Our satellites do not directly measure events; they measure the continuous physical state of the Earth system—a seamless field of temperature, reflectance, elevation, and chemical concentration. An event, from this perspective, is not a direct observation but a derived insight. It represents a significant departure from a baseline state, extracted from this continuous data through the detection of a threshold-crossing (a river exceeding flood stage), a rapid change (a sudden drop in ground elevation indicating a landslide), or a complex pattern recognized by sophisticated algorithms. The primary challenge of event-based science is therefore to transform the torrent of dense, continuous measurements into a catalog of discrete, meaningful events. This is NOT the same as derived variables which is simply a variable transformation of observations under the same spatial-temporal geometry.
This task is complicated further in cases where we cannot observe the event’s source, but only its cascading effects. A volcanic eruption in a remote, cloud-covered location might be initially invisible, but its event signature appears days later and thousands of kilometers away as an atmospheric sulfur dioxide plume. Here, we face a difficult inverse problem: we measure the effect and must use a combination of sensor data and physical models to trace it back to the source event. This requires not just seeing the what, but understanding the how and why of its propagation through the Earth system.
This fundamentally changes the nature of the data itself. While the raw satellite measurements are stored in dense, gridded formats like NetCDF or Zarr, an event catalog is fundamentally sparse and relational. A wildfire is not a discrete cube of pixels; it is a discrete entity with specific attributes: a start time, a location, a final burn area, an intensity, and a probable cause. This structure is more akin to a relational database than a simple data array, where one event (a drought) can be linked to another (a wildfire). To truly understand our planet, we must move beyond thinking in terms of continuous fields and towards building these rich, interconnected event databases.
The Earth’s systems are characterized by long periods of gradual change punctuated by these abrupt, high-impact events. A single satellite, observing at a fixed scale and cadence, struggles to capture both the slow precursory signals and the rapid event dynamics. A system of systems approach is the necessary solution to build these event catalogs. By integrating diverse sensors, we can monitor the long-term build-up of risk with one set of instruments, detect the trigger with another, and track the event’s evolution and consequences with a third, populating our relational understanding of an event-driven world.
🌍 Atmosphere: Capturing Transient Events in the Air¶
The atmosphere is a turbulent, four-dimensional system where a gradual build-up of energy can be released in minutes, creating hazardous events that are difficult to forecast and track.
Event: Hurricane Landfall & Tornado Formation¶
Variables Monitored: Sea surface temperature, atmospheric pressure, wind speed, cloud structure.
ELI5: This is when a powerful ocean storm hits land, bringing destructive winds and flooding that can endanger coastal communities.
Event Definition: The event is defined by a threshold-crossing, where the storm’s center (a point feature) crosses the coastline (a line feature). The intensity is a continuous variable, but the landfall itself is a discrete spatio-temporal event.
The Challenge: Ideally, this requires continuous (sub-minute temporal res), high-resolution (~100m spatial res) 3D snapshots of the storm’s structure. Geostationary satellites (GOES) provide the ideal temporal resolution (5-15 minutes) for tracking cloud movement, but their spatial resolution (~1km) is too coarse to see fine internal details. To compensate, polar-orbiting scatterometers like ASCAT provide vital surface wind data with medium spatial resolution but suffer from low temporal resolution (revisiting only once or twice a day). Microwave sounders (AMSU) add another crucial, albeit infrequent, layer by probing the storm’s internal warm core, which is opaque to optical sensors.
Event: Extreme Air Pollution Episode¶
Variables Monitored: Aerosol Optical Depth, concentrations of NO_2, SO_2, Ozone.
ELI5: This is when the air in a city or region becomes so polluted that it is unsafe to breathe for an extended period.
Event Definition: The event is a threshold-crossing over a duration, where a pollutant concentration exceeds a health-based standard for a specified time. The event’s data structure is a 4D spatio-temporal polygon or raster mask outlining the affected area.
The Challenge: The ideal system would provide hourly (high temporal res) air quality maps at neighborhood scale (high spatial res). Daily mappers like TROPOMI provide excellent spectral resolution to identify specific pollutants but have coarse spatial resolution (~5km) and daily temporal resolution, missing intraday peaks. Aerosol sensors like VIIRS or MODIS offer better temporal resolution (twice daily) for tracking plumes but lack the specific spectral signature for pollutant gases. The high-cadence stare of GOES can track the movement of a dense smoke plume but has the lowest spectral and spatial resolution, highlighting the trade-off between seeing frequently versus seeing clearly.
Event: Volcanic Ash Cloud Incursion into Airspace¶
Variables Monitored: Thermal anomalies, SO_2 concentration, ash particle properties.
ELI5: This is when a volcano erupts and shoots a cloud of fine, abrasive ash high into the atmosphere, creating a major hazard for jet engines.
