The size of urban settlements is increasing worldwide at a rapid expansion rate. This urban sprawl triggers changes in landcover with the consumption of natural areas, and has impacts on the ecosystems with important ecological, climate, and social transformations. Detecting, mapping and monitoring the growth and spread of urban areas is therefore important for many applications, such as urban planning and management, risk analysis, human health or biodiversity.
Earth observation images have long been used to document human settlements. The availability of the Sentinel constellation (S1 and S2) allows monitoring urban sprawl over large areas (e.g. countries) and at high time frequency (with possible monthly updates). The massive stream of data available with these sensors allows proposing new types of urban products at spatial resolutions of 20 m. The method, named URBA-OPT, is object-based, fully automated, uses open source libraries, and optimized for rapid calculation on HPC clusters.
Taking advantage of recent advances in object-oriented image analysis (OOA) and machine learning algorithms, a supervised method is proposed in order to objectify the choice of significant object features and classification thresholds.
The method associates a sequence of image segmentation, spectral and textural feature selection, automated definition of a training set from he High-Resolution Layers Imperviousness, and object classification using a Random Forest classifier. The outputs are binary maps of urban and non-urban areas and a probability of class certainty. The method is implemented for the processing of massive streams of Sentinel 2 images on HPC clusters. Use cases over several French metropolitan areas and over the complete French territory will be presented. Experiments linked to the choice of the initial training set and the machine learning classifiers will also be presented. With an iterative procedure to examine the class-imbalance within the training set, compensation of the effects of class-imbalance and class-overlap on the balance of the error rates is possible. Employing approximately 20% of the data for training, the method results in accuracies between 90% and 95% for the use cases and approximately balanced commission and omission errors close of 0.4 taking the Global Human Settlement Layer (GHSL) as reference.

Multi-temporal interferometric synthetic aperture radar (MT-InSAR) is used for many applications in earth observation with high precision. Most MT-InSAR methods select scatterers with high coherence throughout the entire time series and this technique works well in an urban area as these scatterers are related to the man-made infrastructures. However, as time series lengthen, changes in surface scattering lead to decorrelation, which significantly decreases the number of coherent scatterers. Here we propose a novel method to detect and process temporary coherent scatterers (TCS) by subsequently analyzing the amplitude and the interferometric phase. Assuming a Rayleigh distribution, two hypothesis tests are introduced to develop a single pixel change detection process, which is used to select TCS candidates and identify their corresponding step-times: the moments of appearing and/or disappearing coherent scatterers. Based on the amplitude analysis, phase analysis is conducted using a multi-level network solution. Furthermore, the parameters of interest, e.g. line-of-sight (LOS) displacements, are estimated using the interferometric phase and the initial change detection result is improved in this step. As a result, the coherent intervals of TCS are determined and this approach allows us to distinguish three possible changes, i.e., appearing TCS, disappearing TCS and visiting TCS. The spatial density of TCS is calculated to highlight the changes over the region. Then a deformation velocity map, a step-time map and a point density map are obtained, which show the deformation, step-times and locations of the changes, respectively. The main advantage of this method is the reliable extraction of a subset of time series without using any contextual information and change detection result is obtained both in the spatial and temporal domain. Experimental results by the proposed method show that the TCS significantly increases the number of observations for displacement monitoring and improve the change detection capability in urban construction areas. Both disappearing and appearing objects with their step-times are successfully recognized in Delft, the Netherlands. Additionally, change detection results by ascending and descending TerraSAR-X images are in good agreement and complement each other.

Urban agglomerations play a key role in giving shelter to the growing world population. Mapping the global extent of urban settlements, their land cover composition and growth patterns, as well as interconnectedness contributes to an understanding of city efficiency, resilience and transformation, e.g. in the context of the UN sustainable development goals. Satellite remote sensing data is highly suited for human settlement mapping because of its large area coverage and high potential for standardization. Multiple satellite-based products provide maps of human settlement extent, growth or land use (with varying assumptions and definitions) on different scales and spatial resolutions, e.g. the Global Urban Footprint (GUF), the Global Human Settlement Layer (GHSL), the World Settlement Footprint (WSF) or the EEA Urban Atlas. Still, urban areas are a general challenge to remote sensing because of the heterogeneity of urban features and surfaces. Many products map built-up areas as binary masks without considering the gradual spatial patterns of land cover (LC) composition in human settlements and its important vegetation component.

