Between the Poles

Identifying areas of high risk vegetation encroachment on power lines from satellite imagery

Vegetation is a significant source of outages for many utilities. In some regions from one quarter to one half of all outages can be ascribed to vegetation. Vegetation encroachment initiated the massive North American Northeastern outage of 2003. With the growing incidence of extreme events managing vegetation near power lines is becoming increasingly crucial in tackling potential fire hazards, securing infrastructure increasing resilience. The traditional approach to identifying areas of high-risk vegetation encroachment involves helicopters and ground crews and is time consuming and costly. Drones have been used to reduce the cost of data collection, but it is still a costly process. Now an innovative startup is using satellite imagery together with deep learning to identify high risk areas of vegetation encroachment. By applying machine learning algorithms to high resolution satellite imagery, it is possible to distinguish vegetation from other objects, differentiate trees from grass and bushes and identify trees at risk of encroachment. Comparing a drone-based approach with the satellite imagery methodology reveals that the latter approach is highly accurate in identifying high risk areas and significantly less costly than a drone-based process.

Background

Power line inspections for vegetation management and other purposes are essential in ensuring grid reliability and resilience. They are generally performed by manned helicopters or by a ground crew. Data is collected with cameras and analyzed to detect diseased trees that could fall and hit a power line (fall ins) or trees that are encroaching on power lines (grow ins). These inspections are mandated by NERC in North America and are not optional. Beyond visual line of sight (BVLOS) drones have reduced the cost of collecting line inspection data. Now a new company is reducing the cost of identifying trees at risk by analyzing high resolution satellite imagery to identify high-risk areas, thus reducing the total length of power line that requires more detailed inspection with drones, helicopters, or ground crews.

Evolving machine learning models at Spacept

A Swedish company Spacept has developed a vegetation management solution that applies machine learning models to multi-band satellite imagery to automatically identify areas of high risk for vegetation encroachment on above-ground transmission and distribution lines anywhere on the planet. Spacept develops its own in-house training data for machine learning models which enables follow-up ground inspections of areas of high risk to develop pruning and other response programs. The training data is constantly enhanced with feedback from new projects for continuous improvement of the models.

Satellite-based vegetation encroachment risk assessment

The key business benefit of Spacept’s satellite imagery-based solution is automating the process for identifying the high-risk areas that require ground-based inspection, whether helicopter, drone-based, or ground crew, thus significantly lowering the cost of vegetation encroachment inspections. Spacept’s satellite-based vegetation encroachment technology makes it possible for power grid operators to monitor vegetation encroachment over their entire infrastructure in a cost-effective manner. Spacept uses the most recent, highest resolution (0.50 – 0.30 cm) satellite imagery commercially available which satisfies the requirements for vegetation encroachment detection. Spacept's approach makes it possible to monitor hundreds of thousands of kilometers of power lines every year anywhere in the world.

Workflow

The workflow of identifying and downloading satellite imagery, applying machine learning for feature identification, and the assignment of risk categories to individual trees is currently a semi-automated process. The longer-term objective is to reduce manual intervention and achieve an end-to-end automated workflow. The current workflow proceeds from a geographical layout of the power line to be inspected in the form of a standard GIS file. The most recent available high resolution satellite images satisfying certain criteria are identified and downloaded from data providers. The criteria typically include panchromatic and multi-spectral bands, resolution of at least 50 cm, vegetation capture with leaf on, and minimal cloud coverage. The downloaded images are fed to the current vegetation model which automatically extracts features and distinguishes trees from other man-made and natural objects. One of the important capabilities of the model is that it is very efficient in differentiating trees from bushes. After the automated processing, the imagery and labelling are reviewed by experienced assessors who review each image and make corrections to the labelling for problem areas, for example, for cases of visually distinguishable bushes or edges of trees. After completion of the labelling step, the distance between each tree and the power line is calculated and a risk category is assigned using risk categories provided by the client. The final deliverable is a GIS file containing the location and the corresponding risk categories assigned to the trees.

In-field case study

To assess the potential business benefits of satellite-based analysis Spacept partnered with infrastructure health management platform Sterblue to apply their satellite-based power line vegetation management solution for thousands of kilometres of a French power grid operator grid. Spacept applied their algorithms to selected satellite imagery to generate a risk file for the targeted segment of the grid. Sterblue assigned emergency (risk) levels based on the distance from the power line using a classification of emergency levels provided by the grid operator. Sterblue then carried out in detail inspections of at-risk areas with its drone inspection fleet. A ground inspection team was sent to verify the locations of the trees physically and collect ground truth. A total of 283 locations were identified. The vegetation at each location was manually validated and was compared with the satellite predictions.