Event Definition: The event is the presence of an ash cloud, defined by a threshold-crossing of ash or co-emitted SO_2 concentration. The data structure is a dynamic polygon that moves and deforms with the wind over time.
The Challenge: The ideal is continuous, 3D tracking of the ash cloud with precise chemical speciation. The system begins with GOES, whose excellent temporal resolution (minutes) provides the first detection via thermal anomalies, but its spectral resolution is too poor to definitively confirm ash. This is solved by TROPOMI, which offers excellent spectral resolution to measure the co-emitted SO_2 tracer gas, but its temporal resolution is only daily. Thermal sensors on VIIRS offer a trade-off, with better spatial resolution than GOES and better temporal resolution than TROPOMI (twice daily) to help refine the plume’s boundaries.
Event: Major Methane “Blowout” or Super-Emitter Release¶
Variables Monitored: Atmospheric methane (CH_4) concentration.
ELI5: This is a massive and sudden gas leak from an industrial facility, like a broken pipeline or malfunctioning well, releasing huge amounts of greenhouse gas.
Event Definition: The event is defined by a rapid positive gradient (rate of change) in methane concentration that exceeds a significance threshold. The data structure is a point feature for the source location and a plume polygon for the downwind extent.
The Challenge: Ideally, we need continuous monitoring with meter-scale spatial resolution to detect and attribute any significant leak. The first alert often comes from TROPOMI, which has the necessary spectral resolution to see methane and the temporal resolution (daily) to screen globally, but its coarse spatial resolution (~5km) can only detect enormous events. Once flagged, high-resolution spectrometers like GHGSat are tasked, providing the exquisite spatial resolution (~25m) needed to pinpoint the source, but they lack the global coverage and high temporal resolution, making them unsuitable for initial detection.
🌳 Land: Witnessing Abrupt Changes on the Surface¶
The land surface often changes slowly, but events like fires, floods, and droughts can alter landscapes in a matter of hours or weeks.
Event: Flash Drought & Crop Failure¶
Variables Monitored: Soil moisture, vegetation health (NDVI), evapotranspiration (ET).
ELI5: This is when conditions turn from normal to severe drought extremely quickly—in a matter of weeks instead of months—catching farmers off guard and causing crops to fail.
Event Definition: The event is identified by a high negative gradient in soil moisture and vegetation health indices over a short period. The data structure is a regional polygon that grows as the drought intensifies.
The Challenge: The ideal system would measure plant water stress daily at the individual farm scale (~10m). The event begins with a drop in soil moisture, which microwave sensors like SMAP can measure with high radiometric precision but very coarse spatial resolution (~40km). Thermal sensors like ECOSTRESS then detect the stress response (heating) at a higher spatial resolution (~70m) but with a lower temporal resolution (a few days). Finally, optical satellites like Sentinel-2 see the result—a drop in “greenness”—with excellent spatial resolution (~10m) but can be blocked by clouds, showcasing a trade-off between direct physical measurement and high-resolution visual confirmation.
Event: Illegal Forest Clearing Pulse¶
Variables Monitored: Forest cover, canopy structure.
ELI5: This is when a large area of rainforest is cut down and cleared in a very short amount of time, often to make way for cattle ranching or farming.
Event Definition: The event is a rapid change detection, where the radar backscatter or vegetation index of a forested area drops below a stable baseline. The data structure is a new polygon representing the cleared area.
The Challenge: Ideally, one would have daily, high-resolution (~5m) cloud-free imagery of all global forests. In persistently cloudy regions, all-weather radar on Sentinel-1 provides the essential high temporal resolution (6-12 days) to detect texture changes, but its spatial resolution (~20m) can miss subtle, selective logging. This radar alert then triggers analysis using high-resolution optical imagery from Planet or Landsat, which provide superior spectral and spatial resolution for confirming deforestation but are limited by their lower temporal resolution and vulnerability to cloud cover.
Event: Catastrophic Flood Inundation¶
Variables Monitored: River height, rainfall rate, water extent.
ELI5: This is when a river overflows its banks and submerges surrounding land, threatening homes, infrastructure, and lives.
Event Definition: The event is a threshold-crossing, where the measured surface water extent exceeds the historically normal river channel. The data structure is an inundation polygon or raster mask.
The Challenge: A perfect system would map water depth in 3D, hourly, at building scale. Pre-event monitoring requires radar altimeters (Sentinel-6) which have excellent vertical precision for river height but poor spatial and temporal resolution (they measure sparse lines every 10 days). The trigger, intense rainfall, is monitored by GPM with good temporal resolution (hours) but coarse spatial resolution (~5km). During the event, SAR from Sentinel-1 provides the ideal all-weather mapping capability with good spatial resolution (~20m) but its temporal resolution (days) means it might miss the flood peak.
Event: Wildfire Ignition and Eruption¶
Variables Monitored: Fuel moisture, thermal anomalies, fire radiative power.
