each year as well as annual LUC statistics being essential components for the biennial reporting.

This is possibly one of the few applications where very high resolution images are used on a wall-to-wall format for purpose of reporting of carbon emissions/removals at a country level.

Copernicus Earth Observation data provided by the optical multi-spectral Sentinel-2 satellites (10m to 60m resolution) allow systematic and regular observations (every 5 days) of our planet. Complemented by images from other contributory missions, such as SPOT6-7 (1.50m) or Pléiades (50 cm) very high spatial resolution optical systems, these multi-resolution EO data are particularly adapted for the monitoring of territories at regional and local scale.

Based on the combined exploitation of these EO data and local Data Bases, Copernicus 'downstream' services are being set up at the regional or local level for the delivery of geo-information  in order to support the decision making process of  territorial  actors in charge of  spatial planning and regional or local implementation  of environmental politics within the framework of climate change  adaptation (French Grenelle laws, COP21 decisions, European directives, ...), and especially in their activities linked  to urban planning  management (e.g. for France: PLU, SCOT, …) which includes landscape artificialization limitation and space consumption regulation. 

In the Grand Est French region (North-East France), within the framework of the H2020 Eugenius ongoing project, an Urban Growth Monitoring service has been set up by ICube-SERTIT. This service contributes to fulfill the information needs of urban and land planning actors in charge of the preparation, the setting up and the monitoring of local urban plans, with the yearly provision of objective and up-to-date geo-information on urban developments with associated statistics.  At a global municipality scale, urban and peri-urban growing areas are derived from multi-temporal Sentinel-2 data, whereas the detailed situation assessment, in the local urban plan (PLU) or at the cadastre scale, is obtained using higher spatial resolution EO data, such as Pleiades or SPOT6/7, which allow the refined mapping of new constructions within the authorized built up sectors and the associated evaluation of remaining available space.

These Copernicus data exploitation capacities for local urban development policies assessment are illustrated by urbanization growth monitoring products realized in Alsace (GE region) for the ADAUHR (Urban technical agency of the Haut-Rhin department ) and for the DREAL (Ministry of Environment regional representative), with the use of historical or recent satellite observations between 2013 and 2017.

The idea of Warsaw aeration corridors had been arised in 1916 and was adapted to the present times in 1992 and 2006 in the plannings documents which described Warsaw spatial development conditions. The idea assumes existence of open green areas, wide communication routes and other linear areas which are not bulit-up (for instance Vistula). The goal for creation these corridors has been to establish the air exchange between areas around the city (especially green areas) and downtown. The free movement of air masses is aimed to improve of the urban climate (for example to mitigate urban heat island effect) and to reduce air pollution.

The research involved the analysis of buildings which were on the areas and are specified as a aeration wedges, in the following documents:

Local General Development Plan of Capital City of Warsaw  (approved in 1992);

Ecophsysiographic Study developed for Study of Conditions and Directions of Spatial Development of Capital City of Warsaw (approved in 2006).

Moreover, in Warsaw as in the whole Poland, western wind prevails. Knowing for that fact, the verification of the structure of new buildings in relation to its aerodynamics has been performed.

The analyses were carried out for years: 1992 – based on Landsat 5, 2006 – based on Landsat 7, 2015, 2016, 2017 based on Sentinel-2. There were also analysed  the data published by  Chief Inspectorate of Environmental Protection concerning  air quality and air pollution, especially the data of  PM2.5 particles which causes dangerous effect on human health. Besides, the data of wind speed and wind direction published by the Institute of Meteorology and Water Management were used.

As a result of research, it was found that aeration wedges are constantly built-up. The biggest changes took places in the Mokotów corridor (south part of Warsaw) and the Bemowo (western part of Warsaw) corridor, and in the Bródno corridor on the east of the city. These areas were mostly built-up by multi-familly housing. For instance: in the Mokotów corridor there were built new estates - Marina-Mokotów (8 hectares) and EKO-PARK (18 hectares), in the Bemowo corridor - Artystyczny-Żoliborz ( 11 hecatres) and Schopping Centre  (7 hectares), in the Bródno corridor - Zielone-Zacisze (8 hectares) and estates along Głębocka and Skarbka z Gór streets (67 hecatres and 49 hectares). The fewest changes have taken place in the wedges on the south of Warsaw: Wilanów corridor, Podskarpowy corridor and the corridor along the Vistula.

In addition, during the pre-research it was observed that new buildings have been situated in unfavorable way. Primarily, it applies multi-familly housing in Bródno corridor. New buildings are the walls and barriers to the air masses coming to the downtown. Because of that, air exchange is distrubbed.

The Poster will present the unfavourable conditions for proper exchange of air due to the structure of old and new buildings located in the largest city in Poland in relation to changes in air pollution and meteorological conditions. The satellite data will be applied to present monitor the changes in the city conditions.

In certain land regions with low latitude, as in the Amazon region,
it is essential to use radar images to obtain data where intense
cloud cover prevails. In view of the above, this study aimed to analyze
the data quality of the urban, rural and forest area for the double
polarization HH and VH of the Radar Sentinel 1-SAR after filtering
Speckle noise. A scene was acquired in the Central Amazon region on
January 30, 2017, in the IWS Interferometric Wide Swath mode, and
processed at level 1. The preprocessing of the image was performed in the
following methodological sequence: Appy Orbit File, Radiometric Calibrate
(σ0), Geometric Correction (Range-Doppler Terrain Correction) and Speckle
Filter. The following were performed: Frost 3x3, 7x7 and 11x11; Gamma Map
3x3, 7x7 and 11x11; Lee Refined; Lee 3x3, 7x7 and 11x11; IDAM 1, 2 and 3;
Average filter 3x3, 7x7 and 11x11 and Boxcar 3x3, 7x7 and 11x11 in Snap 5.
0 software. We used the post-filtration evaluation metrics, Mean Square
Error - MSE, Mean Absolute Error - MAE, SNR Signal Ratio, Edge
Preservation Index - EPI, Pearson Correlation Coefficient, Contrast
Distortion - DCON and Equivalent Number - NEL. The frost filter
with 3x3 window obtained the best metrics for VH polarization. The Lee
filter with 3x3 window obtained the best metrics for VV polarization.
The urban area in Parintins, AM presented backscatter values ​​between -5.
1db to -10db, while the Amazon River had values ​​less around -19db and
the Amazon forest -14db. It was concluded that it was possible to
characterize well the spectral responses of the urban area with the VV
polarization. In heterogeneous areas such as the Central Amazon, the VH
polarization has a better ability to characterize targets with spacing,
as in the case of forests.

Urban deformation monitoring is one of the most important interferometric SAR applications. This work will be focused on the deformation monitoring using C-band SAR data acquired by the Sentinel-1A and Sentinel-1B satellites. It will focus on the monitoring of urban and peri-urban areas. It will consider two different and complementary approaches: Persistent Scatterers Interferometry (PSI) and SAR tomography (TomoSAR). The PSI technique is a differential interferometric technique (DInSAR) based on a phase-only model. It was proposed almost two decades ago and has matured considerably in the last years as the main technique to measure and monitor land deformation from space. TomoSAR is a multi-dimensional imaging technique that has proven its ability in localizing the scatterers, reconstructing the structures and analyzing the displacements and thermal contributions. SAR tomography utilizes a stack of complex-valued images to discriminate the superimposed scatterers in the same range-azimuth cell, by synthesizing an elevation aperture to reconstruct a full 3D reflectivity profile along azimuth, range, and elevation. The possible advantage of TomoSAR over classical interferometric methods consists in the potential capability of improving the detection of single scatterers presenting stable proprieties over time (Persistent Scatterers, PS). Additionally, a second advantage is the ability to detect multiple scatterers interfering within the same range-azimuth resolution cell. In an urban environment, when only single dominant scatterers are present in each range-azimuth resolution cell, both methods can be exploited to estimate the altitude, deformation rate and thermal expansion of a subset of reliable scatterers, which are selected on the basis of different criteria.

In this work, we will compare the PSI and TomoSAR results based on a set of C-band Synthetic Aperture Radar data from the Sentinel-1A and Sentinel-1B satellites. The chosen test area is the metropolitan area of Barcelona (Spain). The analysis will consider:

-          The spatial density and distribution of the scatterers;

-          The deformation velocity;

-          The residual topographic error;

-          The thermal dilation component;

-          The quality of the derived deformation time series.

During the last years, we saw a growing interest in satellite image analysis including semantic and quantitative content description of images, physics-related classification of image segments, the quest for the causes and effects of changes over short and long time periods, and disaster relief support tools. These remote sensing applications are most often using optical datasets; fewer users employ SAR (Synthetic Aperture Radar) datasets. However, both cases call for the identification of land cover / land use details within the full image area. As modern imaging sensors are characterized by high spatial resolution and a wide field of view, we face a high diversity of target types, target objects, as well as temporal changes and their spatial interrelationships. Therefore, we need reliable benchmarking tools to ascertain the actual quality of our image analysis and retrieval results. Primary benchmarking tools for automated benchmarking are collections of selected reference images with known semantic content and characteristics that can be employed for quantitative and comparative image analyses. For optical sensors there exist several well-known and publicly available datasets comprising typical remote sensing image patches, while comparable SAR datasets are very scarce.

For our user-oriented application cases no well-known high-resolution and publicly available SAR reference datasets exist. In our case, we collected from all over the world 1,000 urban and industrial areas together with their infrastructure TerraSAR-X images and 75 Sentinel-1 images [1, 2], and assigned them to typical cases of acquisition parameters and target areas. Then about 30% of these images were selected for training and generating a reference dataset using our benchmarking data mining system. We tiled these training images into about 250,000 patches (in case of TerraSAR-X) and 180,000 patches (in case of Sentinel-1), and labeled each patch with a semantic category. The selection of the training images was controlled by SAR experts and was driven by the typical requirements of our application cases and their diversity. When we aim at a systematic and universal benchmarking of our classification quality, we need an established procedure that includes the methodology described in [3], complemented by a manual annotation for the remaining patches left unclassified by the data mining system, and a visual inspection of the results.

The basic finding of this paper is the verification that we can perform and obtain an automated generation of benchmarking SAR datasets with good quantitative performance measures.

[1] TerraSAR-X archive portal, 2018. Available: http://eoweb.dlr.de/.

[2] Sentinels Scientific Data Hub, 2018. Available: https://scihub.copernicus.eu/dhus/#/home.

[3] C. O. Dumitru, G. Schwarz, and M. Datcu, “Land Cover Semantic Annotation Derived from High-Resolution SAR Images”, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 9(6), pp. 2215-2232, 2016.

