In image analysis, Change Detection refers to the capability to identify the location and magnitude of changes between a couple of images acquired at different times. Several analytical unsupervised methods have been proposed and used over time to assess changes occurring in images, ranging from simple image difference to MSE (Mean Squared Error) measures, however most of these fail in accurately identifying perceived changes at a human vision level.
NHAZCA S.r.l., in collaboration with Terradue S.r.l., developed and integrated in the ESA Charter Mapper a Change Detection unsupervised processor that can be used to monitor land changes and provide a fast disaster response exploiting satellite imagery. The processor was initially only available in IRIS - Change Detection and Displacement Analysis Software, a proprietary image analysis software developed and distributed by NHAZCA S.r.l. and is now freely available to authorized users on the Charter Mapper. The Change Detection method implemented makes use of the Structural Similarity Index Measure, an algorithm originally developed to assess perceived quality of digital television and cinematic pictures in which the measurement of image quality is based on an initial image taken as reference. The method here is used at a local scale, iteratively assessing the image similarity on a small subset of pixels in the image with a sliding window approach, allowing to identify the portions of the scene that underwent to alterations and to precisely define the edges of such changes.
The integration in the Charter Mapper was aimed at proving a simple to use processor even for users without any knowledge about the theory behind the algorithm, allowing anyone to run change detection analysis in a fast, unfettered manner thus providing a way to exploit EO data to provide a quick disaster response. The processor was tested using a variety of images before its integration and then different use cases were investigated in the production environment of the Charter Mapper. Here, the results of analyses carried out to evaluate the reduced traffic activity in Rome during the Covid-19 lockdown as well as to assess the effects of a flood event are presented and discussed, an overview of how the service can be used and its future developments and improvements are also described.
The Satellite-based Crisis and Spatial Information Services (SKD, Satellitengestützter Krisen- und Lagedienst), a unit of the German Federal Agency for Cartography and Geodesy (BKG, Bundesamt für Kartographie und Geodäsie) became operational on 01.01.2021 and has responded to more than 250 orders since then (as of 23.11.2021). The setup of SKD was funded by the Federal Ministry of the Interior, Building and Community (BMI, Bundesministerium des Innern, für Bau und Heimat). In this paper, we report how EO data are used by SKD for providing rapid mapping products in security and crisis situations, products of Copernicus Service in Support to EU External Action (Copernicus SEA) and to fulfil the demands of national users with respect to commercial and very high-resolution satellite images.
As one of its services, SKD provides information and individual products with additional specialised information as per requests from federal institutions. These are made using geospatial data provided by BKG and by analysing, and evaluating, very high-resolution remote sensing information. As a result, up-to-date information on challenging situations that require rapid information (natural hazards, humanitarian crises or conflicts) are provided. Having direct contact to the agencies involved in operational planning, strategy, disaster mitigation and investigation, the information provided by SKD can be implemented rapidly in their daily work. Along with providing EO based geospatial information in security and crisis situations, SKD also provides personal consulting services. This is especially important in rapid decision making for such events, where SKD provides competent and targeted advice for users on the possibilities and restrictions that currently exist. In order to ensure efficient follow-up of the products provided by the SKD, various training courses and information events tailored to the respective needs of the federal authorities will be offered.
Another service provided by the SKD is the Federal Service Point of Remote Sensing (Servicestelle Fernerkundung; SF). Based on our survey conducted in 2020, a need for a datahub of commercial EO data was observed and many German federal agencies expressed interest in using EO data in their daily activities. The Federal Service Point of Remote Sensing at SKD attempts to fulfil these needs and requirements of the federal government for using commercial EO data and products. It aims to collect, coordinate and provide a free and assessible path to commercial EO data and products, while providing consultation, capacity building and trainings for interested users in the federal government. More information about the SF is provided in an additional abstract by Mayr et al., 2022.