Sentinel-2 offers free and globally available data with an unprecedented spatial, temporal and spectral resolution, thus representing a highly interesting source for settlement mapping and monitoring at relevant spatio-temporal scales. We use Sentinel-2 A/B Analysis Ready Data for large area fraction mapping of urban LC with machine learning regression. Our approach employs library-based, synthetically mixed training spectra and support vector regression models to map surface fractions of four classes, i.e. impervious surfaces, woody vegetation, non-woody vegetation and soil. Our results show that regression modelling is able to provide accurate fractions of urban LC with a spatial resolution of up to 10m.

We find that using spectral temporal metrics helps to improve the quantitative separation between spectrally similar but temporally variable cover types. Accounting for the temporal dimension improves the robustness and quality of regression models applied over large areas or distant from the reference spectra locations. We systematically study the effect of several spectral temporal metrics as well as temporal statistics of commonly used spectral indices. We find that an aggregation period for summarizing temporal information of one year or more reduces confusion between impervious surfaces and soils as well as uncultivated agricultural land in central Europe.

This study shows that surface fraction mapping with regression models can add valuable information to existing binary or discrete products and that thematic information is extended by including maps of the urban vegetation component. We systematically map different urban settlement types across Europe, and show that the applied method is also able to capture the character of rural settlements.

Remote sensed imperviousness products share a large interest within the scientific community thanks to their importance in a high number of applications such as Urban Heat Island modelling, socioeconomic modelling (population, GDP, etc) and soil sealing monitoring [1].
Currently, Copernicus delivers a 20 x 20 m imperviousness density layer on a yearly basis, derived only from optical images (SPOT 5, Resourcesat-2, Landsat 8) using a regression model of calibrated Normalized Difference Vegetation Index (NDVI) values [2].
In this work, a very simple, innovative and quick methodology to derive such information, based only on the 20 x 5 m Sentinel-1 IW-TOPS SAR night and day, weather independent observations, is presented.
As in the first case, urban features classification depends on the assumption that NDVI value is very low, here we assume that urban features behave in the SAR scene as dihedral scatterers.
According to dihedral scattering theory, effective dihedrals can be only 0 or π valued in the cross-POL coherence, allowing to easily segment the scene in a binary map of urban features [3]. These two assumptions admit a similar degree of robustness with respect to the classification of impervious elements as defined in [2] and are here discussed.
Thus, the 20 x 5m resolution binary map is converted in a 20 x 20 m density product downsampling the grid in the range direction from 5 to 20 m, achieving the capability of describing five degrees of imperviousness, from 0% to 100%.
Furthermore, the six days revisit time of Sentinel-1 constellation produces an abundance of observations, thus allowing a robust estimate of the cross-POL coherence, via multi-temporal averaging, as well as a very high temporal resolution.
The accuracy of the proposed methodology has been assessed over Milan, using as reference a raster derived from the municipality building footprint vector layer, released by Lombardy Region.
Results show a significant robustness of the product derived following the proposed methodology, thus pointing out that the optical and radar-derived products might be successfully integrated in order to obtain better results in overall accuracy.

[1] Xiaojun Yang, Zhi Liu, “Use of satellite-derived landscape imperviousness index to characterize urban spatial growth”, Computers, Environment and Urban Systems, Volume 29, Issue 5, 2005, Pages 524-540

[2] Tobias Langanke, Copernicus Land Monitoring Service – High Resolution Layer Imperviousness: Product Specifications Document, 2016

[3] D. K. Atwood and L. Thirion-Lefevre, "Polarimetric Phase and Implications for Urban Classification," in IEEE Transactions on Geoscience and Remote Sensing, vol. 56, no. 3, pp. 1278-1289, March 2018.