Population maps are dependent on the availability of up-to-date, spatially-explicit population census data, typically created by disaggregating population estimates from census units and attributing them to mapped human settlements footprints. Previous studies have demonstrated the ability to distribute population more precisely over land by using satellite-derived estimates of settlement locations. The Democratic Republic of Congo (DRC) represents a challenge to population estimation because of a large and mostly rural population and a lack of data on settlement locations. We report here a study that generated three products for the DRC: 1) a map of settlement distribution, 2) a new estimate of national population, and 3) a map of population distribution within settled areas. We first mapped human settlements using a supervised, a decision-tree classification of Landsat imagery, assessed map accuracy and bias using stratified random sampling and reported the adjusted areas of settlements with known uncertainty. For the total population, we conducted a stratified-random sampling, with strata based on the settlements map. For each sample, we manually counted observable dwelling units, using high-resolution satellite imagery available on Google Earth, and multiplied by the national mean number of people per dwelling, provided by the 2013-2014 Demographic Health Survey (DHS). For the map of population density, we again used a decision-tree regression, this time of the population values from the same sample points against Landsat imagery. We estimate the total settled area in DRC at 12,930.17 km2, or 0.55% of the national land area (2,345,000 km2). We estimate DRC’s total population in 2013 at 101.4M inhabitants. Of these, 18.8% live within 300-m of all mapped settlements.  DRC remains a very rural country, where over 51.1 percent of the population live beyond 50 km from a major city. This study provides the first model-based estimate of national population as well as the first fine-resolution estimation of settlement and population distribution. These data can allow improved land use planning at the national and sub-national scales.

We have studied the spectral features of reflectance and emissivity in the pattern recognition of urban materials in several single hyperspectral scenes through a comparative analysis of anomaly detection methods and their relationship with amplitude and intensity Sentinel 1 products in order to improve information extraction processes. Spectral ranges of the visible-near infrared (VNIR), shortwave infrared (SWIR) and thermal infrared (TIR) from hyperspectral data cubes of AHS sensor and Sentinel-2 of three test areas of Alcalá de Henares city (Spain) have been used.

In Anomaly Detection (AD) it is assumed no prior knowledge of the targets, thus, the pixels are automatically separated according to their spectral information, significantly differentiated with respect to a background, either globally for the full scene, or locally by image segmentation. Several experiments on three urban scenarios and semi-urban have been designed, analysing the behaviour of the standard RX anomaly detector and different methods based on subspace, image projection and segmentation-based anomaly detection methods. A new technique for anomaly detection using hyperspectral data called DATB (Detector of Anomalies from Thermal Background) based on dimensionality reduction by projecting targets with unknown spectral signatures to a background calculated from thermal spectrum wavelengths is applied. First correlation analyses between hyperspectral anomalies and Sentinel-1 products and their consequences in non-supervised classification in order to discuss the extraction information processes are presented.

Among the most burning global issues, urbanization is considered as one major present and future challenge. The amplitude of this phenomenon and its impacts on the society, the environment and the economies of countries around the world has been used to promote research agendas focusing on the spatial dimension of urbanization. This has stimulated the development of several research activities aiming to improve the knowledge about the presence of human settlements and their dynamics. As a consequence, we are witnessing a proliferation of global Earth Observation (EO)-based settlements products derived from different sensors (Landsat, MODIS, TerraSAR-X, Sentinel-1 and Sentinel-2), using different methodologies and producing usually different thematic contents. With the availability of a range of EO-based settlements products, users are confronted with questions related to the quality of the different global products, their compatibility, comparability and consistency.

These legitimate questions stem from the need to understand the uncertainties and limitations of these datasets in different applications. In parallel to the production of new generations of global human settlements maps, the Global Human Settlement Layer (GHSL) initiative is committed to evaluate the quality of the global EO-based settlements layers. This activity includes wall-to-wall comparison of the settlements layers to existing reference datasets and an analysis of agreement/disagreement between the existing products.

The evaluated datasets comprise: the GHSL products derived from Landsat and Sentinel-1, GlobalLand30, Global Urban Footprint, Global Human Built-up And Settlement Extent and the Global Urban Land Maps (recently produced using a Normalized Urban Areas Composite Index and developed in the Google Earth Engine platform).

In this contribution, the protocols developed for the qualitative and quantitative assessments will be presented. Additionally, we perform a sensitivity analysis to the several parameters that affect the results of the accuracy assessments, namely, the spatial resolution, the size of the sampling unit and the sampling grid.

Beyond the scientific challenges associated with the development of a “fair” validation protocol, the question of how to translate the results of this global quality assessment into recommendations for the end-users of EO-based settlements layers will be also addressed.

  Urban forests are a key component of urban ecosystems. The type of urban forests is an important piece information required for urban ecological regulation and environmental research. In the study, Xuzhou, the winner of the National Ecological Garden City evaluation in 2016, was selected as a case study. We made use of the robust image classifier support vector machine (SVM) to map urban forests of the study and discriminate urban forest vegetation of different types from a 10-m Sentinel 2A image acquired on 24 July 2017. In order to examine whether urban vegetation abundance can improve discrimination, we performed the linear spectral mixture analysis (LSMA) and successfully identified three different vegetation endmembers, namely broad-leaved forest, coniferous forest and low vegetation. Based on including the vegetation abundances for classification or not, four different SVM classification models were constructed for comparative analysis. To confirm the effect of vegetation abundances on classification accuracy, machine learning classifiers like random forest (RF), artificial neural network (ANN), and quick unbiased and efficient statistical tree (QUEST) were also tested in the study. All the classification results were assessed with field-based observations through confusion matrices. Through our results, we conclude that:

   (1) The spectral and spatial resolutions of Sentinel-2A image data are suitable for mapping urban forests and discriminating urban forest types. The Band 8A (Narrow NIR), Band 7 (Vegetation Red Edge) and Band 12 (Shortwave Infrared) of the data show high correlations with urban forest types in the order of feature importance, which were given by RF.

   (2) The LSMA method can derive accurate vegetation endmember information, with a root mean square error of 0.005. Vegetation abundances obtained from LSMA can improve the SVM classification accuracy significantly from 89.13% (Kappa = 0.82) to 89.86% (0.83). Such an improvement in overall accuracy is not only applicable to SVM but also to other classifiers, from 79.71% to 83.45% for QUEST, from 79.29% to 85.00% for ANN, and from 80.58% to 84.21% for RF.

  Our future work will be trying to understand if additional spectral and textual features, derived from the Sentine-2A image data, can further improve the discrimination and to find an optimal set of features for highest classification accuracy.

The application of EO data for climate resilience and disaster management in urban areas relates to various processes. A specific case for the World Bank City Planning Lab programme which was part of the EO4SD Urban project focused on the provision of Flood History and -Risk as well as Land Motion (subsidence) to stakeholders in Indonesia. In addition, related EO products such as Land Use and its Change as well as Transport Infrastructure are also produced, as they are highly relevant for assessing sustainability of cities with respect to climate resilience and disaster management.

The full operational mapping exercise is demonstrated for the Indonesian city of Semarang. Semarang is located at the north coast of Java, Indonesia, in an area of active land subsidence. Flooding is common; specifically the coastal zones have experienced increased flooding due to the sinking land. In future, potential sea level rise can further aggravate the situation.

In a first step, separate mapping products were derived from various EO data sources. We used VHR and HR optical satellite data for the classification of detailed land use classes and transport infrastructure. The Land Use/Land Cover product for Semarang has an overall mapping accuracy of 85.32% with a confidence interval (CI) ranging from 83.04% to 87.59% at a 95% CI.For flood history, we employed all available historic EO data from optical, but also from RADAR systems. In addition, also non-EO data such as maps, in-situ observations, reports, etc. were used to complete the flood history assessment. For assessing the currently ongoing terrain subsidence, time-series data from Sentinel-1 was used in an interferometric processing chain.

In the second step, we generated the current flood risk map by combining the flood history with the subsidence map. Based on this combination, we re-evaluate existing risk areas and identified potential new risk areas. One shortcoming in this evaluation was the lack of a highly accurate terrain model, which is not available for the area. However, rough terrain information was also included in the expert evaluation system to calculate risk areas.

Finally, in a third step, all individual products were combined in a GIS to generate meaningful spatial analytics, such as the share and length of transport infrastructure prone to flooding and/or to damage due to subsidence. Another example is the share of LU class “Residential and Public Urban Fabric Areas”. From these areas, 83 % are not affected by flood risk; about 8 % of the areas are in medium risk areas, 7 % in high and almost 2 % in very high flood risk areas.

The preparation and work done for this article is part of the European Space Agency supported Programme “Earth Observation for Sustainable Development (EO4SD) - Urban”. Joanneum Research would like to thank especially the World Bank City Planning Labs programme and Team for their support; this includes Gayatri Singh the Task Team Leader, Champaka Rajagopal and Aurora Dias Lokita who facilitated the user interactions and data/information collection from the City.

Thematic area :

Resilient Cities ( including disaster risk reduction in cities)

abstact: 

In the last four decades, the city of El Hajeb/Morocco has seen its population tripled, from 12601 population in 1971 to 35282 population in 2014. This population growth has led to the expansion of the city reaching areas less stable and unsuitable for urban sustainable development driven by the presence of sinkholes and other karst landforms around this city. Besides the expansion of urban areas, the growing population are conducting extensive agriculture activities, enhancing groundwater extraction, using chemical and products for crops growth and crops protection, and farming in areas vulnerable to pollution. All these activities have impact on the stability of the landscape, especially vindicated by the recent formation of karst depressions and collapsing sinkholes.  In order to contribute to the recent effort of protecting this region, reducing the disaster risks, and restore the resilience of this  city, we used freely available satellite images of ASTER, Copernicus Sentinel and Landasat programs. On these images we extracted useful information both on the Geology of the Tabular Middle Atlas (TMA) of El Hajeb which is dominated by Liassic carbonate rocks (dolomite and limestone), quaternary basalts and Triassic clays and on Land cover/ land use. By combining various images processing techniques such as supervised classification, normalized difference vegetation index (NVDI) and normalized difference moisture index (NDMI) estimation and Sobel operator for image fractures extraction we noticed a number of depressions in eastern part of El Hajeb which can be warnings for possible future collapse and sinkholes formation, we also identified collapsing sinkholes visible on the surface. Infiltration of surface water in agriculture areas may also fostering this process by increasing gradual erosion of the carbonate rocks. The results showed that a big number of karst landforms in this area are located on NE-SW fault system affecting mainly dolomite and limestone of Lower and Middle Liassic, while NW-SE fault system was identified as pathway which drains water from TMA to the basin of Saiss located in North of El Hajeb. The construction of new buildings should consider all these information in order to minimize the risk of their collapse, at the same time ensuring the quality of groundwater aquifer essential for the prosperity of the City of El Hajeb.