EO data is also actively used to provide Europe and German mosaics. SKD developed and established a process for producing high quality mosaics that allows for the harmonisation of any optical remote sensing data and for any area on Earth. This is particularly beneficial to the Federal Administration in Germany. For our first mosaic of Europe, we used Copernicus Sentinel-2 data from 2018. The Europe mosaic product consists of multiple radiometrically colour balanced Sentinel-2 images that are then assembled to create a single, seamless large area image. The product is produced as an 8-bit image with 3 bands and 10m resolution, with less than 3% cloud cover, coordinate system ETRS89-extended / LAEA Europe (EPSG: 3035) and is available also in the form of web map service (WMS). SKD will continue to expand the time series (past and current ones) of these mosaics and make them available as open data. The next Europe mosaic will be provided for the year 2021. We also provide a complete, almost cloudless and high-quality mosaic of Germany for the years 2018, 2019, 2020 and 2021 made using Sentinel-2 datasets and available as a WMS. These national mosaics have 5 bands, 10m pixel spacing and were requested on average 80,000 times a day by around 4,000 different users.
In January 2021, BKG was named as the national civilian point of contact (PoC) for the Copernicus SEA (Copernicus Service in Support to EU External Action) component and is operated by SKD. This means that SKD is responsible for the retrieval of products and services from Copernicus-SEA for all civil authorities in the Federal Republic of Germany. The services mediated include satellite images of regions outside the EU with additional geographic information or extensive geo-intelligence. SKD also carried out workshops on the services of Copernicus SEA for the federal agencies. Thus, Copernicus SEA complements the services of SKD for the users at the national level.
With the above listed services and activities, SKD aids in increasing disaster risk resilience and security in Germany, and Europe, by providing EO data, products such as mosaics or thematic maps and services to other German federal agencies and research institutions that are vital for their decision-making activities.
Acknowledgements
Special thanks to the other members of the SKD team comprising of Nikolai Adamović, Kristian Ćorković, Marian Graumann, Katja Happe, Tamara Janitschke, Franka Kunz, Matthias Meerz, Robert Oettler and Ulrike Rothe.
Radar backscatter is useful for observing volcanic activity, especially for remote or dangerous eruptions, as it is not limited by access to the volcano or cloud-coverage, but currently it is less widely used for volcano monitoring than radar phase measurements. This is in part because of ambiguity in the interpretation of backscatter signals: there is not always a simple link between the magnitude or signal of the backscatter and the physical properties of volcanic deposits. Here we present three case studies (Pu‘u‘ō‘ō, Kīlauea, Hawai’i, 2010 – 2013; Volcán de Fuego, Guatemala, 2018; and La Soufrière, St. Vincent, 2021) using a range of SAR sensors (CSK, TSX, Sentinel-1, and ALOS-2) to demonstrate how radar backscatter can be used to research and monitor a variety of volcanic eruptions, and especially to extract quantitative information.
Radar backscatter is dependent on the scattering properties of the ground surface (i.e., surface roughness, local gradient, and dielectric properties), each of which can vary during a volcanic eruption and provide information about specific deposits and processes. Pyroclastic density currents and lahars during the 2018 eruption of Volcán de Fuego and the emplacement of lava flows in Hawai’i during 2010 – 2013 were dominated by changes in surface roughness. We identify deposits and their variations based on their different morphologies, calculating the lengths of flows and areas affected by the eruptions. Where a deposit is emplaced over a period of multiple SAR acquisitions, we can map the progression and development of the deposit through time. While backscatter signals associated with eruptions in Hawai’i and Volcán de Fuego were dominated by changes to the surface roughness, backscatter changes during dome growth at St. Vincent were dominated by changes in the local surface slope. Our analysis at La Soufrière is therefore driven by this slope-dominated signal, which provided the opportunity to extract topographic profiles from the SAR backscatter.
