Space oddity: Shining light on uncertainties in satellite observations

New Finnish inverse problem research contributes to estimating uncertainties in interpreting satellite data. Quantifying uncertainty and its sources contribute to an improvement of a satellite retrieval algorithm, while data users benefit from a more reliable uncertainty assessment.

After the Second World War, the rising tensions between the United States (US) and the Soviet Union (USSR) led to a fierce political, technological, ideological, and economical competition known as the Cold War. To prove their technological and intellectual superiority, both great powers announced in 1955 their intention to launch the first satellite into space. With its origins in rocket and missile development arms race, this more peaceful “space race” would in the next 20 years lead to many far-reaching achievements in science and technology.

The first Earth-orbiting satellite in history, the Sputnik 1, or “Fellow Traveller”, was launched into space by the USSR in 1957. Since then, a vast network of satellites has risen to surround the Earth, making global communications, GPS navigation and positioning, and many other necessities of our modern life possible.  

For decades now, satellites have also powered sophisticated weather data and climate monitoring systems by observing the Earth’s atmosphere on a global scale. As this information contributes to understanding our transforming climate system and responding to related science questions, quality monitoring and error identification processes are becoming ever-more essential.

Complex conditions, novel solutions

On 13 June 2024, Researcher Anu Kauppi from the Finnish Meteorological Institute (FMI) defended her doctoral dissertation “Bayesian Model Selection, Averaging and Discrepancy in Satellite Remote Sensing of Atmospheric Constituents” at the University of Helsinki’s Faculty of Science. In her research, Kauppi focused on estimating uncertainties that may arise when satellite observations from the atmosphere are being interpreted with numerical methods. A reliable uncertainty assessment could be used for both developing algorithms and in using satellite data.

“As we try to observe the complex state of the atmosphere, we will need to make simplifications and assumptions. However, this contributes to the uncertainty of the solution”, Kauppi describes the inverse problems related to the atmosphere monitoring. “So, we have focused on quantifying uncertainty arising from parameter estimation, model selection, and inaccurate forward modelling to achieve a more realistic uncertainty estimate.”

To achieve this goal, the research team developed a novel algorithm based on the Bayesian approach. Named after the 18th-century English mathematician Thomas Bayes, this statistical approach allows the combining of prior information on parameters with sample information to provide a revised probability distribution for the parameter. According to Kauppi, the Bayesian model selection and averaging approach has not before been applied to satellite measurements.

Tried and tested        

One of the key areas of Kauppi’s research focused on aerosols and aerosol optical depth (AOD). Aerosols are tiny solid and liquid particles suspended in the atmosphere. Examples of aerosols include dust, volcanic ash, sea salts, smoke from wildfires, and pollution from human environments. AOD is a measurement to which extent these particles can prevent sunlight from reaching the ground by absorbing or scattering light.

For establishing AOD and associated uncertainty, the research team utilised measurements from the Ozone Monitoring Instrument (OMI) together with their algorithm. The Dutch and Finnish-developed instrument was launched to polar orbit in July 2004 on NASA’s Aura satellite. To verify their findings, the team compared their results to ground-based reference aerosol data based on the AErosol RObotic NETwork (AERONET), a network of hundreds of remote sensing stations all over the world established by NASA and the French PHOTONS.

After their method had showed promising potential, the research team tested its algorithm with another instrument, the TROPOspheric Monitoring Instrument (TROPOMI). Launched to polar orbit in October 2017 as part of the European Union’s Copernicus space programme, TROPOMI provides daily information on atmospheric constituents relevant for air quality and climate monitoring.

“After we finished working with OMI, there were discussions about how we could further test our algorithm with some other instrument. It was a very lucky coincidence for us that TROPOMI was launched and started operating at that time”, says Kauppi.

To infinity…

Another focus of Kauppi’s doctoral research work was related to ozone. Ozone is a molecule that occurs naturally in the atmosphere. A vertical ozone profile indicates ozone concentration in the atmosphere at different altitudes. While near the ground increased ozone levels contribute to global warming and affect negatively to air quality, in the upper atmosphere ozone protects the planet from harmful ultraviolet (UV) radiation that damages live cells. Therefore, a thinning of the ozone layer due to natural or human-made causes is a threat to the entire biosphere, emphasizing the need for reliable monitoring.

Kauppi’s thesis assessed the quality of the operational satellite ozone profile data. In addition to time of the year, latitude, and altitude of the atmosphere, the work also considered the ageing of the measuring instrument as an affective factor. By using data from three different satellites for a reference, the team’s verification process achieved a global scale. The research team’s quality verification effort was able to identify potential points of improvement and assist in correction of instrumental degradation.

Having demonstrated her research methods’ potential in quantifying uncertainties and assessing satellite data’s reliability, Kauppi will now pass the ball to the wider scientific community.

“Algorithms especially are now heading towards a massive shift thanks to artificial intelligence and machine learning. With our work now done, my hope is that monitoring instrument manufacturers and algorithm developers will find our research helpful and gain inspiration from it for future endeavours”, Kauppi concludes.

Photo: Anu Heikkilä | From the left: Professor Aku Seppänen from University of Eastern Finland as Opponent, Professor Samuli Siltanen from University of Helsinki as Custos, and Doctoral Candidate Anu Kauppi.