Celebrating SkyTruth’s 2022 Program Accomplishments
Additions to staff and software, as well as global impacts, define a year of robust growth
As we enter a new year, we face two of the greatest challenges of our time: a changing climate and a biodiversity crisis.
At SkyTruth, we’re making sure the impacts of poorly regulated industrial activity on ecosystems and people don’t go unseen. By sharing the view from space, we’re removing the veil that protects those who pollute oceans with bilge, degrade critical habitats through mining, and warm the Earth by emitting methane. If people see with their own eyes the damage being done, they can put pressure on governments and businesses to clean up their acts.
As a society, we must work together to stem the tide of environmental ruin. That’s why our images and data, available to the public for free, turbocharge advocacy and help level the playing field around the world.
In 2022 SkyTruth bolstered its efforts to reveal harm being done to our planet. We filled key positions, including Chief Technology Officer, Director of Impact and Partnerships, Communications Director, and Senior Software Engineer. Crucially, we advanced our programs related to environmental health and justice, land and biodiversity conservation, mountaintop removal mining, watershed protection, and oceans. Our tools expanded greatly, enhancing our efforts to detect harmful algal blooms; stop pollution at sea; track illegal gold mining in the Amazon; monitor gas flaring; and protect landscapes.
Looking ahead, we will continue developing and sharing tools for advocates to investigate and analyze the environmental impacts of extractive industries. On both the climate and biodiversity fronts, Cerulean will be a major factor in the fight to protect our oceans, as well as the diverse species that depend on them. Another tool in the works is Verdant, which will use artificial intelligence and satellite imagery to monitor changes to landscapes, ensuring protected areas remain protected. Finally, we will transform our Alerts platform into the best free public tool in existence for remote sensing of environmental damage caused by human activity. At last, the tools to discover and raise the alarm about environmental damage will be in the hands of those who need them most: the communities directly harmed.
Conservation technology has the power to reshape and protect our planet. By shining a light on the often hidden impacts of human activity, we can hold businesses and governments accountable, help build public pressure for positive change, and promote more sustainable practices for generations to come.
Ending Methane Venting & Flaring
SkyTruth has a long history of tracking the insidious issue of natural gas (primarily methane) venting and flaring. This year, we provided NASA’s DEVELOP team with unfiltered flaring data in the Gulf of Mexico to help them identify likely sources of methane venting, in a successful effort to demonstrate methane leak detection at sea using publicly available satellite imagery. We also worked with the press, helping a journalist in Botswana follow gas flaring in Zimbabwe’s Hwange National Park, part of the Kavango Zambezi Transfrontier Conservation Area (KAZA). Further, we provided the Rio Grande International Study Center with flaring data and influenced an investigative report on flaring sites near marine protected areas.
Cerulean: Stopping Oil Pollution at Sea
SkyTruth is using satellite imagery and artificial intelligence to reveal the hidden impacts of global shipping and offshore oil production on the ocean and the climate. We began developing this technology, called Cerulean, in 2020 to automate the detection of oil slicks caused by illegal bilge dumping from vessels at sea. Our initial machine-learning model has been running globally since August 2020, scanning hundreds of thousands of satellite images and positively identifying thousands of shipping-related slicks within those images.
International law specifies how bilge water should be treated to remove oil and minimize its impacts on ocean ecosystems. However, Cerulean has revealed that many ships bypass costly pollution prevention equipment by simply flushing untreated bilge waste directly into the sea.
In 2022, we advanced Cerulean in several key areas by:
- Improving the automated slick detection model to more accurately detect slicks from both vessels and offshore oil platforms.
- Building a global database of offshore oil platforms and other human-built structures in the ocean.
- Automating the identification of ships associated with slicks using Automatic Identification System (AIS) broadcasts.
- Automating the entire image-handling and analysis pipeline in Google Cloud Platform, improving responsiveness and reliability.
Improved oil slick detection. The slick detection model developed in 2020 was trained to detect oily bilge water from vessels, but not chronic leaks from offshore oil infrastructure. In 2022, we developed a new model designed to detect oil spills from both vessels and offshore oil platforms.
To build out the new model, we hand-labeled hundreds of examples of leaks from offshore oil platforms and trained a new deep learning model to detect them automatically in satellite imagery. This will enable us to more confidently and accurately classify oil from both vessels and platforms, yielding results like those shown below.