ELI5: This is the moment a new wildfire starts and then suddenly “erupts” from a small flame into an uncontrollable blaze.
Event Definition: The ignition is a threshold-crossing where a thermal anomaly appears that is significantly hotter than the background. The data structure begins as a point, which then grows into an active fire polygon.
The Challenge: Ideally, we would detect every ignition within a minute at 10m resolution. The ignition event is a small thermal spike, first detected by GOES due to its excellent temporal resolution (minutes) but its coarse spatial resolution (~2km) means the fire is already of a considerable size. Higher spatial resolution thermal data from VIIRS or MODIS (~375m) can then more accurately map the fire front and measure its energy release, but their temporal resolution (hours) is too slow for real-time firefighting decisions.
🌊 Ocean: Detecting Hazards in the Deep¶
The ocean’s surface is a dynamic interface where atmospheric and oceanic processes converge to create distinct, often hazardous, events.
Event: Marine Heatwave¶
Variables Monitored: Sea Surface Temperature (SST).
ELI5: This is like an underwater heatwave, where a large patch of the ocean gets dangerously warm for an extended time, which can kill coral reefs and disrupt fisheries.
Event Definition: The event is a threshold-crossing over a duration, where the SST exceeds a statistical extreme (e.g., the 90th percentile) for at least five consecutive days. The data structure is a 4D spatio-temporal polygon.
The Challenge: The ideal would be daily, global, cloud-free SST maps at 1km resolution. Sensors like VIIRS provide excellent spatial resolution (~750m) and daily global coverage but are blinded by clouds. To create a complete map, their data must be blended with passive microwave sensors (AMSR2), which can see through clouds but have a much coarser spatial resolution (~25km). This fusion trades spatial detail for temporal and spatial completeness, which is essential for detecting the formation of the heatwave event.
Event: Catastrophic Oil Spill¶
Variables Monitored: Surface roughness, spectral properties of water.
ELI5: This is when a large amount of oil suddenly leaks from a ship or oil rig, spreading across the ocean surface and causing severe environmental damage.
Event Definition: The event is a change detection, identified by the appearance of a new area with anomalously low surface roughness (in radar) or unique spectral properties (in optical). The data structure is a moving polygon.
The Challenge: Ideally, we need hourly, high-resolution (~10m) monitoring of all shipping lanes. The first alert often comes from SAR satellites (Sentinel-1), whose key advantage is all-weather, day/night capability and sensitivity to surface roughness, but its temporal resolution is on the order of days. This low-cadence, wide-area search then triggers urgent tasking of high-resolution optical satellites (Planet, WorldView), which offer superior spatial resolution (~1-3m) to identify the source vessel but are useless if the location is cloudy or it is nighttime.
Event: Harmful Algal Bloom (HAB) Outbreak¶
Variables Monitored: Chlorophyll concentration, water temperature, ocean currents.
ELI5: This is when tiny algae in the ocean multiply out of control, sometimes creating toxins that can poison seafood and make the water unsafe.
Event Definition: The event is a threshold-crossing, where the chlorophyll-a concentration exceeds a regional baseline that indicates a bloom is forming. The data structure is a polygon that drifts and changes with ocean currents.
The Challenge: Perfect monitoring would involve daily, high-resolution hyperspectral imagery of all coastal zones. “Ocean color” satellites (Sentinel-3) provide the necessary spectral resolution to identify chlorophyll but at a moderate spatial resolution (~300m). To forecast the bloom’s path, this is combined with SST data from VIIRS (better spatial resolution, different physics) and current maps from altimeters (Sentinel-6), which have very poor spatial resolution (measuring lines kms apart) but provide the essential ocean velocity data that imagers cannot.
🧊 Cryosphere: Observing a World in Rapid Transition¶
Events in the cryosphere, such as the collapse of an ice shelf, are some of the most dramatic indicators of a changing climate.
Event: Ice Shelf Collapse or Major Calving¶
Variables Monitored: Ice velocity, surface meltwater, crevasse propagation.
ELI5: This is when a massive floating ice sheet at the edge of Antarctica or Greenland suddenly shatters and breaks away, creating huge icebergs.
Event Definition: The event is a rapid topological change, where a single ice polygon fractures into multiple polygons. It is often preceded by an acceleration (high gradient) in ice velocity and the propagation of line features (crevasses).
The Challenge: Ideally, we would have daily, meter-scale 3D measurements of ice shelves. The precursory acceleration is measured with InSAR from Sentinel-1, which offers excellent precision (cm-level motion) but at a low temporal resolution (6-12 days). Surface melt ponds are mapped by Landsat, which provides good spatial resolution (30m) but is useless with clouds. High-resolution imagery from commercial satellites is used to track specific, critical fractures, offering the best spatial resolution but at the lowest temporal resolution and smallest footprint.