ISPRA and Municipality of Rome are working together on land cover analysis in Rome, in order to develop a new cartography which is able to highlight the artificial land cover and soil sealing, a current problem that concerns many cities in the world.

Land cover and soil sealing knowledge is a main topic for environmental protection, risk assessment, climate change adaptation, civil protection and safety. Land degradation affects water and air quality, biodiversity and climate change, as well as economy because of the ecosystem services decline.

This research  concerns the city of Rome, in order to develop a more detailed soil sealing map which covers the whole city area.  Therefore interpretation of multitemporal satellite images (scale 1:2000-1:5000) has been made, integrating Urban Atlas data (2016), Open Street Map data (2017) and Carta Tecnica Regionale of Rome (2002), using the III level classification of the National land consumption map.

It is part of National Environmental Protection System work, coordinated by ISPRA, which annually produces a new map using Sentinel 1-2 data and photointerpretation, through a I/II/III level classification: multitemporal Sentinel images are used to do a semi-authomatic classification in order to distinguish annual land cover changes. For each Sentinel 2 image NDVI index is calculated; the maximum NDVI value of each annual series is compared with the previous one, obtaining a raster image which shows how the vegetation has changed during years: the results are different pixel numbers, which can be caused by soil sealing or other reasons (for example fields which are no more cultivated). Thereafter images are visually verified and classified with this legend:

I level: Artificial Surfaces and Constructions / Natural Material Surfaces

II level: Sealed Artificial Surfaces and Constructions / Non-Sealed  Artificial Surfaces and Constructions

III level: contains different types of land cover (buildings, streets, airports, natural deposits…)

The resulting map shows land consumption and soil sealing in Rome, which will be compared to the National one, through two different methods, in order to verify if vector data are more accurate than raster ones :

-compare a raster image (National soil sealing map) to a vector image (Rome soil sealing map)

-compare two raster images

This map constitutes the base for statistical analysis and to derive several  spatial indicators and landscape metrics.

These analysis will be followed by  a focus on hydrogeological risk areas (i.e. Tevere and Aniene basins), integrating soil sealing data with ISTAT population ones, hydrogeological information and cartography, as well the flood movement analysis, in order to study possible interventions which could increase the area resilience.

This map and statistical studies on soil sealing inside the city area strive for being  available to future administrations for a urban development planning which includes the natural resources defense, real conditions of the city soil and the most urgent Rome Municipality issues.

With the impacts of climate change already being realised in the urban space, many cities are focused on the development and implementation of specific response plans to both mitigate and adapt to two key risks: heat and flooding. With the United Nations estimating that the number of urban residents is growing by nearly 73 million every year, learning lessons across urban systems has never been more critical. In response to the sheer numbers of increasing populations, large financial investment in construction, energy, public transport and other aspects of the urbanisation process will be required. This opens scope for new physical and social infrastructure and economy, as well as creating a need for improved land-use planning and decision making to build risk management and adaptation into urban design.

This presentation focuses on one case study, the Urb-ADAPT Project. Urb-ADAPT aims to identify and assess current and future climate risks for the Greater Dublin Region by developing a range of Climate Vulnerability Indices to support planning for key urban climate impacts. An integrated approach, that analyse current and projected climate data in combination with socioeconomics, environmental and population information by means of different remote sensing and statistical techniques, is the basis of this project. The work here described then demonstrates how these results can be leveraged through international agreements like the New Urban Agenda (NUA) and across other urban spaces and to help stimulate transformation to sustainability. Involving stakeholders and by providing recommendations and action strategies around planning, construction, development, management, and improvement of urban areas such as Dublin, projects like Urb-ADAPT can help to reinforce the NUA tenant that urbanisation can be a powerful tool for sustainable development for both developing and developed countries.

The estimation of artificial areas over Europe is necessary to characterise urban sprawl and contribute to indicators related to sustainable development goals (SDGs). Several land products provide information on the location, extent, structure, nature and changes of artificial land in Europe.

a)       CORINE Land Cover coordinated by EEA

b)      The Copernicus Imperviousness layer for which the production is managed by EEA.

c)       The European Settlements Map (ESM) produced by the JRC.

d)      The Global Urban Footprint+ (GUF+) produced by the DLR

e)      The Copernicus Urban Atlas for which the production is managed by EEA.

f)        The LUCAS point survey conducted by Eurostat.

g)       Validation samples for some the above mapping products.

The four first products are maps with different scales and concepts of artificial land. The Urban Atlas maps artificial land in over 800 Functional Urban Areas (FUAs). LUCAS is a general purpose survey. It presents limitations due to a complex organization, but has a good potential to be combined with maps. The validation samples selected for the validation of specific maps can be also used as independent data sets.

The direct area estimation by pixel counting in a classified image, or equivalent approaches, is known to have a bias because commission and omission errors are not balanced. The risk of bias in pixel counting is often stronger for classes occupy a small proportion of the total area.

The bias can be estimated and corrected with the help of a sample of reference data. This study compares the results obtained from the datasets listed above to best characterize (i) the extent of artificial areas and (ii) their evolution by combining them with a suitable sampled reference dataset.

Differences on area estimation are often due to the heterogeneous concepts of “artificial”. Comparing changes is more delicate but more informative, often raising anomalies in the datasets.

The reference samples must have associated probabilities to be used for the correct extrapolation of confusion matrices. There are several methods to correct the bias of pixel counting, such as calibration and regression estimators. The reasonable choice depends on the type of data.

Initial results based on the assessment of HRL imperviousness 2012 show that the bias of the naïf (direct) estimation of impervious area from classified satellite images is above 20%, even if the overall accuracy of the classification is above 98%. This observation has an implication on the use of remote sensing for area estimation that is not new. New results will be presented based on the assessment of HRL Imperviousness 2015 and the recently re-analysis of the HRL Imperviousness time series for 2006-2009-2012 as well as ESM, GUF+, CLC , UA and LUCAS datasets.

The city of Santo André is located in the Metropolitan Region of the State of São Paulo and constantly suffers from floods. This research has the aim to study the floods that occur in Santo André focusing on the morphometric and hydroclimatological analysis of the drainage sub-basins of the Tamanduateí river that drain the city: Tamanduateí Médio I, Oratório, Ribeirão dos Meninos and Guarará. Initially, the recurrence of the flood events was analyzed through the historical data of the Civil Defense regarding the occurrence of the events from 2001 to 2016. Therefore, it was possible to detect and map the districts/neighborhoods that were most affected by urban floods. After that, the morphometric features of the watersheds were studied, followed by the analysis of the rainfall. The results showed a total of 273 flood events during the 16 years mentioned, most of them occurring in summer. The morphometric analysis indicates that the Guarará watershed (with 53% of the events) is naturally prone to floods because of its relief characteristics. The Tamanduateí Médio drainage basin has a medium tendency to flooding occurrence in terms of geometry and also the relief. The Ribeirão dos Meninos watershed (33% of the cases) is less prone to flood occurrences when compared to the other drainage basins nearby. The Oratório sub-basin is the one with the smallest number of flood occurrences, but according to its morphometric features, it would be more vulnerable to floods than the Ribeirão dos Meninos watershed. Through the spatial analysis of the rainfall for specific summers, it was possible to observe that in the south portion of the urban area the rainfall values were lower, different from the climatic pattern, as well as higher values in the northeast region, where the Oratório watershed is located, indicating the spatial annual variation of the rainfall. According to the frequency of rainfall intervals, the flood events were mostly related to the rainfall that occurred in the same day of the event and also to five days of accumulated rainfall. The precipitation intervals that were more frequently associated with the events were 15-30 mm and 31-50 mm. According to the results the flood events are recurrent even during situations of less rainfall. The drainage basin whose morphometric features make it more prone to flood is the Guarará watershed. For the other basins, the flood events might have a higher relation with the drainage system, incapable to drain the amount of rainfall that it receives.

According to the UN DESA report “World Population Prospects: The 2015 Revision”, The world population is expected to grow 33% by the year 2050. With the highest rate of population growth, Africa is expected to account for more than the half of the world’s population growth between 2015 and 2050. The study area presented in this paper is the Republic of Mozambique, an African country with 70% of its population of 28 million (2016) living and working in rural areas. The Real gross domestic product (GDP) of the country was 3.7% in 2017 shows it’s struggle of poor macroeconomic stability and investment of private sector.

High income countries often have extensive mapping resources and expertise to create reliable and accurate building maps and population databases, but across the low-income regions of the world, relevant data are either lacking or are of poor quality. For low-income regions of the world, accurate maps of human population distribution together with the knowledge of building types and its quantitative measures can play an essential part in planning for elections, calculating per-capita gross domestic product (GDP), poverty mapping, city planning, disaster management amongst countless other applications.

The rapid growth in availability of high resolution satellite imagery, computing power and expansion of geospatial analysis tools over the past decade are providing new opportunities to solve such problems. The use of high resolution images, geospatial data and road network together with state of the art machine learning technology can improve the understanding of human population distribution and building type estimation, which is necessary to predict the future infrastructure management for increasing population because it can be expanded to a bigger scale easily unlike the traditionally used method based on human visual interpretation and survey data collection.

In this paper, we proposed a methodology to classify the types of buildings in three classes; high- rise residential buildings, low-rise residential and non-residential buildings. We have used state- of-the-art machine learning algorithm on the combination of mobile phone sample data collected from a survey, high resolution satellite images, digital surface model and night time light data to extract the building footprints and classify the types of the buildings. The comparing results indicated that our methodology classified types of buildings efficiently with the accuracy of 84%.

Keywords: high-rise residential buildings, low-rise residential, non-residential buildings, digital surface model

The increasing quantity of very high resolution (VHR) satellite data leads to the need of processing procedures that work (semi)automatic, and generate accurate products. The final goal of such automatic processing chains is generation of geometrically corrected images through orthorectification. While accurate automatic orthorectification of medium and high resolution optical satellite images is already viable, the geometric correction of VHR images (resolution 0.5 m or less) is still very challenging.

Our research is oriented towards the development of a procedure for automatic orthorectification of VHR panchromatic images of different sensors. The procedure is divided into four main modules: extraction of ground control points (GCPs), geometric modelling, orthorectification and image mosaicing (only when generating true orthoimages).

The core of the automatic GCPs extraction module is a two-step algorithm that utilizes vector roads as a reference layer and delivers GCPs for high resolution images. The matching is done between the reference data (road map, orthophoto chips) and roads extracted from satellite images with watershed segmentation based on a gradient image and primitive classification (decision tree). The module was upgraded with a super-fine positioning of individual GCPs onto an aerial orthophoto. The final stage utilizes least square matching for sub-pixel positioning.

Geometric modelling is done with a rational function model (RFM). It can work with a single image or a block of images with a joint adjustment. GCPs common to all images are selected during the extraction algorithm that always targets the same crossroads.