Offshore oil platform database. To illustrate the environmental impacts of offshore oil production, we first need to understand where platforms are located. This basic information is not publicly available for many countries that host offshore drilling. We have been working with Global Fishing Watch to use satellite imagery to produce the first global offshore oil platform database, which is slated for release in early 2023. This database will enable us to confidently assign oil slicks and other detected pollution events to offshore oil infrastructure.
The global offshore oil platform database will also facilitate future work with our colleagues at Carbon Mapper, providing a list of observation targets for their methane monitoring mission. An early 2022 study by Carbon Mapper in the Gulf of Mexico indicated that nearly half of offshore platforms had persistent, observable methane leaks, which is a much higher rate than onshore oil and gas infrastructure. By utilizing our comprehensive, global list of targets, Carbon Mapper’s methane monitoring satellites (slated for launch in late 2023) will be able to create the world’s first global picture of methane emissions from offshore oil and gas development.
Automatic ship identification. One of the key value propositions of Cerulean is its ability to explicitly identify ships and platforms associated with oil pollution. Historically, we have identified vessels using a semi-automated process, where a simple algorithm compares ship tracks with slick locations and provides a list of possible candidates. Our analysts use that list to manually assess each candidate to find the best match.
In 2022, we improved this algorithm to achieve full automation where, for each detected slick, we produce a ranked list of the top five candidate vessels. In testing against a dataset of known coincident vessels, the automated algorithm placed the correct vessel in the top five 92% of the time, with the correct vessel ranked first 76% of the time. While there is still room for improvement, we achieved our 2022 benchmark of having the correct vessel appear in the top five candidates over 90% of the time. In 2023, we will work to improve this algorithm, and to account for dark vessels that do not use AIS to report their positions.
Google Cloud Platform automation. To shrink our own carbon footprint, we migrated the prototype Cerulean image-handling and analysis pipeline to Google Cloud Platform (GCP), a key step as we move steadily to achieving operational status in 2023.
The Cerulean cloud architecture is shown in the diagram below. In short, the pipeline is triggered by new satellite images appearing in public cloud storage. When new images arrive, the pipeline checks to ensure that an image is over ocean, at which point our deep learning model scans the image for the existence of probable slicks. Next it processes any detections to record its location and characteristics, then stores the detections in a cloud-based database that is immediately and automatically available for users to query via a programmatic application programming interface (API). This API will enable power users to access all of our detections and their associated metadata.
We will use that same API to serve data on our graphical frontend application, providing a seamless, highly visual experience for all users—both experts and non-experts.
Tracking Floating Production Storage & Offloading Vessels (FPSOs)
Floating Production, Storage, and Offloading vessels (FPSOs and their less-capable cousins, FSOs) are stationary ships used by the offshore oil and gas industry to produce, process, and store oil. FPSOs can be thought of as mobile oil platforms and are often old, repurposed oil tankers. A looming disaster is posed by the FSO Safer: a derelict, unmaintained vessel off the coast of Yemen that threatens to spill over a million barrels of crude oil. Another, the FPSO Trinity Spirit, exploded in 2021, leading to the deaths of several crew members.
FPSOs are poorly tracked and understood, and no comprehensive database exists to monitor their activities. Our objective is to create the world’s first full accounting of FPSOs and their impacts, including FPSO detection and tracking over time; identify vessels and ports most associated with particular FPSOs; monitor for associated oil pollution using Cerulean; and flag suspicious activities.
In 2022 we compiled a list of 586 probable FPSOs from various sources. We are currently verifying each of these using AIS location history and satellite imagery. As we complete this manual data cleaning, we’re using Cerulean to conduct a preliminary analysis of oil pollution associated with FPSOs. To date, we’ve found more than ten with chronic oil leaks over the past two years. This analysis was conducted with only a subset of the FPSO data, and we suspect that chronic oil pollution from FPSOs is a significant, underreported problem that SkyTruth will shine a bright light on in 2023.
Expanding & Improving SkyTruth Alerts
Many skytruthers are familiar with our SkyTruth Alerts platform. It provides an easy way for anyone to subscribe to an area of interest (AOI), and then receive alerts whenever certain environmental incidents, such as a new drilling permit or hazardous chemical spill, occur in those areas.
Throughout 2022 we’ve implemented dozens of user interface changes and software upgrades to make the platform faster, more useful, and easier to use. The following are some highlights of these upgrades:
Planet daily images. In 2022 we made available Planet daily, high-resolution satellite imagery to go along with the Planet monthly composite basemaps. This made SkyTruth Alerts the only public platform—as far as we’re aware—where users can view Planet daily images online at no cost.