The most important conclusion on the business benefit of the satellite imagery-based analysis is that it was able to identify all the trees that the subsequent ground inspection identified as at risk. In a real-world scenario, applying the satellite imagery analysis as a first step in a power line inspection would dramatically reduce the area and trees that would require ground-based inspection using helicopter or drone-based inspection. As a result of this process, Sterblue reduced its operating costs by 80%.

A more detailed analysis of the satellite data attempted to classify trees into detailed emergency levels U0 to U3. Comparing this analysis with Sterblue's classification was used to create a "confusion matrix" showing instances of agreement and disagreement between the two analyses. It was found that there was good agreement between the satellite and ground-based inspections and analyses.

Conclusion

Several companies have recognized that earth observation imagery is of limited value without the application of analytics to interpret the data. For example, Spacept is applying analytics to the huge amount of earth observation data to provide a valuable high risk vegetation encroachment detection service that is safer, faster, and less costly than ground-based approaches to transmission and distribution line vegetation monitoring.

Real time ground deformation monitoring from satellites to reduce risk of damage to utilities and pipelines

Several years ago I blogged previously about an application that used Sentinel-1 radar interferometry to monitor ground displacements with millimetre precision to look for subsidence which might be evidence of a leak in a municipal water system. Now access to data and processing algorithms are available to developers to create workflows to monitor ground deformation to reduce the risk of damage to underground Infrastructure including utilities, oil, gas and water transmission pipelines and associated industrial plants.

Ground movement represents a significant risk to linear structures such as utility, pipeline, and transportation networks as well as to large scale industrial and commercial structures.  Traditionally ground deformation is monitored by fixed high accuracy GNSS stations located at intervals and by periodic surveys along alignments.  Both of these are expensive and are limited by geospatial and temporal gaps. Satellites such as Sentinel-1 that use radar to image the earth are capable of radar interferometry (INSAR) which allows ground deformations on the order of cm and even mm to be detected.  The revisit times for these satellites are on the order of days and the radar antennas on these satellites allow very large areas to be monitored for deformations.  For example, Sentinel-1 is capable of detecting deformations of a few mm over an area of 42,500 km² with a 12 day revisit time. Furthermore INSAR data is available stretching back to the mid 1990's making it possible to monitor historical ground movement over quite long time periods.

UP42 provides the infrastructure of an enterprise software development and deployment platform enabling developers to create scale-able workflows using satellite imagery and customized algorithms. In a recent UP42 webinar (53 minutes in), Shawn Melamed and Sarah Thompson of CATALYST (PCI Geomatics) showed how innovative INSAR technology has made it possible to cost effectively continuously monitor ground deformation along thousands of km of  a linear structure such as a utility, pipeline, or transportation network.  The data can be sourced from any of the satellites capable of INSAR including Sentinel-1 (free) and TerraSAR-X (commercial).  An example is provided in the talk of the development of a UP42 workflow for detecting uneven land subsidence in California resulting from agricultural withdrawals of water from an aquifer that potentially could damage a power plant located in the area.  The code to perform this analysis is available as the Ground Displacement Block on the UP42 marketplace.

 

 

A 3D base map for the globe

At the GeoIgnite virtual event Ryan Hamilton, Senior Manager 3D Products, at Maxar described a remarkable project Globe in 3D to create a 3D model for the earth's entire landmass using the petabytes of 50 cm imagery that Digital Globe (and now Maxar) has collected and is collecting from its constellation of satellites.  Multiview stereoscopy technology enables Maxar to create a 3D model for every point (pixel) on earth by combining multiple views of the same point taken at different times and with different camera locations.  The result is a textured 3D model that has better than 3 meter absolute positional accuracy, avoids the blind spots that result from a single camera view, and averages out errors that are often found in individual images.

In 2015 Digital Globe and Saab formed a partnership which created Vricon to integrate 3D technology with Digital Globe's imagery to enable the creation of an accurately georeferenced 3D model of the earth's landmass without using control points. (Maxar completed the acquisition of Vricon last year.) The process is compute intensive and is still underway. The result will be a 3D TIN model with textures on all sides for the entire landmass of the planet. The Globe in 3D is available in Vricon format, Esri i3S or Cesium 3D Tiles.

A number of derivative products are also available for specific use cases including digital terrain and digital surface models, textured buildings, classification layers identifying different land uses and vegetation, point clouds, 2D ortho images, 3D vector representations of buildings, vegetation and transportation infrastructure with coverage in dense urban, suburban and rural areas. A limitation resulting from the algorithm that combines images captured at different times is that it is not possible to assign a specific date to  the 3D model at any location. But Maxar does provide tools to enable change detection over longer periods of time.