Orthorectification can be done either by an indirect or a direct model.

The developed algorithm was tested on Pleiades and WorldView-2 panchromatic images of 0.5 m resolution. The studied images cover the area of two major cities in Slovenia. Some of the images were acquired during the same pass.

The two-step point extraction algorithm is capable of achieving accuracy of approximately two pixels on panchromatic data. The enhanced super-fine matching improves the accuracy to values around a pixel. An even better accuracy can be achieved with the use of least square matching: when compared to the reference national aerial orthophoto the accuracies of the orthorectified images automatically produced with the RFM usually reach sub-pixel values at independent check points.

Besides the generation of orthoimages the developed procedure is also designed to support the creation of satellite true orthoimages. It combines GCPs extraction for urban areas and the block adjustment algorithms, which are the basis for an accurate true orthoimage generation procedure. Although the development of this final module is not yet completed, the first results are very promising.

Supported by the ESA SAR4Urban and Urban-TEP projects, the German Aerospace Center (DLR) has lately produced the World Settlement Footprint (WSF) 2015, i.e. a binary mask outlining the extent of human settlements globally derived by means of 2014-2015 multitemporal Landsat-8 and Sentinel-1 imagery acquired at 30m and 10m resolution, respectively. For quantitatively assessing the high accuracy and reliability of the layer, DLR has recently carried out in collaboration with Google an unprecedented validation exercise based on a huge amount of ground-truth samples labelled by crow-sourcing photointerpretation. In particular, to this purpose a statistically robust and transparent protocol has been defined following the state-of-the-art practices currently recommended in the literature.

In the response design, since input data with different spatial resolutions have been employed to generate the WSF2015, a 3x3 block composed of 9 cells of 10x10m size has been used as spatial assessment unit. Furthermore, to cope with the different existing definitions of settlement, 4 possible labels were collected, namely buildings, building lots, roads/paved-surfaces and none-of-the-previous. Here, by means of an ad-hoc tool implemented by Google, each operator was iteratively prompted a given 3x3 unit on top of the available Google Earth reference VHR scene closest in time to the year 2015 and given the possibility of assigning any of the 4 labels to each cell.

In the sampling design, to include a representative population of settlement patterns, 50 out of the ~18.000 tiles of 1x1 degree size composing the WSF2015 have been first selected based on the ratio between the number of estimated settlements (i.e., disjoint clusters of pixels categorized as settlement) and their area. Then, for each of them, stratified random sampling has been applied; specifically, since the settlement class covers a sensibly small proportion of area compared to the merger of all other non-settlement classes (~1% of Earth’s emerged surface), an equal allocation has been chosen where 1000 samples have been randomly extracted both for the settlement and non-settlement class from the WSF2015 and used as centroid of the 3x3 reference units to label by photointerpretation. Such a strategy resulted in an overall amount of 50 x (1000 + 1000) x (3x3) = 900.000 cells labelled by the crowd.

Finally, in the analysis phase different standard measures for assessing the accuracy of the settlement extent maps have been computed, including the percentage overall and average accuracies, the kappa coefficient and the percentage producer’s and user’s accuracies for both the settlement and non-settlement class. This has been carried out by iteratively considering as “settlement” all areas covered by: i) buildings; ii) buildings or building lots; and iii) buildings, building lots or roads/paved-surfaces.

Overall, results assess the great effectiveness of the WSF2015, which also outperforms all other currently existing similar global layers including the Global Urban footprint (GUF), the Global Human Settlement Layer (GHSL) and the Global Land Cover (GLC) 30.

The RUS Service aims to promote the uptake of Sentinel data and supports the scaling up of R&D activities with data coming from the Copernicus programme of European satellites.

RUS Service is configured in a scalable cloud environment that offers the possibility to remotely store and process EO data. The RUS Service is providing support from a helpdesk and a team of EO and IT experts, who can address any request coming from beginners or skilled practitioners. Cloud ICT resources are procured with Free and Open-Source Software and are tailored to meet the user’s needs. The RUS service also proposes on-site training sessions, like webinars and online materials. The RUS Service is offered at no cost and is available for a large community of users and various types of institutions.
The objective of this poster is to present and introduce different examples of the exploitation of the RUS Virtual Machines with Sentinel datasets related to urban areas applications.

The RUS Service is funded by the EC, managed by ESA, and operated by a consortium of European companies led by Communications & Systèmes – Systèmes d’Informations (CS-SI France) and its partners.

Mapping the Earth’s surface and its rapid changes with remotely sensed data is a crucial tool to understand the impact of an increasingly urban world population on the environment. However, the impressive amount of freely available Copernicus data is only marginally exploited in common classifications. One of the reasons is, that measuring the properties of training samples, the so-called ‘features’, is costly and tedious. Furthermore, handling large feature sets is not easy in most image classification software. This often leads to the manual choice of few, allegedly promising features. In this research, we use the computational power of Google Earth Engine and Google Cloud Platform to generate an oversized feature set in which we explore feature importance and analyse the influence of dimensionality reduction methods. We use Support Vector Machines (SVMs) for object-based classification of satellite images - a commonly used method. A large feature set is evaluated to find the most relevant features to discriminate the classes and thereby to contribute most to high classification accuracy. In doing so, one can bypass the sensitive knowledge-based but sometimes arbitrary selection of input features.

Two kinds of dimensionality reduction methods are investigated. The (1) feature extraction methods, which transform the original feature space into a projected space of lower dimensionality; i.e. Linear Discriminant Analysis (LDA) and Independent Component Analysis (ICA). And (2) filter-based feature selection methods that rank and filter the features according to a chosen statistic; i.e. being chi-squared test, mutual information and Fisher-criterion. We compare these methods against the default SVM in terms of classification accuracy and computational performance. The classification accuracy is measured in overall accuracy, prediction stability, inter-rater agreement and the sensitivity to training set sizes. The computational performance is measured in the decrease in training and prediction times and the compression factor of the input data. We conclude on the best performing classifier with the most effective feature set based on this analysis.

In a case study of mapping urban land cover in Stockholm, Sweden, based on multitemporal stacks of Sentinel-1 and Sentinel-2 imagery, we demonstrate the integration of Google Earth Engine and Google Cloud Platform for an optimized supervised land cover classification. We use dimensionality reduction methods provided in the open source scikit-learn library and show how they can improve classification accuracy and reduce the data load. At the same time, this research gives an indication of how the exploitation of big earth observation data can be approached in a cloud computing environment.

The preliminary results highlighted the effectiveness and necessity of dimensionality reduction methods but also strengthened the need for inter-comparable object-based land cover classification benchmarks to fully assess the quality of the derived products. To facilitate this need and encourage further research, we plan to publish our datasets (i.e. imagery, training and test data) and provide access to the developed Google Earth Engine and Python scripts as Free and Open Source Software (FOSS).

Since the 1950s, the Nile Valley and Delta (NVD) in Egypt have faced an accelerated human settlement growth accompanied by weak planning, governance and coordination. Regardless of many physical plans which aimed to control this accelerating growth and preserve the agricultural land, many urban and rural settlements have expanded at the expense of the best fertile lands in the NVD. With 98% of Egypt’s population (estimated 104 Million inhabitants in 2018) settled in the NVD, vast areas of agricultural land have been lost to human settlement growth. Although many studies have addressed human settlement growth on agricultural in the Nile Delta (from Cairo to the Mediterranean Sea), there is a lack of knowledge on human settlements growth in the Nile Valley (south of Cairo). Therefore, we analyze human settlement growth on agricultural lands in Assiut Governorate (AG) as a case study between 1999 and 2016. AG consists of 235 villages and 12 cities with a combined population of 4.7 million inhabitants. The 1999 built-up area of AG was obtained from the National Authority for Remote Sensing and Space Sciences (NARSS). The built-up area of 2016 was extracted from one Landsat 8 scene using Masada 1.3 software from the Joint Research Center (JRC) . The standard Global Human Settlement Layer model settings were tuned to improve the accuracies for the study area. The results show that the built up area of AG between 1999 and 2016 has expanded by 490%, mainly at the expenses of the agricultural land. However, the rate of human settlement expansion varies over the study area. Most settlements (66%) have expanded by more the 200%. The comparison of  actual growth with recent development plans for most settlements shows the lack of government control over settlement expansion processes.  The variation of human settlement growth rate indicates that there may be different driving forces which accelerate or restrict human settlement growth in AG. To manage human settlement growth in AG and the NVD, it is important to identify and investigate such forces through further research.

The 21^st century is the first “urban century” according to the UN. In 2014, more than 50% of the global population (3.9 billion people) lived in cities and recent estimates suggest this will grow to 75% by 2050. The rate of increased urbanization is likely to be the largest in the developing world, creating new challenges for sustainable development.

The growth, often chaotic, of urban agglomerates has a significant impact on the environment. The UN Conference on Human Settlements has pointed at cities as the main cause of some of the global problems such as waste production and air and water pollution. However, it is not merely the natural environment, but the newly created urban environment itself that can pose risks to human health, key among which is the Urban Heat Island (UHI).

The UHI is an increase in air temperature inside an urban area due to the presence of materials that increase radiation absorption, the reduction of green areas, the reduction of natural wind circulation, and the reduction of shade. The increase in temperature associated with the UHI is usually in the order of 1°C to 3°C but can reach values up to 12°C. UHI effects are well known in principle, but they are difficult to measure, due to the limited number of urban meteorological stations and to the continuously changing urban landscape, especially in developing countries.​

Satellite data represent a new opportunity to reconstruct the dynamics of rapidly growing urban areas and correlate it with the occurrence and intensity of UHI effects. Hence, our objective is to explore the relationship between the urban growth and the intensity of UHI effects that pose a great problem in areas where heat waves occur regularly and are projected to worsen in the future.

Here, we use the city of Chennai (India) as a case study, with well-known recurrent heat waves. We investigate growth in urban area from the 1990s to date using archive synthetic aperture radar (SAR) data (ERS, Envisat) and the most recent Sentinel-1 acquisitions. These products are combined using innovative multitemporal and multisource SAR processing to reconstruct the evolution of the urban landscape. In particular, remote sensing data are used to classify urban and rural areas. The variation of the temperature over the time is then observed in areas classified as permanently urban, permanently rural, and changing from rural to urban. This way, it is possible to understand different warming trends and their correlation with the changing landscape.

During the last few decades, Doha, the capital city of the State of Qatar in the Arabian Gulf, has witnessed deep changes due to aggressive urbanization which involved wide-scale residential, industrial and commercial construction as well as the development of articulated infrastructure including modern roads and underground transportation facilities that totally changed the shape of the city. These changes have generated also a wide spectrum of environmental adverse effects such as the reduction of green areas, marine coastal areas fillings and modification of the city-sea interface. The city also witnessed around 10 folds increase in population since 1998. This caused severe pressure of services and other environmental components.