Our users jumped on this new feature and applied it in a variety of ways across several geographical regions. In the United States, one person monitored petrochemical destruction of Louisiana wetlands and damage to homes from LNG plants and CO2 injection wells; another researched the history of mine sites in the Santa Rita Mountains in southern Arizona to detect new mining since late 2019.
The Planet imagery in SkyTruth Alerts had a global impact, as well. It helped journalists identify suspected, and potentially illegal, drilling sites in the Kapinga Kamwali Community Conservancy in Namibia. Also, in Ukraine, it helped determine the extent of pollution caused by the collapse of waste collection systems, as well as fuel and oil spillages due to shelling. Further, the imagery was applied to monitor deforestation caused by the Russian army in Ukraine as a result of constructing fortified positions. When a November 2022 explosion destroyed road and rail sections along the Nova Kakhovka HPP dam, Planet imagery helped determine damages and human/environmental risks.
Other applications of the feature included: following the footprints and damages related to tailings dam failures; providing imagery to a Twitter user monitoring floodplains, volcanic eruptions, and other natural and environmental disasters; and groundtruthing machine learning detections for several SkyTruth projects, including Project Inambari in Peru.
New data layers, custom maps, and imagery. These are some of the additions made available in 2022:
- Dynamic World land cover products (see Verdant section)
- Map and data of 2022 fire locations and perimeters from the National Interagency Fire Center (NIFC)
- Map of the Nord Stream pipeline leak in the Baltic
- Map of Hurricane Ian’s track, including extent of hurricane and storm winds and locations of Florida phosphate mines at risk from flooding
- Map of Kapinga Kamwalye conservancy boundaries, including Dynamic World layer for detecting mining activity
- PALSAR-2 imagery: Synthetic Aperture Radar (SAR) imagery from the Japan Aerospace Exploration Agency (JAXA)
- Pipeline routes from Oil and Gas Watch, which focus on highlighting new and proposed pipelines and expansion projects
- Summary of monthly gas flaring data from the Earth Observation Group: now available in Alerts for any AOI or bounding box
- The European Space Agency’s WorldCover 10-meter global land cover map for 2020
Saving very large SkyTruth datasets in our Postgres database allows us to serve up subsets of data with much better load times for an online browser environment. Some current implementations include:
- Our offshore infrastructure dataset, which has 1.7 million entries
- Inambari mining detections: >200,000 complex polygons that can be visualized for any date range or zoom level, enabling a user-friendly timeline of detection footprints
Other developments. Aside from the core changes to Alerts, we have developed several other features:
- New “Combine Scenes” option that helps stitch together multiple images
- A method to reduce the number of unwanted images when searching for imagery
- Improved backend deployments by switching to Docker containers
- Switched to Google managed, auto-renewing SSL certificates
- Added CSV files to drag-and-drop capabilities
- Implemented a new tile server for Planet imagery
- Enabled users to create new AOIs from shapefiles dropped on maps
- Converted keys and API keys to use Google Secrets
Detecting Harmful Algal Blooms
Over the past four years SkyTruth has collaborated with Professor Joseph Ortiz’s lab at Kent State University (KSU) to implement a novel water quality analysis that identifies harmful algal blooms using satellite imagery in Google Earth Engine.
Harmful algal blooms affect every coastal and Great Lakes state in the U.S., and they also are a problem globally. Most algae are harmless, but under the right conditions, they can grow out of control. Some even release toxins that kill wildlife and make people sick. When categorizing blooms, the many types of algae, suspended sediment, and other aquatic materials can make it difficult to distinguish signals (e.g. cyanobacteria, the algae often responsible for algal blooms) from noise (everything else); this makes it hard to detect when blooms are harmful. The KSU method—a process called spectral decomposition—makes it possible to separate these different components and detect dangerous blooms.
Since late 2021, we have worked to enhance the automation of this workflow. Using Google Cloud Platform’s suite of automation tools, a human no longer needs to be in the loop to physically hit the go button in order to get back results from the spectral decomposition method. Now, the KSU team can go into cloud storage and find new imagery that has already been run through spectral decomposition.