We examine the use of various methods to reduce (1) noise (e.g., speckle filters and extended timeseries), (2) satellite geometry (e.g., radiometric terrain correction), and (3) constellation influences (e.g., principal component analysis) present in backscatter signals and to improve the identification of volcanic changes. The addition of supplementary datasets (e.g., high-resolution DEM, rainfall data, pre-eruption land cover) are important when performing detailed analyses of deposits.
We demonstrate through the three case studies the ways in which backscatter can be used to understand and monitor a range of volcanic eruption styles. We highlight a number of quantitative volcanic outcomes (e.g., flow lengths, deposit thicknesses, areas and volumes), a variety of SAR methods (e.g., change difference, extended timeseries, flow mapping, pixel offset tracking) and corrections (e.g., radiometric terrain correction, satellite dependency).
Global atmospheric warming and associated deglaciation effects lead to the increasing development of slope instabilities in glacier fore-field environments. The primary drivers are de-buttressing effects due to retreating glaciers, exposure of previously contained rock masses and thawing of permafrost. Such effects can lead to a decrease in slope stability and possible resulting failure in the generally rough and steep terrain encountered in high mountains, calling for extensive hazard analyses of such features.
As part of ESA’s research project Glacier Science in the Alps (AlpGlacier, https://eo4society.esa.int/projects/alpglacier), we apply advanced Differential Interferometric Synthetic Aperture Radar (DInSAR) techniques to detect and map slope instabilities in selected regions of the European Alps and assess associated geohazards. In particular, we apply differential interferometry including atmospheric and unwrapping corrections, a multi-temporal stacking analysis and finally persistent scatterer interferometry. The combined use of these methods enables the detection of a wide range of surface motion in terms of displacement rates and size.
We mainly use data from both ascending and descending orbits of ESA’s Sentinel-1 satellite constellation. This allows us to assess slope instabilities in the last six years with temporal baselines of 6 to 12 days at a relatively high spatial resolution as given by C-band. The complementary use of past SAR sensors (e.g., ERS-1/2, JERS-1, ENVISAT, ALOS-1 PALSAR-1) additionally enables a historic analysis, while current SAR sensors (e.g., TerraSAR-X, Cosmo-SkyMED, Radarsat-2, ALOS-2 PALSAR-2) and optical sensors (Sentinel-2) are used for validation purposes and an integration of higher resolution data or longer wavelengths to detect fast motion. Such a broad and systematic coverage enables the generation of displacement maps, revealing the spatial distribution of surface movement in glacier fore-fields. Mapping of slope instabilities is done manually through a comparison of all available data. In order to assess associated geohazards, the detected features are classified into movement type, based largely on geomorphological characteristics, and an activity state, which is mainly defined by the observed velocities.
We will discuss particular features of interest such as sliding complexes observed near Mer de Glace in France and the Findel Glacier in Switzerland. For such selected features, time series can be generated to highlight the temporal evolution and possible acceleration/deceleration patterns, which are important for geohazard analyses. In light of the obtained results, we will outline the practical considerations of the applied techniques in mountainous regions and discuss the advantages and disadvantages of each method. This includes limitations and uncertainties as well as using alternative methods and additional sensors to overcome some of the limitations.
Developing an understanding as to how activity changes for slowly moving landslide impact on assets has significant value to support the safe operation of pipelines across North America. As landslides in various regions typically move very slowly, infrastructure and assets are often designed to accommodate movements and require ongoing maintenance and mitigation. A recent paper by Porter et al. (2019) estimated the annual costs associated with management and mitigation of landslide hazards in the WCSB to be over $400 million with much of these costs associated with interventions needed to minimize damage caused by local increases in landslide velocity.
Asset owners and consultants manage the impacts of slow-moving landslides through planning and design, monitoring, and timely interventions. Over the past decade, many regions Western Canada specifically have experienced significant storm events and increases in moisture infiltration, associated with precipitation and snow melt, which appears to have caused an increase in landslide activity and movement rates from recent historical levels. It is not currently known if the changes in water infiltration are a result of climate cycles, or a response to global climate change. Irrespective of the cause, the observed increase in landslide activity and movement rates threatens to undermine the benefits being realized from our geohazard management programs, for which mitigations are currently designed based on current climate conditions.