The 3D model can be used for many use cases including estimating cut and fill for a road, railway or other linear structure, estimating tree height for vegetation management, planning 5G telecom networks and others.  But perhaps it's biggest achievement could be to provide a shared 3D base map for the entire planet. To assist in realizing this vision Maxar provides precision 3D registration (P3DR) tools to allow other data to be corrected (conflated) to the Maxar base map. P3DR is a stand-alone software solution that automatically georegisters imagery from any source to Maxar's Globe in 3D base map.

Making Landsat data free and open generated billions of dollars of economic benefits

Open data has been gathering momentum among governments. First enunciated in the United States, the principle that citizens should not be required to pay twice for government data, free and open access to government geospatial data has been adopted by many governments including US Federal, Canada, Japan, South Africa, Brazil, California state and counties, and by the City of Vancouver and other cities.  In 2013 G8 leaders (Canada, France, Germany, Italy, Japan, Russia, United Kingdom, United States, and the European Union) signed a charter on open data.  The Ordnance Survey in the UK began releasing some of its data with open licensing as part of its OpenData initiative in 2010, but it wasn't until 2018 that the OS Mastermap was made openly available for the public and businesses to use. But being able to access prepared maps and other derived material is just a first step, governments need to provide access to raw geospatial data in commonly used Web-friendly formats. In the U.S. Data.gov was launched in late 2009.  In 2013, President Obama issued an Executive Order making open and machine readable the new default for federal government information in the United States.

A key transition occurred in 2008 when the U.S. Geological Survey announced a policy of open data. Prior to this Landsat images were made available on a cost recovery basis, costing hundreds to thousands of dollars per scene. But in January 2008 Barbara Ryan, the Associate Director for Geography at the U.S. Geological Survey, and Michael Freilich, NASA’s Director of the Earth Science Division, signed off on a Landsat Data Distribution Policy that made Landsat images free to the public. The USGS announced the free-and-open data policy on April 21, 2008. In a recent talk at the Geospatial Knowledge Infrastructure Summit Barbara Ryan described the impact of that decision. Landsat downloads jumped from 53 scenes per day to 5,700 scenes per day. The economic benefit derived from the Landsat data was estimated at US $1.7 billion domestically, $400 million internationally for a global total of $2.1 billion In 2011.

New satellite constellation will capture 30 cm imagery with up to 15 revisits per day

I remember back in August 2014, when the first images from WorldView-3 were being downloaded. WorldView-3 was the first one-foot-resolution commercial satellite. It operated at an altitude of 617 km, flew in a sun synchronous orbit and could collect up to 680,000 km² per day.  It had 31 cm panchromatic resolution, could capture 680,000 km² per day, and had an average revisit time of less than a day.  In 2021 MAXAR is planning to launch the WorldView Legion constellation, six satellites with 30 cm resolution, better than 5 meter absolute accuracy, able to capture 5 million km² every day, and with up to 15 revisits per day.

At GeoIgnite Online 2020 Kumar Navulur, Chris Formeller, and Kathryn Seitz presented an overview of the specifications and capabilities of the Legion constellation. The total land mass of the Earth is 148.9 million km2.  The new constellation will be able to capture 30 cm imagery of over 3 % of that every day.  Over much of the lower 48 states in the U.S. it will be able to revisit the same location 15 times per day. Over Canada it will be closer to 8 revisits per day.  Some of the satellites will fly sun synchronous orbits and others mid-inclination orbits.  What this means practically is that it will be possible to capture imagery at different local times of the day, unlike with sun synchronous orbits where imagery is captured at the same local time every day. For many customers imagery will be available within 24 hours of capture.

Using satellite imagery to prioritize vegetation management for utilities

A recent startup LiveEO is focusing on the digitalization of energy and transportation network assets using satellite imagery. LiveEO uses current multi-band imagery including radar and hyperspectral data at various resolutions from earth observation satellites to provide services to railway, electric power and pipeline operators.  The services include vegetation, third party encroachment, and ground distortion monitoring. I recently had the opportunity to chat with Sven Przywarra, co-founder of LiveEO.

Transmission line inspections are essential in ensuring grid reliability and resilience.  They are generally performed by manned helicopters often together with a ground crew.  There are serious safety issues when inspections are conducted by helicopter. Data may be collected with cameras and analyzed to detect a variety of conditions including corrosion, evidence of flash over, cracks in cross arms, and right-of-way issues such as vegetation encroachment. in North America annual inspections are mandated by NERC  and are not optional. With over 200,000 miles of high-voltage transmission lines and 5.5 million miles of distribution lines in the United States, improving the efficiency and reducing the risk of inspections would have a major impact on the reliability of the power grid. 