Since SkyTruth’s GCP workflow became operational at the end of 2021, 92 Sentinel-2 images and 11 Landsat 8 images have been processed over the location of interest chosen by KSU for testing: the Muskingum Watershed Conservation District. The current implementation allows for a simple reorientation of the AOI, as well; simply set a new AOI in each of the Cloud Functions and processing will begin in your desired location. This year’s development has opened the door to fully automate the process of interpreting the outputs of the spectral decomposition, which presents a new opportunity for the SkyTruth team.
Measuring Recovery From Coal Mining
Mountaintop removal mining (MTM) remains a critical driver of ecosystem destruction and habitat loss in Central Appalachia. Over the past year, SkyTruth has continued to document this destruction, while simultaneously working to assess the legacy of harm that burdens Appalachian communities and ecosystems.
Beginning last year, we moved beyond solely documenting the location of surface mining operations in the Central Appalachian region and also began exploring the long-term ecological legacy mining has on landscapes. In 2022, SkyTruth led a team of researchers from Appalachian Voices, Colorado State University, The University of Nevada, and The University of Wyoming to explore these legacies. Our research, published in the journal Restoration Ecology, found that only 7.9% of land disturbed by mining activity has recovered to a level that is consistent with the unmined, forested sites across the region. This research expanded upon our analysis of mine health conducted in 2021 with Appalachian Voices and provides a more nuanced understanding of how mine sites experience recovery. We made this data publicly available so that it can be used by policymakers to improve reclamation policy, by regulators to improve their assessments of post-mining restoration, and by advocacy groups to support transparency efforts in the restoration space. The approach we take for assessing recovery on mined lands was designed to be useful for assessing recovery from other disturbance types in other ecosystems, as well.
In addition to our recovery assessment, we completed an annual update of our mining-extent dataset, delivering on our promise to provide regular updates and ensure the data remain useful to environmental advocates and researchers. The original research, published in 2018, has been cited nearly 100 times to date by other scientists and policy analysts.
We also have begun to explore avenues for future research and data creation for MTM. We have created a prototype model, which runs on early Landsat imagery, that will allow us to backdate our annual mining-extent dataset to 1974. This application is particularly important: it will allow us to better understand the impact of mining that occurred before the passage of the Surface Mining Control and Reclamation Act of 1977, and the pre-mining conditions of many areas mined after 1985. We also began research into calculating the biomass potential and storage capacity of post-mining landscapes, which is a precursor to designing carbon credit-generating restoration projects on mined landscapes.
Inambari: Detecting Illegal Mining in the Amazon
This year we were thrilled to publicly launch Project Inambari, an early alert system for tropical forest mining, at an event in Lima, Peru. Named after a river that flows down from the Andes and through the mining region of Peru’s southeastern rainforest, Inambari is the culmination of two full years of work. In that time, our engineers developed a machine learning model to specifically detect mined areas on satellite imagery and display the expansion of mining on an open map platform, freely available to the public.
A critical need. To protect biodiversity and human well-being in the Amazon, we must find a way to effectively monitor and intervene in illegal mining.
The tropical forests where we have developed our mining detection model have the highest biodiversity on Earth. Since gold mining in the Amazon results in total deforestation and removal of soil over a large area, the effect of unregulated mining on critical habitat has been catastrophic. Forests are left with no natural way to recover; for example, former rainforest environments at sites like La Pampa in Peru and Las Claritas in Venezuela are barren moonscapes of mud and rock, even in areas miners have long abandoned.
Small-scale gold mining in the Amazon depends on the use of mercury to extract and concentrate the gold. Miners boil off this mercury in the purification process, releasing toxic fumes that eventually settle into waterways. Once mercury enters the river, it converts to methylmercury and enters the food chain. This process puts people at risk: much of the population depends on fish from the river as a primary source of protein.
Unregulated and illegal resource extraction often becomes closely tied to criminal activity and labor exploitation, and investigations have shown this to be the case with mining in the Amazon. In many mining areas, the reach of the state is limited and inconsistent at best. Some regions like Venezuela’s mining arc are now clearly under the control of armed, non-state groups who derive the majority of their financing from illegal mining and systematic extortion of the local economy. High profits and territorial rivalries lead to acts of extreme violence involving miners and the local population. Native people of the Amazon, including a small number of tribes in voluntary isolation, are particularly vulnerable as mining activities encroach on their lands, introduce toxic levels of mercury, and make it more difficult to pursue sustainable economic activities like agriculture and tourism.