In this paper, an innovative Tomographic Synthetic Aperture Radar (TomoSAR) approach for the 5D reconstruction of the illuminated scenes is presented. The added value of the TomoSAR approach, compared to the well-known SAR Interferometry (InSAR) is the possibility of separating of the scatterers interfering within the same range-azimuth resolution cell. TomoSAR is able not only to provide the scatterers position in the 3D space, but also the estimation of the deformation velocity (4D reconstruction) and the estimation of the thermal dilation (5D reconstruction).
The proposed 5D tomographic processing is based on the use of an extended phase model having a strong impact on the size of the unknown reflectivity vector, which becomes very large; for this reason, the practical application of a statistical detection approach to 5D SAR tomography fully benefits from the approximation introduced by the Fast-Sup-GLRT detector. Anyway, an important issue that has to be investigated to state its applicability to 5D SAR tomography is the evaluation of the probability of detection achievable with a given probability of false alarm and of the estimation accuracy for height, deformation velocity and thermal dilation, in the case of the extended phase model.
Performance analysis is carried out on StripMap TSX data of an area of Barcelona (Spain) where, due to the high resolution, the layover effect is present. Presented results show the effectiveness of the proposed approach in detecting reliable single and double scatterers in urban areas, useful for the measurement of surface deformations of manmade structures. Moreover, it is shown that taking into account velocity deformation and thermal dilation helps in recovering more scatterers, both single and double, and also in detecting the top of tall buildings. The two contributions, linear temporal deformation and thermal dilation, are separated and estimated. A physical interpretation of the detected thermal dilation phenomenon, in accordance with the geometry and material of the structure considered, is given. In addition, it is shown that the separation of double scatterers allows for detecting a larger number of reliable scatterers and for obtaining more accurate estimates’ height, deformation velocity and thermal dilation coefficient in urban structures monitoring.

The growing urbanization raises issues such as air quality deterioration or rising temperatures. Many actions are currently undertaken to understand and improve the comfort and quality of life in cities as well as to protect citizen health. To evaluate the relevance and performance of the implemented or planned actions, their effectiveness must be quantified. To that purpose, Earth Observation (EO) has a strong potential to support more effective urban planning. Many studies dealing with the urban heat island phenomenon strongly rely on the estimation of land surface temperature (LST) derived from thermal infrared (TIR) space borne sensors. Upcoming high-resolution TIR satellite missions such as HyspIRI from NASA, TRISHNA joint-mission between CNES (Centre National d’Etudes Spatiales) and ISRO (International Space Research Organization), or the Copernicus High Spatio-Temporal Resolution Land Surface Temperature Mission, will provide new opportunities to evaluate LST changes in urban areas. With minimum 3 TIR bands, those sensors should allow to retrieve LST at 60-m spatial resolution every 5 to 3 days over cities, improving the urban environment monitoring capacities. LST can be retrieved from TIR data using two methods, Split-Window (SW) or TES (Temperature Emissivity Separation). Those widely applied methods provide good results over homogeneous surfaces but their performance remains limited over urban areas due to 3D shapes and the large variety of manmade and natural materials. The main challenge with SW lies in the characterization of the large diversity of materials to provide the suitable emissivity values required by this method. Another major issue to consider when applying SW or TES over urban areas is the impact of the surface roughness generated by the 3D structure of the city. With the aim of quantifying the uncertainty inherent to each method in the perspective of using them to retrieve LST over urban areas from the future TIR data, works were conducted to evaluate their respective error budget and to investigate the improvement brought by combining those data with ancillary data such as land use maps, surface digital models or model simulations outputs. The benefits of providing a representative material database, derived from a classification map, when using SW were explored and validated based on airborne or satellite data. The impact of the 3D geometry on TES estimations was also evaluated using 3D thermo-radiative model simulations. The results showed that calibrating the SW algorithm with an urban-oriented database leads to an intrinsic RMSE of 1.06 K, 0.94 K for the validation dataset and a mean difference of 2 K for daytime and 0.7 K for nighttime between airborne measurements and satellite ones. They showed as well that neglecting the impact of the surface geometry while performing TES could lead to errors up 4°C in LST estimations. This research, investigating the main issues faced when estimating LST in urban areas and exploring solutions to tackle those issues considering upcoming mission characteristics, paves the way to future research integrating EO, modelling and ground data.