Vegetation is a significant source of outages for many utilities. In some regions from one quarter to one half of all outages can be ascribed to vegetation. Vegetation encroachment initiated the massive North American northeastern outage of 2003. It is so critical for utility operations that some utilities have a vice president responsible for vegetation management. It can be costly for utilities to identify areas of high risk vegetation encroachment. Previously I have blogged about an electric power utility in Brazil that used Airbus Pléiades satellite imagery combined with GIS data to identify high risk areas for vegetation encroachment in a power grid.

A recent startup LiveEO is focusing on the digitalization of energy and transportation network assets using satellite imagery. LiveEO analyzes current multi-band imagery including radar and hyperspectral data for vegetation, third party encroachment, and ground distortion monitoring. LiveEO uses multi-band imagery at various resolutions to distinguish vegetation and network assets, stereo imagery to estimate tree heights, difference detection to identify changes, and interferometry to detect subsidence and other ground distortions at the cm level.  LiveEO is developing its own inhouse training data for machine learning algorithms. 

For electric power utilities LiveEO applies machine learning multi-band satellite imagery from multiple sources to automatically identify areas of high risk for vegetation encroachment and fallover anywhere on the planet. The results of these analyses are available to clients over the web on handheld and other devices.  This enables follow-up ground inspections of areas of high criticality to develop pruning and other response programs. LiveEO has applied this approach to transmission as well as distribution circuits.  To demonstrate the wide ranging capability of their approach LiveEO is offering a demonstration of the analysis of 50 km of line anywhere in the world.

A number of companies, realizing that earth observation imagery in itself is of limited value, are not in the business of selling imagery, but the results of analytics applied to the imagery.  I have blogged about a new company GHGSat who use their own nanosatellite to estimate emissions from industrial plants, leaks from flow and gathering infrastructure in natural gas fields, and even livestock.  LiveEO is another example of a company who apply analytics to the huge amount of earth observation data to provide a valuable service that is safer and less expensive than the traditional approach to transmission and distribution line vegetation monitoring.

Open source geospatial technology enables cost-effective solutions in many domains

At the GITA Pipeline Forum in Houston, an enthralling presentation by Rich Henry of Pivvot, a recent startup that builds solutions for pipeline routing and other applications, highlighted how open source geospatial technology is transforming applications in a variety of sectors. Pivvot builds SaaS solutions using open-source components on Google Maps and Amazon Web Services to provide a web-based cloud computing platform with a curated database from over 900 public and private sources.  The advantages of this approach for siting pipeline routes are dramatic; applying infinite computing resources, completely avoiding the arcane art of managing ESRI ArcGIS licences, and integrating hundreds of data layers making it possible to run alternative pipeline right of way scenarios in hours rather than weeks or months.

Recent technical advances have made it possible to implement cost-effective solutions for utility, telecom and pipeline asset management, integrating BIM and geospatial, and geospatially-enabled machine learning.  Standards are a key underpinning in the development of geospatial applications for a number of vertical industries.  There are now many widely adopted geospatial standards; WMS, WFS, GML, KML, InfraGML, PipelineML, WaterML to name just a few from the Open Geospatial Consortium (OGC) that were not available two decades ago.  Furthermore, open source geospatial developers like open standards.  Many open source libraries and applications implement standards from the Open Geospatial Consortium (OGC), W3C and GeoJSON. Since 1999 there has been rapid development of open source geospatial projects including MapServer, GDAL, PostGIS, GeoServer, OpenLayers, Leaflet, QGIS and others.  The Open Source Geospatial Foundation (OSGEO) was founded in in 2006 and the open source code-base has enabled entrepreneurs to rapidly develop low cost geospatial applications because the cost of maintaining most of the code is shared across a broad developer community.  I'll describe a few of the solutions built on open source components that I have encountered in the last few years.

Records management for utilities

One of the challenges faced by large utilities with multiple GIS, outage management, customer information and other systems, which are often the result of mergers, is how to make spatial and associated data, in different, often incompatible systems, accessible to all users not just GIS professionals.  At Distributech 2018 I discovered how Duke Energy is running an off-the-shelf web-based product based on open source components that is being used by 8000 users, 2000 of them using it 24/7 every day, to access GIS and associated data in several GIS systems, both online and offline, using laptops, tablets and phones. A single, common easy-to-use web tool permits viewing geospatial asset data stored in GE Smallworld, ESRI ArcGIS, Intergraph GTechnology, spatialNET and Google Maps as well as non-GIS data in Maximo, SAP, three different outage management systems, four or five customer information systems, CAD drawings and photos and images.