A novel monitoring solution. From the ground, the scale of destruction caused by mining is shocking; however, it mostly takes place in remote areas rarely visited by outsiders. Our Inambari platform creates a striking visual testament to the relentless advance of mining over past years. It also serves as a tool to detect new mining and guide intervention by local authorities.
Detection of mining activity such as that shown in the example above relies on satellite imagery from the European Space Agency’s Sentinel-1 and Sentinel-2 satellites.
By using SkyTruth’s Alerts platform (see below), we have been able to take advantage of a wide range of existing analyses and visualization tools. Alerts has also proved useful to our partners, allowing them to visually verify and investigate mining detections using a large catalog of recent and historic satellite imagery.
Verdant: Protecting Landscapes Worldwide
Verdant is SkyTruth’s project to build and deploy a free, real-time surveillance system for protected landscapes around the world. Verdant is being designed to promote protected area creation and effectiveness by providing transparency and public accountability for protected area integrity, ultimately preserving and restoring biodiversity while supporting and empowering local and Indigenous communities.
In 2022, we kicked off Verdant by building a simple, accessible habitat-change detection tool that can be applied instantly to landscapes anywhere in the world by anyone—not just technical experts—for free. After seeing an early prototype of our change detection tool, Google invited SkyTruth to present our work during a keynote talk at their Geo for Good conference in October 2022.
To build a tool that enabled global ecosystem change detection, we leveraged the Dynamic World land cover dataset. This near real-time, 10-meter-resolution global land use land cover dataset, produced through a partnership between Google and the World Resources Institute, leverages Sentinel-2 satellite imagery. Sentinel-2 observes the entire planet every five days and the Dynamic World models automatically classify every scene into nine land cover classes, including forest, grassland, urban, agricultural, and water.
We are using this unique dataset to detect habitat destruction by monitoring for persistent changes in land cover, such as new buildings, deforestation, desertification, and changes in surface water boundaries. When coupled with existing features of Alerts, such as dynamic mapping and satellite imagery, we are creating a unique tool for terrestrial change monitoring.
Using SkyTruth Alerts, we’ve made the massive, complex Dynamic World dataset available to all, through just a few simple steps for users to begin monitoring change in their ecosystem of interest:
- Define an AOI;
- Define the “before” and “after” periods to be compared;
- Define the land cover type to monitor for change (e.g. forest, water, all).
That’s all that’s required! To our knowledge, this is the simplest, custom, global-change-detection interface available today. Once those parameters are defined, the model automatically detects changes and displays them on an interactive map. The tool also automatically displays high-resolution imagery from Planet Labs, so you can visually verify detected changes that occurred during the monitoring period (see figure below).
Building Capacity: Staff & Interns
At SkyTruth, we consider people the most important resource of our organization. This mentality led us to bring on critical full-time staff in 2022, including Jason Schatz (Chief Technology Officer), Mitchelle De Leon (Director of Impact and Strategic Partnerships), Joel Goldberg (Communications Director), and Aemon Malone (Senior Software Engineer). Thanks to these strong additions to the SkyTruth team we now have the capacity to accelerate our innovation, enhance our product offerings, uplift our level of service, and expand our reach among industry groups, users, and the general public, ultimately generating greater positive impact for conservation.
Interns are a cornerstone of our day-to-day productivity here at SkyTruth, and 2022 was as productive as ever. SkyTruth interns from various backgrounds aided project Cerulean by performing cumulative oil spill analyses, helping Cerulean discriminate among oil slicks from vessels, offshore platforms and other sources. They also learned how to use a combination of our offshore infrastructure dataset and our satellite flaring data. For another project, they created an NDVI function that could be applied to the Lake Maracaibo region on any given day, making it possible to symbolize the data—this helped identify the extent of surface algal blooms while omitting other materials that might be recognized as oil by the Cerulean model.
Other projects interns supported include the expansion of the MTM process to Pennsylvania; accuracy assessment for our annual MTM update; and point validation for the offshore infrastructure dataset. Norway-based intern Dhiya Sathananthan reported on the EPA’s decision to temporarily suspend pollution reporting requirements during the COVID pandemic. Dhiya found that the EPA’s temporary policy could be linked to a 14% reduction in pollution reporting from March 13–August 31, 2020, suggesting that a significant number of pollution events went unreported to government officials during that time, leaving the public in the dark—a potential risk to health and property that typically afflicts economically and politically disadvantaged communities.
Our interns often continue their careers in the Remote Sensing/GIS field, showing how SkyTruth is a springboard for long-term success.