With the preparation of the country for the 2022 World Cup and the implantation of Qatar Vision 2030, it is expected that the rate of urbanization will be accelerated even more during the coming few years with a sharp increase in population due to the arrival of more working force.

The main goal of this paper is to monitor the increment of the urban area relative to the Doha City area by using a change detection approach based on Synthetic Aperture Radar (SAR) images. Innovative multitemporal (from the ‘90s of the past century up to today), multisensory (ENVISAT, RADARSAT-1, TerraSAR-X and Sentinel-1), multifrequency (C- and X-band) and multipolarization (HH, VV and VH) processing techniques to detect changes in the urban landscape are here exploited.  Urban maps are retrieved by evaluating the different backscattering responses of urban and non-urban areas due to their different geometric and electromagnetic characteristics. This different behaviour results in bright and dark pixels in the SAR image for urban and non-urban areas, respectively.

Preliminary analyses show an urbanization rate of about 5% for the 2015-2017 period, and an impressive 455% for the 2000-2017 period. Comparisons with ground-truth data show consistency with analysis and confirm suitability of SAR in urban growth monitoring.

After the launch of the Copernicus Sentinels 1, 2 and 3 by the European Space Agency, the availability of free and open Earth Observation (EO) data streams provides totally new opportunities for innovative scientific and commercial geo-information services. With the given spatial resolution and revisiting times, the potential of Sentinels missions to support a wide range of environmental, regional and urban planning and monitoring applications is high. Recently, the ERA.Net-RUS project GEOURBAN developed a set of EO-based environmental indicators for urban planning and a software tool for their on-line evaluation. GEOURBAN mainly focused on the local city level; however, planning in peri-urban and rural areas is particularly important for Russia, given its huge territory and its high number of large cities and scattered settlements.  Standard EO-based spatial datasets, such as the European Urban Atlas have proven to be quite valuable for various urban and spatial planning applications. However, this data just exists for the large and medium European cities, but not for the Russian ones. To this end, the main objective of the SEN4RUS (exploiting Sentinels for supporting urban planning applications at city and regional levels in Russia) project, funded by ERA.Net-RUS Plus, is to take into account the specific requirements of spatial and urban planning in Russia to develop indicators that effectively and efficiently exploit the information content provided by Sentinels mass data streams in support of city and regional planning. SEN4RUS is based on GEOURBAN outcomes, therefore using the expertise and basic techniques developed in the context of GEOURBAN, SEN4RUS will design and implement EO-based services for planners and decision makers that are specifically tailored to the Russian requirements. A key instrument in this context is the further development of a Web-based Information System (WIS), capable of evaluating the EO-derived indicators and provides them in a form that allows easy access and direct implementation into planning procedures. Three Russian cities with different typologies and planning perspectives have been included as case studies: St. Petersburg, Omsk and Vladivostok. To engage the users in the project, a Community of Practice approach is employed. The innovation of SEN4RUS lies in the development of robust techniques for information extraction and derivation of geo-information products from Sentinel satellite imagery in combination with an improved WIS that is adapted to and optimized for the Russian urban and regional planning system and can be easily understood and controlled by non-experts. Adaptation of the SEN4RUS WIS to forthcoming missions have also been planned, therefore a fully operational tool is expected in the future.

Land-cover maps at large scale provide important information for bettering our understanding of the relations between human activities and global change. The maps can further be used to provide critical information for climate change studies. Although urban areas only represent a relatively small fraction of the Earth’s land surface, the detection of urban area extent and the derived estimation of population and monitoring of population migration are pre-requisites for accurately assessing the impact of human activities on the environment.

Here we propose an automatic algorithm that aims to map buildings based on intensity backscattering recorded with Synthetic Aperture Radars (SAR) and Interferometric SAR (InSAR) coherence time series. The two main hypotheses that pave the way for the proposed algorithm are: built-up areas in SAR images appear very bright and they are coherent in time. The reason for the first one is that in the presence of buildings an effect commonly termed as ‘double bounce’ leads to a very high backscattering; while the second hypothesis can be formulated giving that manmade structures are hard target and InSAR coherence is mostly constant in time. As a matter of fact the built-up areas show a distinctive behaviour in terms of backscattering and coherence values, which can be used to differentiate them from all other land cover classes.

To identify pixels of high backscattering values, representative of buildings, we make use of a statistically based algorithm, which requires the parameterization of distribution function of backscatter values attributed to building class. To parameterize accurately such distribution function, a sufficiently high number of pixels is needed. However, this criterion is not always satisfied, because the building class usually occupies only a small fraction of the scene. To this aim, we propose an algorithm that makes use of a Hierarchical Split-Based Approach (HSBA) to calibrate the buildings distribution. Subsequently, based on the class distribution, the building map is generated by applying a sequence of region growing and histogram thresholding processes. Finally, the InSAR temporal coherence is used to remove from the classification those pixels that show an high backscattering but low InSAR coherence, as certain type of vegetation.

The algorithm has been developed in the framework of the Urban Round-Robin exercise, supported by the European Space Agency (ESA) through the ESA Land Cover Climate Change Initiative (CCI). It was tested on Sentinel-1 data from five different test sites located in semiarid and arid regions in the Mediterranean region and Northern Africa providing building maps at 20m resolution. The algorithm is efficient in terms of processing time, since it only needs as input a limited number of images and also in terms of scalability to different sites, due to its capability to automatically adjust its parameters for all test cases.

A comparison with the Global Urban Footprint (GUF) from TerraSAR-X data has been done. The GUF product has been under sampled to 20m and the comparison has shown an overall agreement between the two products in the range of 92% and 98% for the five tested areas.

Africa is projected to triple its urban population by 2050, making it one of the most rapidly urbanising regions on the planet. The agile urbanization leads to many challenges linked to urban population distribution monitoring and urban health. The African megacities can be incubators for new epidemics allowing infectious diseases to spread in a more rapid manner and to become worldwide threats. Adequate population monitoring and estimation are essential tools to improve our understanding of infectious disease dynamics within cities and target interventions.

For tackling these challenges, the Belgian Science Policy (BELSPO) is funding two projects addressing population distribution and its link to infectious diseases. The Modelling and forecasting African Urban Population Patterns for vulnerability and health assessments (MAUPP) project, uses advanced big-data processing scheme to map the dynamic morphological areas of 48 African cities from 1995 till present at 12.5m resolution. The robust and fully automatic data scheme extracts training information from OpenStreetMap (OSM) and applies machine learning to map the multi-temporal built-up densities, for improving the prediction of spatially-disaggregated population densities. The project advances as well the state-of-the-art urban population models by integrating land-use maps obtained using VHR optical and SAR data. The multi-temporal data also allows the development of urban growth and population distribution forecasts.

The Remote Sensing for Epidemiology in African CiTies (REACT) project uses population prediction methods developed in MAUPP for spatial epidemiology purposes, and more specifically the risk of malaria within cities. The project aims to better understand the various factors responsible for focal malaria transmission in urban African settings using earth observation and spatial modelling. The methodology is developed at inter- and intra- urban levels in Sub-Saharan Africa by studying the demographic, socioeconomic, climatic and environmental conditions associated with the spatial variation of the incidence of malaria.

The lecture will present the multi-sensor and multi-scale spatio-temporal approaches and advanced processing techniques implemented in the two projects to extract various input-variables from earth observation. It will discuss the benefits and the limitations of ‘big-data continental approach’ versus the ‘local approach’ and debate about the importance of fusing the two approaches for obtaining accurate population and epidemiological predictions for African cities. 

The revisit time and high spatial resolution of Sentinel-2, combined with thermal infrared satellite images and the increased availability of urban geometric databases, brings new opportunities for mapping urban radiative budget, which is essential for studying urban processes (climate, heat island,…). The anisotropic behavior of urban reflectance ρ(Ω_v) and brightness temperature T_B(Ω_v), where Ω_v is the viewing direction, is a challenge for extrapolating the satellite signal from its unique viewing direction Ω_sat to the upper hemisphere 2π⁺. Remote sensing (RS) models that simulate satellite ρ(Ω_v) and T_B(Ω_v) can achieve this angular integration. In order to be accurate, they must consider the urban 3D architecture, which explains that they need to be physically based. DART (Discrete Anisotropic Radiative Transfer; www.cesbio.ups-tlse.fr/dart) is one of the most comprehensive physically based 3D models of Earth-atmosphere optical radiative transfer, from ultraviolet to thermal infrared. It simulates the optical 3D radiative budget and signal of proximal, aerial and satellite imaging spectrometers and LiDAR scanners, for any urban and natural landscape and any experimental / instrumental configuration. DART was used in the H2020 project URBANFLUXES (http://urbanfluxes.eu/) dedicated to the determination of anthropogenic heat fluxes in three cities (Heraklion, Basel, London) with the help of satellite images. The two terms of the urban radiative budget Q^* (i.e., short Q_sw^* / long Q_lw^* wave radiative budget) were calculated by DART using satellite derived data. The short wave approach was realized with Landsat and Sentinel 2 images. It includes three major steps. (1) An iterative satellite based inversion method gives a map of material reflectance ρ_i,j (x,y) per type of urban element i (e.g., roof, street, vegetation), per spectral band j at the satellite image spatial resolution. At each iteration, the comparison of satellite and DART images gives new estimates of the reflectance of urban materials. (2) DART computes the urban exitance M_j(x,y) and albedo A_j(x,y) maps. (3) These maps are spectrally integrated to get urban exitance M(x,y) and albedo M(x,y) maps. Because time series of Q_sw^* maps are needed, Q_sw^* (x,y,t) maps must be computed in the absence of satellite images. DART can achieve it using material reflectance derived from the closer satellite image, and direct sun E_sun,dir(t) and atmosphere E_sun,dif(t) irradiance. To reduce DART simulation time of entire cities, Q^*(x,y,t) is computed as the sum of black and white sky albedo maps weighted by E_sun,dir(t) and E_sun,dif(t). Black sky albedo maps are pre-computed for a number of sun directions that sample the period of interest. The accuracy of Q^*(x,y,t) maps was assessed by comparing Q^* simulated by DART and measured at urban flux towers. Results are very encouraging: RMSE ≈ 15 W/m²; i.e., 2.7% mean relative difference. As expected, the improved spatial resolution of Sentinel 2 gives better results than Landsat. The URBANRADS project is aimed to make operational and transferable the developed inversion methodology for estimating Q^* with Sentinels.

