Satellite data: reading wildfires, water and air with AI

This article shows how satellite data is used together with ground measurements, research flights and artificial intelligence to understand phenomena ranging from large wildfires to rare clouds to urban air quality. Satellites indicate where to look, ground sensors provide continuous readings, flights collect samples at altitude, and algorithms help sort enormous flows of images and measurements. What follows are concrete examples, numbers, and the limits of what is currently known.

When smoke rises above the mountains and changes the atmosphere

Large fires can create clouds called pyrocumulonimbus, abbreviated pyroCb: thunderstorms fed by the fire itself. These clouds can throw smoke into the lower stratosphere, the layer just above the troposphere where the air is drier and particles can persist for a long time. Researchers have recorded roughly 70 pyroCb a year globally, and at least 13 over the continental United States in 2026. During the INSPYRE campaign an aircraft sampled aerosols at about 12 kilometres, an altitude ordinary models rarely observe directly. Injected particles can remain for months or years, changing how much energy Earth absorbs and reflects.

Images from the International Space Station showed smoke around the Cascade volcanoes, including Mount Hood. The Grasshopper fire had burned nearly 84,000 acres, about 34,000 hectares, and contributed to worsening air quality in inhabited areas. Satellites such as MODIS and multispectral sensors measure cloud-top temperature: readings well below −40°C indicate very high cloud tops, and a very cold cloud top is a sign the cloud has climbed to the upper troposphere or beyond. Those observations guide flights and models. What the models still cannot reliably predict is why only some fires generate pyroCb at all.

Integrating satellite observations, ground measurements and flight data is what makes extreme fires comprehensible.

Water forecasts and operational decisions from the same data

The same observation chain, combining satellites, ground sensors and models that learn from data, is also used to manage water and energy. During the 2025-2026 winter the Pacific Northwest recorded extremely low snow cover: January, February and March each had the lowest snow cover for those months in the MODIS record since 2001. In many mountain areas the melt no longer delivered the gradual flow expected, with snowpack between 20% and 50% of normal in many zones, reducing the summer water release.

An operational forecasting provider integrates weather forecasts, river measurements and satellite products derived from the VIIRS instrument on the Suomi-NPP satellite. HydroForecast, by Upstream Tech, updates flow estimates every two hours. Tacoma Power, a public utility, uses those forecasts alongside gauging stations and operational judgement to manage two dams, whose production is equivalent to the annual consumption of 151,000 homes. Thanks to those satellite inputs operators were able to keep reservoirs higher in advance, compensating for a dry spring, while still balancing the risk of insufficient capacity in case of heavy rain.

Air quality, urban networks and the role of citizens

Combining ground stations, flights and observations from space also helps separate the sources of particulate matter in cities. PM2.5 is fine particulate with a diameter of 2.5 micrometres or less; a micrometre is a thousand times smaller than a millimetre. In Addis Ababa, a network of 10 stations from the MAIA project, the Multi-Angle Imager for Aerosols, measured PM2.5 continuously between 2022 and 2025. The network found a three-year average of 30 micrograms per cubic metre, more than three times the annual reference standard of the US Environmental Protection Agency. MAIA also estimated black carbon levels roughly 4 to 9 times higher than in three US metropolitan areas monitored by the same project.

MAIA pairs ground measurements with multi-angle observations from space to distinguish different types of aerosol, the term for particles suspended in air. That combination makes it possible to recognise increases linked to rush-hour traffic or to bonfires during festivals. The project includes public health researchers who will use concentration maps to study possible links with health outcomes. The broader global health assessment estimates that exposure to PM2.5 is associated with about 4.9 million deaths a year.

In parallel, citizen science reduces the human workload with algorithms. Space Cloud Watch collects photographs of noctilucent clouds, which reflect sunlight after sunset. Volunteer Namai Chandra built a machine learning pipeline to classify the images: it pre-screens photographs, classifies the clouds, and routes low-confidence images for human review. That keeps expert judgement where it is needed and automates the repetition where it is not.

The challenges are shared. Connecting measured exposures to climate or health effects requires more data: more ground sensors and more observations from space, properly integrated. Human interpretation should not be replaced. Concretely, whether forecasts, fire response and health decisions improve over the coming years depends on increasing the quantity and quality of satellite data and on how operationally it is used.

Better data and better integration are what turn observation into decisions.

Sources: