Sentinel 1 methods

Urbanization and its impact on the environment is without doubt one of the major global challenges. Satellite-based Earth Observation is the only reasonable approach in order to detect and delineate urban area extents on the global scale. Due to their independence to daylight and weather conditions, SAR data have become not only a complementary source of information, but also has proved their usability as a stand alone source for accurate built-up area extractions. Since its launch in April 2014, the Sentinel 1A SAR sensor acquired thousands of images in different modes. Due to the intended global coverage at high spatial resolution Sentinel 1A data hold a tremendous potential for accurate urban area detection and characterization at the planetary scale.

This work deals with the adaption and optimization of an already existing and intensively validated urban area extraction algorithm (Urban EXTractor) for the application on Sentinel 1A data. The UEXT algorithm has been originally developed for ENVISAT ASAR Wide Swath Mode data at a spatial resolution of 75 meters in the framework of the ESA Climate Change Initiative – Land Cover (CCI-LC). The main idea of the UEXT method is based on the assumption that urban areas do not significantly change within short time frames and therefore they can be easily be recognized in multi-temporal images stacks due to the fact that – in contrast to built-up areas - the surrounding areas usually change their appearance throughout annual phenological cycles. The same multi-temporal filtering, averaging and equalization procedure has been applied on Sentinel 1A data at a spatial posting of approximately 20 meters. Starting from the multi-temporal image stack, the algorithm searches for very bright pixels which, in SAR imagery, usually correspond to artificial structures and are consecutively used as starting points for a region growing procedure which is iterated until a specific threshold is reached. Due to the increased spatial resolution of the Sentinel 1A data the parameters of the UEXT algorithm had to be adapted and additional object-based post-processing steps have been applied in order to refine and improve the results.

The optimized UEXT 2.0 method has been so far tested on multiple test sites in arid and semi-arid region, because these were the areas where the previous version of the approach, based on ASAR WSM data, provided the poorest results. Advantages and drawbacks of UEXT 2.0 are presented in detail in this work. Furthermore, the potential of cross-polarized VH backscatter is evaluated in the context of global human settlement characterization. To this aim, in all test areas (located in Portugal, Tunisia, Turkey and Israel) an objective reference data set was generated through random selection and consecutive manual labeling of discrete global grid hexagons. The quality of the urban extent extractions have been assessed visually and quantitatively. Results show that the UEXT method can be successfully applied on Sentinel 1A data and that this new generation SAR sensor bears a huge potential for urban area extractions at the local, continental and global scale.

Starting from the beginning of the years 2000, more than half of the global human population is living in urban environments and the dynamic trend of urbanization is growing at an unprecedented speed. Rapid urban growth brings several challenges, including meeting accelerated demand for basic services, infrastructure, and affordable housing (particularly for the nearly 1 billion people living in informal settlements). Moreover, as cities develop, their exposure to climate and disaster risk increases (e.g., almost half a billion urban residents live in coastal areas, thus increasing their vulnerability to storm surges and sea level rise). In this framework, an effective monitoring of urban sprawl represents a key issue to analyze and understand the complexity of urban environments and ensure a sustainable development of urban and peri-urban areas.

To this purpose, the ESA DUE Innovator III SAR4Urban project aims at implementing - in support of its users the World Bank and GEO Global Urban Observation and Information Task for Societal Benefits (GEO SB-04) - a novel service that allows to automatically and reliably derive maps of past and current extent of urban areas by means of archived ERS/ASAR and novel Sentinel-1 data, respectively.

The basic assumption of the intended approach is that given a series of multi-temporal images for a given study area, the temporal dynamics of urban settlements are sensibly different than those of all other non-urban classes. As an example, the backscattering temporal mean of urban areas (due to double bounce reflection) is higher than that of forest areas (which might result in high backscattering in one/few acquisitions due to specific conditions, but in general exhibit lower values).

After applying orbit correction, calibration, and terrain correction to the multi-temporal images available over a region of interest in the selected time interval, for each pixel we extract key temporal statistics (i.e., backscattering temporal mean, standard deviation, minimum, maximum, etc.). It is worth noting that for different pixels in the study area, different number of scenes might be available. However - in the hypothesis of a sufficient minimum number of acquisitions for computing consistent statistics - this does not represent an issue. Indeed, we always expect a more stable behavior of the urban class compared to the others (for which the temporal variability is higher). Heterogeneity features are also extracted to ease the detection of lower-density settlements and, finally, specific unsupervised classification schemes are applied to ERS/ASAR and Sentinel-1 data, respectively.

Output of SAR4Urban will include the 2002-2003 urban extent map of entire Africa derived from ASAR WSM data, as well as the urban extent maps of Athens, Beijing, Los Angeles, Mexico City, Atlanta and the Pearl River Delta derived from ERS-1/2 PRI and ASAR IMP scenes. Moreover, the current built-up extent of both these and several African cities will be delineated by means of Sentinel-1A imagery. Preliminary results are extremely promising and confirm the great potential of ESA SAR data for mapping urbanization over time.

Human activities on the Earth's surface are rapidly altering our environment. Ongoing urbanization is a global phenomenon and one of our world’s most pressing challenges. For the monitoring of human-induced changes and to better understand global environmental changes, consistent and reliable global urban area maps are urgently needed. This study was conducted in the framework of the Land Cover CCI project as part of the ESA Climate Change Initiative (CCI) and focuses on the development of an automated processing chain for the updating / improving of the urban class of the CCI-LC global land cover product (resolution ~ 300m) using multitemporal Sentinel-1 SAR data with 20m spatial resolution. The test sites are located in semiarid and arid regions in the Mediterranean and Northern Africa, namely Portugal, Turkey, Israel, Egypt and Tunisia.

As demonstrated by several studies SAR data reveals a high potential for urban area mapping. Due to numerous occurrences of double bounce effects, urban areas are characterized by high radar backscatter values in radar images. Multitemporal backscatter statistics and texture measures are particularly valuable tools for the detection of built-up areas and will be used in this study. The proposed methodology takes the UADP (urban area detection parameter) texture measure and the Sentinel-1 multitemporal mean values at VV- and VH polarization as input. The UADP measure calculates the mean radar backscattering difference for each pixel and its surrounding pixels under a neighborhood constraint. The proposed update algorithm consists of two main processing steps: first, an initial urban area mask is produced applying an unsupervised ISODATA classifier. Based on this initial map the update process is performed in the second processing step, considering further parameters such as object size, neighbourhood functions etc. The algorithm was developed for the Tunisia data set. To demonstrate the transferability of the proposed methodology the processing chain was applied to the other four semiarid / arid test sites mentioned above without any adaptions.

In the end, the produced map updates of the urban class of the CCI-LC land cover product will be validated. This validation process includes (a) a qualitative validation based on visual comparison with high resolution imagery, (b), a statistical product validation using reference hexagons and (c) a product intercomparison with other existing land cover products.

Although the use of  satellite data for land cover/land use monitoring is one of the most addressed topics in scientific remote sensing literature, the implementation of processing chains for the production of regularly updated land cover maps at decametric spatial resolution and at national scale is still a rare case. In fact, on one side, the cost and the not always systematic availability of the data may represent a limiting factor, on the other side, the algorithms developed by the scientific community may not have the necessary robustness for being applied nationwide, which involves a lot of manual corrections and, in turn, unbearable costs. In this context, the launch of the Sentinel 1 and 2, coupled with the free distribution of the data, can yield significant improvements. In particular, the SAR images provided by Sentinel 1 guarantee data availability with a high revisit time even in case of cloudy sky, which can be very important especially in northern regions.

In this paper we present the preliminary results obtained by the implementation of a processing scheme using satellite images to provide, and regularly update every six months, land cover maps for the whole Italian territory. Indeed although in Italy some regional institutions have developed local GIS including land cover maps, a consistent nationwide product , at least to our knowledge, is still not existent, especially if an update every six months is considered.

Aiming at keeping high both the level of automation and the final accuracy the scheme consists of four steps. In the first step the whole Italian territory is divided in a certain number of tiles. In the second step for each tile a pixel based classification is performed using a multi-layer perceptron neural network (MLP-NN) algorithm [1]. By mosaicing all the classified tiles we obtain what we call the “Master” land cover map. In the third step the update of each classified tile is addressed using new satellite data. To this purpose a change detection algorithm based on Pulse Coupled NN (PCNN) is considered [2]. PCNN is a relatively new technique based on the implementation of the mechanisms underlying the visual cortex of small mammals [3]. Only the changed pixels detected by the PCNN are reclassified with the MLP-NN and the whole updated land cover map is obtained. In the fourth step an accuracy evaluation of the final product is carried out.

In our study the “Master” land cover map has been produced using Landsat acquisitions while for the updated versions of the map either Landsat or Sentinel 1 images have been considered. To assure enough robustness and accuracy a restricted number of land cover classes has been considered so far: forest, built areas, water, other natural surfaces. The final results are encouraging: first of all a consistent land cover map of the whole Italian territory with a spatial resolution of 30 m has been produced with an overall accuracy of about 92%. Moreover the PCNN procedure allows us to update the maps using a very high level of automation and keeping the same final accuracy.

 [1] Del Frate, F., F. Pacifici, G. Schiavon, C. Solimini, “Use of neural networks for automatic classification from high resolution imagery,” IEEE Transactions on Geoscience and Remote Sensing, vol. 45, n. 4, pp. 800-809, April 2007.

[2] F. Pacifici and F. Del Frate, “Automatic Change Detection in Very High Resolution Images with Pulse-Coupled Neural Networks,” IEEE Geoscience and Remote Sensing Letters, vol 7, n. 1, pp. 58-62, January 2010

[3] R. Eckhorn, H. J. Reitboeck, M. Arndt, and P. Dicke, “Feature linking via synchronization among distributed assemblies: Simulations of results from cat visual cortex,” Neural Comput., vol. 2, no. 3, pp. 293–307, 1990.

Urban extent and land cover have been mapped using a range of datasets and algorithms. With the recent launch of Sentinel-1A, SAR data with global coverage, operational reliability and quick data delivery became freely available, thus provide excellent opportunity for developing SAR-based methods for global urban mapping.  The objective of this research is to evaluate Sentinel-1 SAR data for quick and reliable urban extent extractions in selected cities around the world using the KTH-Pavia Urban Extractor, developed in collaboration between KTH and University of Pavia.  This study is part of the EO4Urban project funded by the ESA DUE INNOVATOR III program.

Multitemporal Sentinel-1A SAR data over Bejing, China, Milan, Italy, Stockholm, Sweden and Lagos, Nigeria were acquired for this research. The methodology is based on the original approach developed by Gamba et al. (2011) using both spatial indices and texture measures.  The overview of the methodology in this research is illustrated in Figure 1 with the improvements highlighted in light and dark green. The improvements mainly involve preprocessing, contrast enhancement, post-processing as well as decision level fusion using multitemporal and multipolarization data.  The original method shown in blue is based on “Local Indicators of Spatial Association” (L.I.S.A.), including the Moran index, the Geary index and the Getis-Ord index and GLCM variance and correlation textures. The detailed methodology can be found in Ban et al. (2015) and Gamba et al. (2011). 

The preliminary urban extraction results showed that urban areas and small towns could be well extracted using a single-date Sentinel-1A SAR data with the KTH-Pavia urban extractor. The urban extraction results are further improved using multitemporal dual orbit Sentinel-1A SAR data. Rigorous accuracy assessments are being performed and will be reported at the workshop.

REFERENCES

Y. Ban, and A. Jacob & P. Gamba, 2015. Spaceborne SAR Data for Global Urban Mapping at 30m Resolution Using a Robust Urban Extractor.  ISPRS J. of Photogrammetry & Remote Sensing.

P. Gamba, M. Aldrighi, M. Stasolla, “Robust extraction of urban area extents in HR and VHR SAR images”, IEEE Journal on Selected Topics in Applied Earth Observation and Remote Sensing, Vol. 4, no. 1, pp. 27-34, March 2011.