Event: Rapid Sea Ice Breakup¶
Variables Monitored: Sea ice concentration, motion, and thickness.
ELI5: This is when a solid sheet of sea ice, which may have been stable all winter, rapidly cracks and breaks apart into smaller, drifting floes.
Event Definition: The event is a rapid negative gradient in sea ice concentration and a rapid positive gradient in the number of leads (cracks). This transforms a single, large ice polygon into a field of many smaller polygons.
The Challenge: Ideal monitoring would provide hourly, meter-scale maps of ice thickness and type. The broad weakening is monitored by passive microwave sensors (AMSR2), which provide daily coverage with excellent sensitivity to ice type but at a very coarse spatial resolution (~10km). For navigation, ships need meter-scale detail, provided by SAR (RADARSAT) with high spatial resolution (~5m) and all-weather capability, but its swath is narrow and revisits are less frequent, making it a “zoom” tool guided by the coarse daily maps.
Event: Rain-on-Snow Flood¶
Variables Monitored: Snow cover, snow water equivalent (SWE), rainfall rate.
ELI5: This is a dangerous type of flood that happens when heavy, warm rain falls on a deep snowpack, causing the entire pack to melt almost instantly and rush downstream.
Event Definition: This is a co-occurrence event, defined by the presence of a snow-covered area (polygon) and a rainfall rate that exceeds a critical threshold within that same polygon.
The Challenge: The ideal system would map snow water content and rainfall at a high resolution (~100m) hourly. The risk is assessed by mapping snow-covered area with MODIS or VIIRS, which have excellent daily temporal resolution but cannot see the water content. The GPM mission’s radar detects the trigger—a warm rain storm—with excellent temporal resolution (hours) but coarse spatial resolution (~5km). The fusion is essential, as neither satellite alone can capture the full physics of the event.
⛰️ Solid Earth: Capturing a Restless Planet¶
The ground beneath us moves constantly, with the slow build-up of strain released in sudden, catastrophic events like earthquakes and landslides.
Event: Volcanic Eruption¶
Variables Monitored: Ground deformation, thermal output, gas emissions (SO_2).
ELI5: This is when a volcano erupts, sending ash, gas, and sometimes lava out of a vent in the Earth’s crust.
Event Definition: A complex event defined by multiple threshold-crossings: ground deformation rate exceeds a background level, a significant thermal anomaly appears, or gas emissions spike. The source is a point feature (the vent).
The Challenge: Ideally, we would have continuous, multi-parameter monitoring of every active volcano. The slow swelling of the ground is measured in millimeters by InSAR from Sentinel-1, offering unparalleled spatial precision but low temporal resolution (days). A thermal spike might be seen by MODIS, which has good temporal resolution (4x daily) but coarse spatial resolution (~1km). The definitive detection of an eruption can be an SO_2 plume seen by TROPOMI, which has excellent spectral resolution but only daily revisits. Each sensor provides a different, crucial piece of the warning puzzle.
Event: Catastrophic Landslide Failure¶
Variables Monitored: Ground creep, soil moisture, rainfall.
ELI5: This is when a slope of land, like a hillside or mountainside, suddenly gives way and collapses, sending rock and soil rushing downhill.
Event Definition: The event is a phase transition, identified by an acceleration (a high gradient) in ground displacement that signifies the shift from slow creep to rapid failure. The data structure is a polygon (the landslide scar and debris field).
The Challenge: A perfect system would monitor slope stability in 3D at daily intervals. The precursory creep is detected by InSAR (Sentinel-1), which has the required millimeter-level precision but a temporal resolution (6-12 days) that is too slow to catch the final acceleration. The immediate trigger, intense rainfall, is monitored by GPM with high temporal resolution (hours) but at a coarse spatial resolution (~5km) that can miss localized, convective downpours. The system must fuse the precise but infrequent deformation data with the frequent but coarse rainfall data.
Event: Infrastructure Failure from Ground Subsidence¶
Variables Monitored: Ground elevation, groundwater mass.
ELI5: This is when slow, invisible sinking of the ground causes critical infrastructure like a bridge, levee, or building to suddenly crack and fail.
Event Definition: The event is a threshold-crossing, where the cumulative strain from a long-term, slow subsidence (a low gradient) exceeds the known structural tolerance of a piece of infrastructure. The data structure is a point or polygon representing the asset.
The Challenge: Ideally, we would have millimeter-precision elevation maps of all critical infrastructure, updated weekly. The slow subsidence is mapped by InSAR (Sentinel-1) with excellent spatial resolution and precision, but its temporal resolution is just enough to track the slow process, not sudden changes. The underlying cause—groundwater withdrawal—can be inferred at a massive regional scale by GRACE-FO, which has extremely coarse spatial resolution (>150km) but is sensitive to the total mass change that drives the process.