According to UN 2018 report the urbanization is going to increase from 55% in 2018 to 68% by 2050 and impacting the global climate. With outward urban growth and increasing temperature within the cities is leading to the emergence of Urban Heat Island (UHI) and the factors responsible are urban geometry, urban morphology, change in Land Use and Land Cover (LULC) and the anthropogenic heat emission. To study the urban climatic condition, Stewart & Oke (2012) defined the Local Climate Zone (LCZ) classification. The LCZ classification differentiate itself into 17 types of classes of which 10 are urban built classes and 7 are land cover classes. Study area considered for the research is Chandigarh (India) due to its high rate  of outward growth and development of satellite towns (Zirakpur, Mohali and Panchkula) around it, which has significant impact on the planned city. For performing LCZ classification, the method proposed by World Urban Database and Access Portal Tool (WUDAPT) is adopted. The LCZ classification was performed using two widely available open source Satellite data sets Landsat-8 and Sentinel-2.   Landsat-8 is the default dataset for WUDAPT. Both the datasets were downloaded and processed. The training areas for classification were collected from Google Earth following the approach of WUDAPT. It was noted that most of the land cover classes especially agriculture, bare soil and bare rock areas has very high seasonal variability and selection of date for collecting training areas is of significant importance. Collection of training areas from wrong season image leads to very high inaccuracies in overall classification. The results obtained from Landsat 8 RS data provides higher accuracy of 81.04% than Sentinel-2 data which has accuracy of 70.56% for LCZ classification. It is to be noted that although the Sentinel-2 data has higher spatial resolution(10 m) as compared to Landsat -8 (30 m) , It has provide lower LCZ accuracy which may have been due to absence of thermal bands in Sentinal-2 dataset.

Key Words: Local Climate Zone, World Urban Data Base and Access portal tool, Landsat-8, Sentinel-2

Urban classification is important to monitor urban sprawl, soils impermeabilization and to predict their further evolution . Supervised classification approaches using satellite imagery have been widely studied to automate the process of land cover (LC) classification. However, new sensors such as Sentinel-2 and Spot-6/7 with interesting spatial, spectral and temporal characteristics bring new opportunities. Indeed, VHR sensors, e.g. Spot-6/7, enable the delineation of small features and the use of texture, but often have not enough spectral information to distinguish fine LC types. Conversely, sensors such as Sentinel-2 have more spectral bands and an important revisit frequency (time series) but a less geometric resolution. Fusion of both sources would combine their advantages to reduce spatial and semantic uncertainties.

Thus, this study investigates the joint use of Spot-6/7 and Sentinel-2 data for the detection of urban areas, defined as the aggregation of buildings, roads and enclosed small structure. A late fusion scheme is adopted, considering original data have been classified earlier and independently by specific methods. This workflow consists of 3 steps.

The Spot-6/7 image and Sentinel-2 time series are classified independently for a simple 5-class legend (“buidlings”, “roads”, “forests”, “other vegetation and crops”, “water”), with training samples extracted out of French national land cover and topographic databases. The Sentinel-2 time serie is classified using a Random Forest (RF) classifier , so as to have a framework similar to the French “OSO” LC maps, which are intended to be used in the future. Spot-6/7 mono-date image is classified using a deep Convolutional Neural Network (CNN) , because of its high ability to efficiently exploit context and texture information from VHR image. Both classifiers produce membership probabilities for the 5 classes for each pixel.

Both 5-class classifications are merged at decision level, aiming especially at the best extraction of building objects.

Detected buildings are then considered as the seeds of urban areas and used to define a prior measure for being in urban area. This measure is then merged with a binary urban/non-urban Sentinel-2 classification derived from its original 5-class one within a second fusion at decision level.

Both fusions (for the 5-class and the binary classifications) are performed following a same two-step scheme involving a per- pixel decision fusion followed by a spatial regularization to cope both with semantic and spatial uncertainties. The per-pixel decision fusion relies on state-of-the-art fusion rules : fuzzy, Bayesian, evidential or supervised (i.e. involving learning methods). The spatial regularization involves an image contrast constraint so as to have results fitting to strong image contours.

The fusion improves results. Similar results are obtained for several rules. For the 5-class classification, slightly better ones are reached for supervised fusion. Such supervised methods can not be used for the binary classification fusion, because no reference training data are available. For the same reason, obtained urban areas maps can only be assessed visually.

There has been substantial progress in the characterization of both urban built ups on the one hand and the derivation of 3D building models on the other hand. For characterization of urban areas at neighbourhood level there are various approaches with urban structural types (UST), however, they lack consistency and thus comparability. Thus we argue that there is a certain scale (and semantic) gap to map cities at intermediate scale. This is particularly evident for the purposes of urban climate science, which requires consistent descriptions of form and function of cities to represent their climatic impact on larger scales and evaluate climate mitigation and adaption options. In this field, the typology of Local Climate Zones (LCZ) (Stewart & Oke 2012) was proposed as generic description of urban areas at km scale and quickly became a standard. Bechtel and Daneke (2012) demonstrated that LCZ could be derived using multi-spectral Landsat data and standard classifiers such as Random Forest and Neural Networks. Thus the methodology was adopted by the World Urban Database and Access Portal Tool (WUDAPT) (Bechtel et al. 2015), an international crowd-based initiative, which aims to derive climate relevant data on cities worldwide. More recently, other scientific fields started being interested in the concept and LCZ mapping raised attention in the wider image analysis community, most visible by the data fusion contest of IEEE GRSS (Yokoya et al. 2018).

While the database of classified cities is growing, the spatial coverage is limited to about 100 cities worldwide and a few regional maps. Larger-scale efforts are mostly inhibited by the variable spectral appearance of the LCZ due to different building materials and codes as well as biophysical characteristics related to the climatic background. Moreover, data handling and processing become more demanding at larger scales. More recently, some efforts have been undertaken to upscale this framework to larger regions, enabled by the use of cloud computing resources of the Google Earth Engine. As a first step, the availability of a larger number of input features (such as Sentinel 1, Sentinel 2, and DMSP-OLS night-time lights) allowed to analyse the transferability of training. These results then further allow to up-scale the LCZ framework to countries and continents.

References

Bechtel B, et al. (2015) Mapping Local Climate Zones for a Worldwide Database of the Form and Function of Cities. ISPRS Int J Geo-Inf 4:199–219

Bechtel B, Daneke C (2012) Classification of Local Climate Zones based on multiple Earth Observation data. IEEE J Sel Top Appl Earth Obs Remote Sens 5:1191–1202

Stewart ID, Oke TR (2012) Local Climate Zones for Urban Temperature Studies. Bull Am Meteorol Soc 93:1879–1900

Yokoya N, Ghamisi P, Xia J, Sukhanov S, Heremans R, Tankoyeu I, Bechtel B, Le Saux B, Moser G, Tuia D (2018) Open data for global multimodal land use classification: Outcome of the 2017 IEEE GRSS Data Fusion Contest. IEEE J Sel Top Appl Earth Obs Remote Sens 11:1–15

With the emergence of spaceborne imaging spectroscopy missions, novel opportunities for describing urban composition globally arise. To move beyond case studies for individual cities, transferable models for repeated urban mapping across space and time are desired. Here, we present an approach for quantifying vegetation-impervious-soil (VIS) fractions of multiple cities with a single, generalized regression model. We used the combination of support vector regression (SVR) and synthetic mixtures generated from spectral libraries for fraction mapping. This library-based approach is independent from the spatial context of the work with training areas and appears better suited for more generalized mapping assessments. Three spectral libraries from Berlin (Germany), Brussels (Belgium), and Santa Barbara (U.S.) were used to develop one generalized model from a combined multi-site library, and three local models from the separate single-site libraries. In a first assessment, all model variants were applied to simulated Environmental Mapping and Analysis Program (EnMAP) data for these cities. Results indicated that the VIS composition of all three cities was accurately mapped by the single generalized model (average MAEs of 8.5, 12.2and, 11.0%). No loss in performance was found compared to the local models. In a second assessment, all model variants were transferred to the San Francisco Bay Area (U.S.) and Munich (Germany), i.e., two unknown cities not represented in the models. Results indicated that vegetation and impervious fractions were accurately mapped by the generalized model (MAEs within a range of 5 to 15%). In contrast, all local models were only useful either on one of the two cities or for individual VIS classes. For all models, deficiencies remained for accurate soil mapping. Combining SVR with synthetically mixed training data from multi-site libraries constitutes a flexible modeling framework for generalized urban mapping, e.g., across cities from the same biomes. Future work will focus on developing generic urban spectral libraries and investigating model transferability across data from different sensors.

An Earth Observation based set of smart tools is developed in support of smart urbanization and in view of quality urbanization. The tools reflect a methodology for the detection of hot and cold spots in the urban web onn the basis of a downscaling technique for land surface temperature and likehood probability assessment and z-scoring, the extraction of urban characteristics, such as the aspect ratio, with the potential to influence ventilation in the city and to this end the dispersion or trapping of heat and air pollutants, the multi-criteria manual definition of local climate zones, the corresponding spatial distribution of thermal comfort and the assessment of the impact of greenery through the estimation of the surface park cooling intensity. Results refer to the city of Athens and demonstrate the need for differentiated urban mitigation and adaptation measures at sub urban scale, the delicate balance between the thermal environment and urban characteristics as well as the potential of the developed tools to improve urban governance. For the purposes of the study, satellite data sets in the visible and thermal infrared are used, mostly from Sentinel 2 and 3, along with VHR ground data.

Urban areas are dominated by man-made structures like buildings, streets and impervious surfaces. In the last years there is a growing demand of urban footprint mapping: is fundamental to monitor the rapid and often uncontrolled growth of cities as well as the population shift from rural to urban area. Industrialized countries suffer from “sprawl” defined as the spreading of urban developments (such as houses and shopping centers) on undeveloped land near a city. All these phenomena cause “soil sealing” and a consequent loss of ecosystem services. City planners need efficient methods for management of large urban areas andcertainly remote sensing data helps providing area-wide and up to date land cover information. The Italian Institute for Environmental Protection and Research ISPRA is part of a network known as National System for Environmental Protection (SNPA), which is made up of 21 Regional Environmental Protection Agencies (ARPA / APPA). According to the law n.136/2016, ISPRA, yearly, publishes the report of the land consumption in which it analyses the evolution of land cover; considering the territorial transformations at different levels, this report provides new assessments of the impact on lost or threatened natural functions, landscape fragmentation and the economic costs of artificial soil growth.

Within the Copernicus program of ESA, ISPRA makes use of multitemporal Sentinel-2A MSI and Sentinel-1A SAR data as first input to produce the maps of different land cover classes and therefore of land consumption among consecutive years. Historically, for urban mapping application and for land cover classification, optical multispectral sensors have been used; nowadays thanks to the frequent revisit time of S-1 SAR data, the two kinds of data are adopted. Built-up areas, like buildings, are typically characterized by high backscattering values and high values of the InSAR coherence: urban areas do not significantly change within short time and therefore they can be easily recognized in the multitemporal images stacks. In fact, using proper coherence and backscatter threshold values, after masking out the vegetation contribution with the max NDVI within a long timeseries, it is possible to identify urban areas. The classification obtained is therefore validated through comparison with the land consumption map provided by SNPA based on photointerpretation of high resolution orthophoto.