At Distributech 2019 a new entrant IQGeo emerged into the utility and communications space and announced an alternative to traditional GIS.  myWorld Capture is a mobile solution that prioritizes direct data maintenance from the field.  Building on an open source geospatial platform that supports data versioning and interfaces to legacy utility/communications GISs such as ArcGIS and Smallworld, myWorld provides tools that make it possible to develop and maintain a high quality near real-time GIS with the information that is required for a complete network asset data model including equipment location, descriptions, connectivity, condition, and status.

Open access to municipal geospatial data

At GITA 2018 in Phoenix, Bob Basques, GIS Systems Developer at the City of St Paul, described a system called COMPASS he and his team have developed an easy to use tool that allows city employees and the public access to all of the City's spatial and associated data including, for example, scans of surveyors' notebooks, 2.2 million street level photos, and permitting and licensing information from 200 different applications. Based completely on open source components the system is compact and efficient enough to run on a Raspberry Pi.

Integrating BIM and 3D geospatial

At the GEO|Design+BIM conference I discovered mago3D a 3D web-based geo-platform developed by Gaia3D.  It is open source and integrates many well-known open source geospatial products and APis. It has an engine for handling massive, complex BIM objects that supports LOD (level of detail) control and efficient spatial queries and culling (the removal of objects that are not visible to the observer which is key to efficient visualization or large, complex structures.) It integrates with two of the best known open source 3D virtual globe tools, Cesium and NASA WorldWind, for visualization in a geospatial context.  Viewing BIM structures in mago3D has some similarities with a Google Earth experience. The source code is available under open source licenses.

Machine learning

Deep learning algorithms have been developed by academia and as a result the code for the most part is open source. For example, a deep neural network model developed originally for medical image segmentation called U-Net is open source and has been applied to identifying building footprints. Successful training of deep networks requires thousands of labeled training samples. Labeled data involves people on the ground manually ground-truthing land use types and other features so that the deep learning algorithms can learn what to recognize.  Machine learning is computationally intensive and requires a distributed engine.  Spark is a open source (Apache) project which enables distributed processing for global scale computation. Machine learning has been used successfully to identify building footprints and transportation networks and deforestation in satellite imagery.

RasterFrames is a free and open source toolkit allowing scientists, data scientists, and software developers to process and analyze geospatial-temporal raster data with the same flexibility and ease as any other data type.  Open source training data is beginning to become available, for example, the System for Terrrestrial Ecosystem Parameterization (STEP) has 2000 manually labeled sites covering 17 different land cover types including five forest types scattered across all continents.

Pipeline siting

To identify an optimal route for a pipeline on your own local hardware involves stitching together many DEMs, calculating slopes, finding, downloading and integrating hundreds of data layers, and avoiding certain areas (protected or uncooperative owners).  You have to worry about having enough disk space and the calculations can take a long time with no certainty of completion.  Running these calculations on several machines in parallel requires an ArcGIS licence for each of the machines.  Together this could result in months of data collection and processing.  In contrast even for long pipelines an open source solution from Pivvot running in the cloud means you don't have to worry about file management and disk capacity or getting enough machines and licenses to run it in parallel. Using Pivvot's curated datasets data collection and processing can be done in a few hours and it is scalable to many data layers.

Migrating to an open source platform is easier than you think

Samuel Song is responsible for the National Cemetery GIS implementation within Veterans Affairs and like many in government increased use of GIS means a greater sensitivity to cost.  Samuel outlined his experience in migrating to open source at a GIS in the Rockies conference in Denver.

 He had been using ArcGIS Desktop as his desktop GIS, SQL Server as a database, ArcGIS Enterprise as a web server, and ArcGIS API for Javascript for his web GIS. To look at ways to lower his costs Samuel experimented with open source geospatial software; QGIS for his desktop GIS, PostgreSQL/PostGIS for his database, GeoServer for his web server, and Leaflet or OpenLayers for his web GIS (all part of the OSGEO open source geospatial stack). His most important findings are that QGIS is a very stable, powerful alternative to ArcMap, PostgreSQL/PostGIS is a stable alternative to SQL Server and perhaps most importantly migrating data to an open source spatial database from an ESRI Geodatabase was much easier than he expected. His overall conclusion was that with the currently available open source software stack it is possible to move to a fully open source platform from an existing ArcGIS implementation and that the migration is much simpler than expected.