In order to support better decisions around infrastructure planning and design, monitoring, and timing of interventions a study was initiated to support answering the following questions:
• Based on the current velocity of a landslide (or inventory of similar landslides), what are the average likelihoods of slower or faster velocities in the near future (weeks to months), based on observations of past landslide behaviour?
• How do different combinations of hydro-meteorological influences, such as changes in water infiltration associated with snow melt and precipitation, change the likelihood that landslides with similar characteristics within a region will increase or decrease in velocity?
Based on the answers to these questions a historical understanding of regional landslide activity and associated hydroclimatic drivers is required. As part of the overall warning system development, EO data and derived models play significant roles in process understanding and the ability to provide proactive warning. As understanding of the historical deformation trends are key to predicting the future, regional coverages of Sentinel-1 and ALOS-2 Stripmap data have been acquired and utilized to supplement ground measurements to build historical deformation time series plots for over 40 landslides over a region in Western Canada. This data has then been integrated with hydroclimatic data (ERA-5, SMAP-4) into visualization tools to support identification of broad trends and into machine learning models to develop clear relations to support tracking of critical trends to support warning.
The presentation will review how data on landslide activity has been integrated with satellite and ground hydroclimatic data (rainfall, soil moisture, snow melt) and used to support the development of operationally appropriate thresholds and response plans.
The cost of a disaster, both in terms of economic loss and fatalities, is dependent upon the rapidity and efficacy of the event response. Considering volcanic disasters specifically, the remoteness of the terrains combined with potentially incapacitated lifelines (e.g., disturbed transportation network) prevent ground-based surveys for timely assessment of damage extents. This is where emergency managers can most benefit from remote sensing tools.
To that effect, we have been working on using optical and Synthetic Aperture Radar (SAR) data to rapidly delineate the areas impacted by volcanic flows during an eruption (e.g., pyroclastic flows, lava flows, lahars), which can in turn be used to target and organize the response efforts. Multi-sensor analysis allows to alleviate the limitations from each sensor type and obtain imagery at various spatial resolutions. While optical data can provide direct observations of the areas covered by volcanic flows, they require cloud-free skies, which is often restricted during an event due to heavy ash clouds or rain. SAR data compensate for these limitations with all-weather and all-day imaging capabilities. Moreover, short data latency is a critical factor to enabling rapid access to volcanic flow extent maps, which is why combining multiple datasets from multiple sources may allow for better temporal coverage during an event.
In this research, we used the 2015 eruptions of Colima (Mexico) and Calbuco (Chile) volcanoes to calibrate detection thresholds of different types of volcanic flows, from optical and SAR imagery. Specifically, optical imagery was used to calculate Normalized Difference Vegetation Index changes (U+2206 NDVI) between pre- and post-eruption images, that were caused by the presence of erupted materials on the surface, and SAR amplitude images were used to detect changes in surface roughness (sigma0) attributed to the emplacement of new volcanic flows. Linear rescaling of minimal and maximal threshold signals were used to create probability maps of volcanic flow deposits extent, and then combined into a joint probability map to maximize the accuracy of the deposit extents. Finally, very-high-resolution imagery was used to validate the flow extent footprint, and the True-/False- Positive/Negative technique was used to evaluate the performance of our detection method.
In a second part of this project, we tested our capability to generate volcanic flow extent maps during recent volcanic disasters response work (i.e., the eruptions of La Soufrière St Vincent in April 2021, La Palma in September 2021, and Mount Semeru in December 2021), using this detection method and the calibrated threshold values. As the list of available sensors grows, we hope to continue improving the use of multi-sensor analysis to reduce data processing latency and therefore increase disaster response efficacy. Testing this methodology at different spatial and temporal-resolution can also provide pointers to what will be relevant in future spaceborne and airborne missions.