Convolutional neural networks (CNN) have recently become the new state-of-art solution for several computer vision problems, such as image classification, object detection, semantic segmentation, etc. [1]. CNN can learn hierarchical image representations from the input data by successively abstracting higher level features from the previous low-level ones. In the last years, these type of neural networks have been demonstrated excellent performance on different remote sensing problems, such as pixel-based classification for land use [2], target recognition like buildings, roads, ships, aircrafts and vehicles [3], image pansharpening [4]. In this study we seek to exploit the potential of CNN to address the problem of identifying artificial changes altering the natural soil using Sentinel-2 data. The continuous growing of buildings, roads and infrastructures has raised the attention toward this issue and the knowledge of detailed changes across time is highly desired since it can be useful to understand the urbanization process and make proper urban planning proposals. In the paper, we use a pre-trained CNN on ImageNet to extract features from Sentinel-2 images pairs. The proposed approach consists of three steps: 1) For each image of the input pair we apply the feature extraction with the pre-trained CNN. 2) We apply the euclidean distance between the two multidimensional matrices produced by the feature extraction. 3) Finally, we obtain the change classification results through KMeans algorithm. It will be shown how the results obtained with the method improve those yielded by other state-of-art approaches [1] R. Girshick, J. Donahue, T. Darrell, and J. Malik, “Region-based convolutional networks for accurate object detection and segmentation,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 38, no. 1, pp. 142–158, Jan 2016. [2] Wenzhi Zhao and Shihong Du, “Learning multiscale and deep representations for classifying remotely sensed imagery,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 113, no. Supplement C, pp. 155 – 165, 2016 [3] Volodymyr Mnih and Geoffrey E. Hinton, Learning to Detect Roads in High-Resolution Aerial Images, pp. 210–223, Springer Berlin Heidelberg, Berlin, Heidelberg, 2010. [4] Wei Huang, Liang Xiao, Hongyi Liu, Zhihui Wei, and Songze Tang, “A new pan-sharpening method with deep neural networks, IEEE Geoscience and Remote Sensing Letters, Vol. 12, NO. 5, 2015

The growth of settlements and the associated increase of the exploitation of natural resources is an ongoing trend in the Arctic. Buildings and other infrastructure are endangered from destabilization and collapsing due to the climate change induced thawing of permafrost in northern regions. The majority of human activity in the Arctic is located near permafrost coasts. Coastal settlements are additionally vulnerable because of coastal erosion, caused by rapid warming and thawing of coastal permafrost.

The European Union (EU) Horizon2020 project “Nunataryuk” aims to assess the impacts of thawing land, coast and subsea permafrost on the climate and on local communities in the Arctic. One task of the project is to determine the impacts of permafrost thaw on coastal Arctic infrastructures and to provide appropriate adaptation and mitigation strategies. For that purpose, a circumpolar account of infrastructure is needed.

The two polar-orbiting Sentinel-2 satellites of the Copernicus program of the EU are continuously providing multi-spectral images with high spatial and temporal resolution. Sentinel-2 data is of great value for mapping land cover. However, most traditional land cover classifications only contain one class for built-up areas. By using a multi-sensor approach, such as the combination of multispectral and Synthetic Aperture Radar (SAR) data, additional information can be derived that goes beyond the identification of built-up areas. Different types of infrastructure can be distinguished, as it its commonly needed.

We assess the potential of combining Sentinel-2 multispectral data with SAR data for mapping and characterizing Arctic infrastructure. A first evaluation is carried out for two test sites: Tuktoyaktuk, located on the Beaufort coast in the Inuvik Region of the Northwest Territories in Canada and Longyearbyen on Svalbard, Norway. First results show that L-band from ALOS PALSAR 2 data are particularly useful, but limitations are encountered due to the coarser spatial resolution when compared to Sentinel-2. In addition, the benefits of using higher resolution C-band SAR data are discussed.

The High Resolution Settlement Layer (HRSL) provides estimates of human population distribution at a resolution of 1 arc-second (approximately 30m) for the year 2015. The population estimates are based on recent census data and high-resolution (0.5m) satellite imagery from DigitalGlobe. The population grids provide detailed delineation of settlements in both urban and rural areas, which is useful for many research areas—from disaster response and humanitarian planning to the development of communications infrastructure. Computer vision techniques are used to identify the presence of buildings in the mosaic of imagery; the resulting Boolean surface is used to represent settlement extents. Population is allocated from census administrative units to settlement extents using proportional allocation. An overview of the methodology will be presented, including an assessment of accuracy of the classification algorithm, along with a description of recent modeling efforts to improve detail in the population distribution through the use of covariates.  

The number of satellites having on board SAR instruments has significantly increased. In addition, the spatial resolution of the currently provided products is improved reaching the metric dimension for the TerraSAR-X and the COSMO-SkyMed missions. At the same time, the Sentinel-1 Copernicus SAR data, will be free distributed for many years and with a few days revisit time. Moreover, new SAR configurations, such as Argentinian SAOCOM, are planned to be launched in the very next future. This new situation opens new challenges and new fields of research in in Earth Observation and, in particular, in SAR image processing. Indeed, one of the main issues is that such a wealth of information might be difficult to be managed by the final users who therefore might not take fully advantage of the operations carried out by the space segments. In this context the importance of data fusion algorithms is steadily increasing. Various data fusion techniques have been already presented in literature by researchers. For example a review regarding data fusion among multispectral and hyperspectral data has been recently published [1] while in [2] also fusion techniques between SAR and optical images or including lidar and vectorial data are considered. However, even if multi-band SAR analysis of specific targets has been often performed, the design of specific pixel-based data fusion algorithms for SAR images acquired at different bands has been so far scarcely addressed. Neural networks (NN) have been already used in the past as an effective approach to pixel based data fusion. Their competiveness for SAR and optical fusion has been shown in [3] while in [4] they have been exploited for dimensionality reduction of hyperspectral data. In this paper, an Auto-Associative NN for the generation of X-band COSMO-SkyMed and C-band Sentinel-1 fused products is described. Indeed, with their capability of positively handling nonlinear relationships among data, NN seem a suitable technique for the management of the different scattering mechanisms stemming from the multi-band acquisitions. In the paper one on side the fused products are compared with the original ones by means of the accuracy obtained on the corresponding classified urban maps. On the other side, a Principal Component Analysis (PCA) technique is considered as a benchmark algorithm for data fusion.

[1] N. Yokoya, C. Grohnfeldt, J. Chanussot, “Hyperspectral and Multispectral Data Fusion: A comparative review of the recent literature,” IEEE Geoscience and Remote Sensing Magazine, Vol. 5, No. 2, 2017
[2] J. Zhang, “Multi-source remote sensing data fusion: status and trends,” International Journal of Image and Data Fusion, 1:1, 5-24, 2010
Environment, v. 183, pp. 26-42, 2016
[3] F. Pacifici, F. Del Frate, W. J. Emery, P. Gamba, J. Chanussot, “Urban mapping using coarse SAR and optical data: outcome of the 2007 GRS-S data fusion contest,” IEEE Geoscience and Remote Sensing Letters, vol 5, n. 3, pp. 331-335, July 2008
[4] G. Licciardi and F. Del Frate, “Pixel unmixing in hyperspectral data by means of neural networks,” IEEE Transactions on Geoscience and Remote Sensing, vol. 49, n. 11, pp. 4163-4172, November 2011

The Copernicus satellite missions Sentinel-1 and Sentinel-2 offer an important tool to routinely monitor the urban development by exploiting SAR and multispectral acquisitions. Indeed, the high revisit time of both satellites allows to recursively update the change detection map over urban landscapes. Satellite remote sensing can be usefully used to provide an objective and consistent view of urban areas [1]. Anyway, to efficiently exploit the increasing amount of data, like those acquired by the Sentinels missions, novel automatic procedures are needed. Since the effects of the electromagnetic interaction with human made structures largely differ moving from optical passive to microwave active acquisitions, the fusion of these different information may result in more accurate results in mapping urban changes, contributing to reduce the number of false alarms and the miss-detections with respect to the case on which only one data type is considered.

In this study a new approach based on the fusion of Sentinel-1 and Sentinel-2 products to detect urban changes and to observe suburb’s development is presented. The fusion algorithm is based on two Pulse Coupled Neural Network [2] blocks, applied to Sentinel-1 and Sentinel-2 images respectively, and on a spectral difference computation block, applied only to multispectral images. Pulse Coupled Neural Networks have already proven their capabilities in addressing change detection problems with SAR [3] or optical data [4] and here have been considered to develop the data fusion algorithm for the change detection procedure.

The latter processes the data in a fast, automatic and accurate way and can be also applied to multitemporal acquisitions. The three outputs of each computational block are fused together by means of a weighted average formulation that compute the fused result at each time step, which is represented by an acquisition date. When running in multitemporal mode, the algorithm identifies the presence of the detected objects at different time steps and, applying a majority criterion (e.g. an object is detected three times out of five temporal steps, or five times out of eight temporal steps), produces a reduction of false alarms on the final results. Such false alarms may be due for instance to the variation of natural surfaces optical reflectivity or backscattering mechanisms not related to urbanization processes.

The proposed procedure has been tested on Sentinels data acquired over different European cities like, Basel, London and Rome, obtaining encouraging results.

[1] J.-F.Mas, Monitoring land-cover changes: a comparison of change detection techniques. In: Remote sensing, vol. 20, no. 1, 139-152. 1999.

[2] Zhaobin Wang, "Review of pulse-coupled neural networks", 2009.

[3] Pratola C, Del Frate F, Schiavon G, Solimini D. Toward fully automatic detection of changes in suburban areas from VHR SAR images by combining multiple neural-network models. IEEE Transactions on Geoscience and Remote Sensing. 2013 Apr;51(4):2055-66.

[4] Pacifici F, Del Frate F. Automatic change detection in very high resolution images with pulse-coupled neural networks. IEEE Geoscience and Remote Sensing Letters. 2010 Jan;7(1):58-62.