Records management for utilities

One of the challenges faced by large utilities with multiple GIS, outage management, customer information and other systems, which are often the result of mergers, is how to make spatial and associated data, in different, often incompatible systems, accessible to all users not just GIS professionals.  At Distributech 2018 I discovered how Duke Energy is running an off-the-shelf web-based product based on open source components that is being used by 8000 users, 2000 of them using it 24/7 every day, to access GIS and associated data in several GIS systems, both online and offline, using laptops, tablets and phones. A single, common easy-to-use web tool permits viewing geospatial asset data stored in GE Smallworld, ESRI ArcGIS, Intergraph GTechnology, spatialNET and Google Maps as well as non-GIS data in Maximo, SAP, three different outage management systems, four or five customer information systems, CAD drawings and photos and images.

Samuel Song is responsible for the National Cemetery GIS implementation within Veterans Affairs and like many in government increased use of GIS means a greater sensitivity to cost.  Samuel outlined his experience in migrating to open source at a GIS in the Rockies conference in Denver.

 He had been using ArcGIS Desktop as his desktop GIS, SQL Server as a database, ArcGIS Enterprise as a web server, and ArcGIS API for Javascript for his web GIS. To look at ways to lower his costs Samuel experimented with open source geospatial software; QGIS for his desktop GIS, PostgreSQL/PostGIS for his database, GeoServer for his web server, and Leaflet or OpenLayers for his web GIS (all part of the OSGEO open source geospatial stack). His most important findings are that QGIS is a very stable, powerful alternative to ArcMap, PostgreSQL/PostGIS is a stable alternative to SQL Server and perhaps most importantly migrating data to an open source spatial database from an ESRI Geodatabase was much easier than he expected. His overall conclusion was that with the currently available open source software stack it is possible to move to a fully open source platform from an existing ArcGIS implementation and that the migration is much simpler than expected.

New Radarsat Constellation imagery will be available to public at no cost

Canada's first synthetic aperture radar satellite was launched in 1995. Since then Radarsat-2 was launched in 2007 and continues to operate.  In June of this year the Radarsat Constellation (RCM) was launched. At the GeoIgnite conference Sergey Samsonov, Canada Centre Remote Sensing (NRCan), explained that in addition to the technical improvements such as a revisit time of four days and data latency for some applications as low as 10 minutes, RCM data will be made available at no cost.  This is a different business model from Radarsat-1 and Radarsat-2 and is intended to reduce entry costs for developers and entrepreneurs.  The RCM data will be available through the Earth Observation Management System.  In addition earlier this year the Canadian Space Agency and the Canada Centre for Mapping and Earth Observation made RADARSAT-1 images of Earth available to researchers, industry and the public at no cost. The 36,500 images are available through the Earth Observation Data Management System.

One of the primary benefits of radar is that it is not affected by cloud cover.  Radarsat-2 and RCM are capable of one meter resolution, although five meter is more typical.  In addition the Radarsat satellites are capable of using interferometry to measure centimeter movements and deformations of the Earth's surface in the horizontal and vertical directions.  In his talk Sergey Samsonov showed some examples from Radarsat-2 of subsidence in Vancouver, Seattle, Mexico City, Alberta, and other locations detected by radar interferometry.  I have blogged about an application Network Alert developed by Planetek that uses satellite radar interferometry to detect ground subsidence and to alert municipal water network operators of possible water leaks.

The new constellation dramatically improves the ability to monitor changes on the Earth's surface. The revisit time for the new constellation is much shorter. While the previous satellites had a revisit time of 24 days, the new RCM satellites will revisit the same location every four days with a greater frequency in Canada's fat north. The time between data acquisition and the data being available with the RCM constellation is remarkable. The data latency for the new RCM constellation depends on the type of data. For ships in Canadian waters it is 10 minutes. For other maritime surveillance it is 30 minutes. For Canadian and global disaster management it is 2 hours - and remember it is not affected by cloud cover.  For ecosystem monitoring applications the daya will be available in 24 hours. Pre-defined recurrent observation scenarios (standard coverages) will be made available to users in advance of acquisition.

The RCM is designed for three main areas: maritime surveillance (ice, surface wind, oil pollution and ship monitoring), disaster management and environmental monitoring. While the mission design initially focused on maritime security requirements, land security, particularly in the Arctic, will be dramatically enhanced. In addition economic growth is also being targetted.  Developers and entrepreneurs are being encouraged by an open data policy to develop a wide range of applications in Canada and internationally.

Applying machine learning to identifying housing starts from satellite imagery

Deep learning is increasingly finding practical applications using geospatial data, specifically, satellite imagery. At the GeoIgnite conference in Ottawa, Nicolas Martinez of Statistics Canada described a project scheduled to kick-off in July to use machine learning to identify housing starts across Canada.  The goal of the project is to improve the accuracy and coverage of housing starts survey, specifically to fill data gaps for smaller remote and indiginous communities which are generally excluded from the current survey program. 

Canada Mortgage and Housing Corporation (CMHC) currently employs a significant number of people at their head office in addition to field agents across the country who physically confirm residential starts and completions.  The objective of the project is to apply convolutional neural network (CNN) machine learning to satellite imagery to extract information about housing starts and completions. The imagery will be preprocessed to prepare it to be used in developing a machine learning processing model.  The largest challenge will be developing training and testing imagery data sets.

The original application of deep learning to photo imagery was by Geoffrey Hinton at Google who used it to distinguish dogs and other objects in photos that people uploaded.  More recent applications of satellite imagery targeted landuse and agricultural identification. For example, remote sensed imagery from a satellite can be used to create a model that will differentiate between corn and potatoes, using factors that can be calculated from satellite imagery such as normalized difference vegetation index (NDVI) (min, max, mean), texture (min, max, mean), vegetation height, and geometric factors such as orientation.  I blogged about how open source code and publicly available training data has been applied to track deforestation and reforestation in Mato Grosso, a state in the central amazon region of Brazil. A study in the Netherlands used CNN to identify blocked waterways using overflight imagery with a success rate of 97 %.

An important application of satellite imagery is identifying built structures. I have blogged about how satellite imagery has been used to identify buildings and transportation networks by applying a neural network model together with OpenStreetMap layers and high resolution Worldview multispectral imagery.  NVIDIA has demonstrated the ability to automate detection of many road networks using deep learning algorithms and multi-spectral high resolution imagery.  Another application of CNN and imagery in the Netherlands was able to differentiate residential roofs with dormer windows from those without.

The essential key to effective application of deep learning is good training and test data - ground-truthed data that involves, for example, someone on the ground identifying whether the fields seen in the imagery are corn or potatoes or something else. There are  publicly available training data sets that can be used to train satellite imagery CNN models. For example, I blogged about the System for Terrrestrial Ecosystem Parameterization (STEP) dataset which has 2000 manually labeled sites covering 17 different land cover types scattered across all continents. A large publicly available training dataset derived from Sentinel 2 imagery contains 30,000 polygons of land use training data for ten classes of land use: annual crop, forest, herbaceous vegetation, highway, industrial, pastures, permanent crop, residential, river and sea or lake.

The Statistics Canada project is scheduled to begin in July with initial work on a proof of concept and CNN testing to begin in August and September.  The communities chosen for the initial phase are  Kitchener-Cambridge-Waterloo, Red Deer, and Iqaliut.  The first phase is scheduled to be completed by winter 2019.  The project plans to use Tensorflow and Keras, open source libraries and a python framework for applying machine learning.  Artificial intelligence (AI) has often suffered from inflated expectations, complex code, heavy processing requirements and not very practical applications. There is a growing body of practical applications in the geospatial domain that shows that as a result of the immense processing power available in the cloud machine learning is able to generate fairly easily practical and useful results.  As Chris Holmes pointed out in his talk at FOSS4GNA, the challenge for Statistics Canda will be to develop reliable and publicly available training and test datasets to enable deep learning models to be created for a broader range of applications using the huge volumes of satellite imagery and other geospatial data that are now available.

Deep learning is increasingly finding practical applications using geospatial data, specifically, satellite imagery. At the GeoIgnite conference in Ottawa, Nicolas Martinez of Statistics Canada described a project scheduled to kick-off in July to use machine learning to identify housing starts across Canada.  The goal of the project is to improve the accuracy and coverage of housing starts survey, specifically to fill data gaps for smaller remote and indiginous communities which are generally excluded from the current survey program. 

Canada Mortgage and Housing Corporation (CMHC) currently employs a significant number of people at their head office in addition to field agents across the country who physically confirm residential starts and completions.  The objective of the project is to apply convolutional neural network (CNN) machine learning to satellite imagery to extract information about housing starts and completions. The imagery will be preprocessed to prepare it to be used in developing a machine learning processing model.  The largest challenge will be developing training and testing imagery data sets.

The original application of deep learning to photo imagery was by Geoffrey Hinton at Google who used it to distinguish dogs and other objects in photos that people uploaded.  More recent applications of satellite imagery targeted landuse and agricultural identification. For example, remote sensed imagery from a satellite can be used to create a model that will differentiate between corn and potatoes, using factors that can be calculated from satellite imagery such as normalized difference vegetation index (NDVI) (min, max, mean), texture (min, max, mean), vegetation height, and geometric factors such as orientation.  I blogged about how open source code and publicly available training data has been applied to track deforestation and reforestation in Mato Grosso, a state in the central amazon region of Brazil. A study in the Netherlands used CNN to identify blocked waterways using overflight imagery with a success rate of 97 %.