Sealed surfaces in urban areas have been shown to have differing thermal behaviors relative to their degree of permeability, in addition to factors such as albedo and water availability. However, they are not typically included in remote sensing-based studies of surface heat fluxes, partially due to their heterogeneity and detection difficulty. Considering that these surface types should have differing thermal behaviors over time, the present study examines the utility of a thermal time series approach. Urban block data from 2015 with percentages of four surface sealing-type classes (extreme, high, medium and low, represented cover types such as asphalt, cobblestones, etc.) were obtained for the city of Berlin, Germany. This was assessed against a Landsat thermal time series of 13 images over a period of two years around 2015. The initial tests were performed on a reduced dataset, with building density less than 20% per block unit (leaving 6426 blocks). K-means clustering was used to divide the mix of sealing types into 10 classes. Mean temperatures were derived for each block, and the mean total urban surface temperature was subtracted from each block time series to reduce the trend effect. Then, Spearman's rank correlation was then used to determine the relationship between each block unit and all other blocks, which were then averaged for each k-means cluster. The Mann-Whitney-Wilcoxon Test was used to determine if a significant difference was present within and between clusters. The initial findings show that 7 out of 10 k-mean clusters have significantly different time series temperature patterns. These are mostly classes which are dominated by extreme, high and low sealing, as well as a class where no type is above 50% coverage. These results suggest that surface sealing types, which are typically only represented as asphalt or concrete in remote sensing classifications, can be better characterized through the use of thermal time series data.

Public urban spaces such as streets, open spaces, green areas, parks, and public buildings are a big part of cities that are often overlooked. Inadequate, poorly designed, or privatized public spaces often generate exclusion and marginalization and degrades the livability of the urban environment. That is why the importance of green areas and open spaces are now embedded within the Sustainable Development Goals, particularly Goal 11.7 on universal access to safe, inclusive and accessible, green and public spaces.

This presentation showcases how the use of geospatial information derived from very high resolution (VHR) satellite imagery can contribute into development of a framework to examine how urban spaces can transform urban environments by promoting inclusive green growth and enhancing livability in megacities – starting with Karachi (Pakistan) and Dhaka (Bangladesh), which are pilot cities that the World Bank Urbanscapes Group has initiated an Advisory Services and Analytics (ASA) activity to support ongoing investment operations on the ground.

Overall cooperation under this ASA aims to:

(i)              Develop an enhanced diagnostic on the nature of public spaces, and the opportunities and challenges;

(ii)            Provide a concrete body of evidence on public spaces-related policies and programs, for the Bank operations to assist cities with strategic advice and inputs to public spaces;

(iii)           Gain a better understanding of the state and problems of urban public spaces, focusing on selected cities, and to identify future investments and implementation strategy;

(iv)           Provide a platform for related knowledge exchange and policy dialogue with practitioners, academics, and clients.

As part of this program, the UrbanScapes Group multidisciplinary team cooperated with GISAT - remote sensing information company on developing a series of analytical work focused on characterizing land use and identification of public municipal assets (e.g. public spaces) using high-resolution satellite imagery, and to better define intervention areas during project preparation. The presentation will provide an overview on a spatial analytical work done in Phase 1 and how it was applied to two Bank operations during the preparation phase as well as the potential for streamlining the approach to future Bank project preparation with the UrbanScapes ASA and beyond.

Earth Observation data potential exploration part is carried out by cooperation in the frame of the EO4SD-Urban project supported by the European Spatial Agency.  EO4SD (Earth Observation for Sustainable Development) is a new ESA initiative, built on previous successful ESA-WBG collaborations, aims to achieve a step increase in the uptake of satellite-based environmental information in the IFIs regional and global programs. It follows a systematic, user-driven approach in order to meet longer-term, strategic geospatial information needs in the individual developing countries, as well as international and regional development organizations (http://eo4sd.esa.int).

The goal of this paper is to propose an innovative approach for Satellite Image Time Series (SITS) classification considering the data series as a hyperspectral image. The data volume of a SITS describing a ground region could be equally large or even greater than the volume of data in a single hyperspectral scene. Additionally, the difference in spectral information between two adjacent wavelength bands in a hyperspectral scene is usually very small, their grayscale images appearing almost identical. Most of the data in a scene therefore would seem to be redundant, but embedded in it is critical information that often can be used to delineate the ground surface materials. This case is similar with a high temporal resolution SITS, where spectral characteristics of the features composing a scene may slightly differ between consecutive scenes.  Due to the high spectral resolution, the hyperspectral imagery offers the capability to distinguish for example between several ground materials while the multispectral imagery is not able to make the distinction. One can assume, correspondingly, that the end members  (i.e. spectra chosen to represent pure surface materials in a hyperspectral image) would represent in SITS the center of the semantic clusters describing the scene informational content. Starting with this assumption we propose the use of an end members selection algorithm to set these “pure pixels” as initial centers of the classes, followed by a classification algorithm in order to derive a classified time series. Such a result would clearly highlight the regions of the scene that have the same evolution in time.

In the end we validate this approach using a SITS based on the NDBI (Normalized Difference Build Index) spectral indices computed for Landsat scenes acquired between 1992-2011 centered on Bucharest, the capital of Romania.

The classification result clear highlights several types of  evolutions of urban areas in this region, our next step being to validate this approach on Sentinel 2  SITS.

The THEIA Land Data Services Centre is a French national inter-agency organization designed to foster the use of Earth Observation images for documenting changes on land surfaces. THEIA is offering to the science community and to public policy operators a broad range of services, from the access to raw and corrected images at different resolutions, to the development of methods and services. THEIA is further proposing distributed processing infrastructures to operate the methods and deliver an ensemble of products and mutualized services. THEIA is further organized in Competence Centres in order to provide scientific and technical innovations on several aspects of Land surfaces monitoring. One of this competence centre is the "Urbanization and Artificialization Centre / Urban Expert Centre" clustering experts in multi-sensor urban remote sensing.

The objective of this poster is to present recent (>2016) innovations of the Urban Expert Centre in terms of 1) development of algorithms useful for urban remote sensing, 2) validation of the urban products provided by the THEIA Space Data Infrastructure, and 3) demonstration of user-tailored applications from any type of optical sensors. The Urban Expert Centre brings together researchers and engineers from several institutes (UMR LIVE – Strasbourg, IGN-LaSTIG - Univ. Paris Est, CESBIO – Toulouse, LETG-Rennes, Irstea – Montpellier, UMR TETIS - Montpellier, INP Bordeaux, IRD, UMR ESPACE-DEV - Montpellier, UMR ESPACE - Nice, ONERA - Toulouse.

Research results and methods linked to 1) the detection and mapping of urban areas at national scales (URBA-OPT), 2) the identification of urban fabrics, 3) the monitoring of green areas and corridors within the cities will be presented. The results are associated to the processing of high-frequency optical images of the Sentinel Constellation and of SPOT 6&7 and Pléiades data.

Temporally stable night light imaged from space provides a globally consistent record of lighted development and nocturnal human activity since 1992.  In addition to urban development, changes in night light also reveal expansion of infrastructure, large scale resource extraction and consequences of conflict.  The defense meteorological satellite program (DMSP) operational linescan system (OLS) has imaged emitted light from Earth’s surface since the 1970s. Since 1992, a digital archive of DMSP-OLS annual composites has been provided by the NOAA National Geophysical Data Center (NGDC). Following the launch of the NASA/NOAA Suomi satellite in 2012, NGDC also began producing monthly composites of stable night light from the Visible Infrared Imaging Suite (VIIRS) Day Night Band.  We combine both of these products to characterize and map interannual and seasonal night light dynamics of the region of Southwest Asia extending from Turkey southward through Yemen and eastward through Iran.  To quantify decadal interannual changes in night light, we use OLS+VIIRS image fusion to produce sharpened night light change composites showing changes in night light since 1992 from OLS at VIIRS improved ~500 m spatial resolution.   For the later interval from 2014 through 2017 we use Empirical Orthogonal Function (EOF) analysis of VIIRS monthly composites to characterize and map the diversity of temporal trajectories of night light.  The EOF analysis clearly distinguishes annual cycles from progressive interannual changes and serves as a pixel-scale anomaly detector.  The most conspicuous changes are related to urban growth and expansion of lighted infrastructure, but increases are also related to expansion of electrification in some areas.  For example, widespread increases in night light brightness are observed throughout Turkey post-2002 resulting from urbanization as well as electrical infrastructure development.  The most concentrated expansion of lighted settlements and infrastructure occurs on the Arabian Peninsula and Mesopotamia.  Throughout the Gulf Cooperation Countries (GCC) there was a major increase in lighted highways since 2002. In the same region, we observe both seasonal and interannual changes related to oil production and gas flaring –particularly in Iraq.    We also observe a variety of changes related to conflict in Iraq and Syria, generally resulting in abrupt decreases and occasionally in gradual increases.  Overall, the most consistent and widespread increases, over a range of settlement sizes, have been in Turkey and Saudi Arabia.  We present a range of location-specific examples, along with vicarious validation of land cover changes as observed in Landsat and Sentinel 2 imagery.  This spatiotemporal characterization + vicarious validation provides a conceptually simple and computationally trivial, yet comprehensive, approach to large area detection of temporal patterns of night light change.

Throughout the years Brunei is a popular place for mining pit areas, having gas-oil resources and also due to ancient history of civilizations lived in. In last years, coastal cities in the country have also increasing tourism factor besides rapid urban growth. In the most of the cities of the country the population is getting higher year by year. It brings also land cover and land use changes with acceleration in that places. Remote sensing technologies are widely used all around the world for environmental monitoring in different scales. Urban growth, changes in rural areas, deforestation, clear-cut areas are some of the topics can be count under the regional monitoring targets. Brunei-Muara is chosen as a study area because of the city's old important history and significant changes in the region. Land use land cover change is monitored with using image processing techniques and different classification methods. Sentinel-2 satellite images are used as data of the study and land use classes areas are calculated. It's showed that especially urban areas are changed significantly in the last 5 years in the city. In the end of the study land use and land cover maps are produced and compared in some aspects scientifically.

Intensive settlement activities create necessity of their control and documentation.

Activities related to collection and aggregation of data for buildings registers and topographic mapping used in many administrative processes on the national, regional and local levels are often insufficiently coordinated and out- of- date. This causes inconsistency, incompleteness and misuse of public registers.

Intensive and frequently uncoordinated settlement processes leading to excessive dispersion and sprawling of the urban areas have serious, adverse influence on transportation and/or media supply costs. New built-up areas and connected infrastructure extension causes fragmentation of ecological corridors and continuous air and soil pollution.

This project aims at designing, implementation and operationalization of a dedicated web service platform providing at demand and almost continuously the reliable information on newly detected changes of the settlement network. This information is supposed to be derived automatically from Sentinel-2 (S-2) satellite imagery using special-purpose image classification algorithms.

Regional authorities responsible, inter alia, for the processes of maintaining and updating of topographic data-bases have been identified as the main end-users of the proposed service.

The service will be at first tested and implemented for the territory of the Mazovian Province in Poland. Hence, this first implementation is called Built-up Areas Monitoring Service - Mazovia (BAMS - Mazovia). However, the platform’s overall architecture and its flexibility will allow for extension of the provided services for other provinces and regions.