An important application of satellite imagery is identifying built structures. I have blogged about how satellite imagery has been used to identify buildings and transportation networks by applying a neural network model together with OpenStreetMap layers and high resolution Worldview multispectral imagery.  NVIDIA has demonstrated the ability to automate detection of many road networks using deep learning algorithms and multi-spectral high resolution imagery.  Another application of CNN and imagery in the Netherlands was able to differentiate residential roofs with dormer windows from those without.

The essential key to effective application of deep learning is good training and test data - ground-truthed data that involves, for example, someone on the ground identifying whether the fields seen in the imagery are corn or potatoes or something else. There are  publicly available training data sets that can be used to train satellite imagery CNN models. For example, I blogged about the System for Terrrestrial Ecosystem Parameterization (STEP) dataset which has 2000 manually labeled sites covering 17 different land cover types scattered across all continents. A large publicly available training dataset derived from Sentinel 2 imagery contains 30,000 polygons of land use training data for ten classes of land use: annual crop, forest, herbaceous vegetation, highway, industrial, pastures, permanent crop, residential, river and sea or lake.

The Statistics Canada project is scheduled to begin in July with initial work on a proof of concept and CNN testing to begin in August and September.  The communities chosen for the initial phase are  Kitchener-Cambridge-Waterloo, Red Deer, and Iqaliut.  The first phase is scheduled to be completed by winter 2019.  The project plans to use Tensorflow and Keras, open source libraries and a python framework for applying machine learning.  Artificial intelligence (AI) has often suffered from inflated expectations, complex code, heavy processing requirements and not very practical applications. There is a growing body of practical applications in the geospatial domain that shows that as a result of the immense processing power available in the cloud machine learning is able to generate fairly easily practical and useful results.  As Chris Holmes pointed out in his talk at FOSS4GNA, the challenge for Statistics Canda will be to develop reliable and publicly available training and test datasets to enable deep learning models to be created for a broader range of applications using the huge volumes of satellite imagery and other geospatial data that are now available.

A killer application for satellite imagery collected daily for the entire earth

At the GeoIgnite conference in Ottawa, Wade Larsen of Urthecast described what may be a killer application for satellite imagery captured daily. Farmers in the US over-irrigate by 25% per year. A recent study has demonstrated that monitoring vegetation growth on a daily basis can save farmers $50 000 in water per year. With 260,000 pivots operating in the US, this amounts to billions of dollars of savings every year.

Urthecast has been known for having two cameras (Iris and Theia) on the International Space Station and two Deimos earth observation satellites. Last year Urthecast announced the acquisition of Geosys Technology, a software company that processes earth observation imagery for agricultural applications.  Their latest initiative is called UrtheDaily with the objective of capturing nine bands of imagery from six satellites covering the Earth's entire landmass every day. The nine bands are identical to Sentinel-2's visual and near-infrared bands. The cameras are cross-calibrated with Sentinel-2 to assure data compatibility. The six cameras enable a daily revisit for higher latitudes and somewhat less for tropical areas. It captures imagery at the same time, from the same altitude every day.  About 150 terabytes will be acquired and downloaded every day. The imagery has five meter resolution (GSD) and represents scientific grade data quality that will be available on the cloud for user access within hours of acquisition. UrtheDaily is designed to support machine-learning and AI ready geo-analytics applications on a global scale. The UrtheDaily platform will provide the ability to discover data, run user algorithms, and deliver data to customers.

UrtheCast reports that it has already signed contracts primarily in the agriculture sector. An example of a potential killer app in agriculture is monitoring irrigation. Farmers in the U.S. overwater by 25 %. A four year in Texas by HydroBio which assessed the value of frequent monitoring of plant health and growth using satellite data concluded that to be effective this required high-frequency (daily) monitoring with highly calibrated sensors. With its resolution UrtheDaily will be able to monitor vegetation for individual pivots on a daily basis in order to optimize water utilization. It is estimated that this could save farmers $50,000 per pivot per year. (A pivot typically irrigates 120 acres.) There are 260,000 pivot irrigation systems in the U.S. which translates into the potential to save farmers billions of dollars per year.

To date UrtheCast reports that it has signed $350 million worth of contracts in the U.S., Russia, India and other countries. It plans to launch six satellites in 2021 and become operational in 2022.