SPACE/COSMOS
Using machine learning to overcome blind spots in satellite-based PM10 monitoring
image:
This workflow involves three main steps: (a) data integration, which processes previously developed SV data and other multisource data; (b) model building, which employs automated ML methods to develop an optimal, self-iterating model, and (c) model deployment, which fine-tunes the model using dynamically updated data from the past two years and retrieves hourly PM10 concentrations for the most recent day.
view moreCredit: ©Science China Press
This study is led by Prof. Huizheng Che and Dr. Ke Gui from the State Key Laboratory of Severe Weather & Key Laboratory of Atmospheric Chemistry of CMA, Chinese Academy of Meteorological Sciences. Sand and dust storms pose an ongoing threat to environment and public health. PM10, the primary pollutant of sandstorms, is directly detected by sparse ground-based stations, which fail to capture the incursion process of large-scale events. While satellite spectral products integrating with ancillary data provide an indirect method, real-time PM10 monitoring remains constrained by delays in meteorological data inputs. "Satellite-based PM10 retrieval limited to coarse daily scale, and most studies focus on reconstructing historical datasets rather than tracking real-time PM10 levels.” Huizheng says.
Huizheng and Ke, together with lab member Xutao Zhang, sought to build a real-time surface PM10 retrieval (RT-SPMR) framework powered by interpretable automated machine learning with dynamic updates. The framework comprises three core modules (see Figure 1 below) and uniquely integrates the team’s custom-developed surface visibility dataset as a key input. These ingenuities enable the RT-SPMR to provide real-time gridded PM10 data across China, with a gapless coverage of spatial resolution of 6.25 km, which temporally updates every hour (see Figure 2 below).
The team found that the RT-SPMR model demonstrates robust generalization and stability, achieving higher daily retrieval accuracy than previous studies, as confirmed through cross-validation and rolling iterative validation experiments. "These performance tests highlight the readiness of RT-SPMR for operational deployment." Ke says.
During a severe sandstorm event that began on March 14, 2021, in northern China, the RT-SPMR showcased exceptional performance in real-time tracking of the fine-scale evolution of dust intrusion. It successfully captured the dynamic PM10 variations in areas beyond the reach of geostationary satellite imagery and ground observation networks (see Figure 3 below).
"This new framework overcomes the limitations of current satellite-based PM10 monitoring. We remain committed to advancing our models by incorporating more detailed information to enhance retrieval accuracy. Our goal is to produce even more reliable datasets, providing robust support for atmospheric environmental monitoring." Ke says.
The seamless real-time PM10 data products generated by RT-SPMR are expected to provide more accurate initial field information for dust storm forecasting models, improving prediction accuracy. This research also supports the “virtual network” objective outlined in the WMO’s Scientific and Implementation Plan: 2021-2025 for the Sand and Dust Storm Warning Advisory and Assessment System (SDS-WAS). Additionally, it offers a "China solution" for building refined monitoring systems in other countries frequently affected by dust storms.
Central panels show the multiyear (2020–2022) annual average PM10 from the RT-SPMR model (shading) and observations (points). Surrounding panels show the multiyear averaged 24-hour (China Standard Time, CST) PM10 maps.
Credit
©Science China Press
The 6-hourly evolution of dust plumes as revealed by gapless PM10 retrievals (on the left), Himawari-8 dust RGB composite images (in the center), and PM10 observations (on the right) during a dust storm event that occurred from 21:00 (CST) on 14 March to 15:00 (CST) on 17 March 2021. GIF credit: Xutao Zhang.
Credit
GIF credit: Xutao Zhang.
Real-time mapping of gapless 24-hour surface PM10 in China
https://doi.org/10.1093/nsr/nwae446
Journal
National Science Review
Oxygen for Mars
Direct splitting: electrochemical process uses carbon dioxide to produce oxygen
To mitigate global climate change, emissions of the primary culprit, carbon dioxide, must be drastically reduced. A newly developed process helps solve this problem: CO2 is directly split electrochemically into carbon and oxygen. As a Chinese research team reports in the journal Angewandte Chemie, oxygen could also be produced in this way under water or in space—without requiring stringent conditions such as pressure and temperature.
Leafy plants are masters of the art of carbon neutrality: during photosynthesis, they convert CO2 into oxygen and glucose. Hydrogen atoms play an important role as “mediators”. However, the process is not particularly efficient. In addition, the oxygen produced does not come from the CO2 but from the absorbed water. True splitting of CO2 is not taking place in plants and also could not be achieved at moderate temperatures by technical means so far.
Ping He, Haoshen Zhou, and their team at Nanjing University, in collaboration with researcher from Fudan University (Shanghai) have now achieved their goal to directly split CO2 into elemental carbon and oxygen. Instead of hydrogen, the “mediator” in their method is lithium. The team developed an electrochemical device consisting of a gas cathode with a nanoscale cocatalyst made of ruthenium and cobalt (RuCo) as well as a metallic lithium anode. CO2 is fed into the cathode and undergoes a two-step electrochemical reduction with lithium. Initially, lithium carbonate Li2CO3 is formed, which reacts further to produce lithium oxide Li2O and elemental carbon. In an electrocatalytic oxidation process, the Li2O is then converted to lithium ions and oxygen gas O2. Use of an optimized RuCo catalyst allows for a very high yield of O2, over 98.6 %, significantly exceeding the efficiency of natural photosynthesis. As well as pure CO2, successful tests were also carried out with mixed gases containing varying fractions of CO2, including simulated flue gas, a CO2/O2 mixture, and simulated Mars gas. The atmosphere on Mars consists primarily of CO2, though the pressure is less than 1 % of the pressure of Earth’s atmosphere. The simulated Mars atmosphere thus contained a mixture of argon and 1 % CO2.
If the required power comes from renewable energy, this method paves the way toward carbon neutrality. At the same time, it is a practical, controllable method for the production of O2 from CO2 with broad application potential—from the exploration of Mars and oxygen supply for spacesuits to underwater life support, breathing masks, indoor air purification, and industrial waste treatment.
(2677 characters)
About the Author
Dr. Ping He is a Professor and Head of the Department of Energy Science and Engineering at the College of Engineering and Applied Sciences, Nanjing University. His research interests focus on transformative electrochemical energy technologies, including high-energy-density batteries, electrochemical CO₂ reduction, and lithium resource extraction and recycling. He is a Fellow of the Royal Society of Chemistry and serves as an Associate Editor for ACS Energy & Fuels.
Journal
Angewandte Chemie International Edition
Method of Research
Experimental study
Subject of Research
Not applicable
Article Title
Artificial Carbon Neutrality Through Aprotic CO2 Splitting
Article Publication Date
24-Mar-2025
Delft University of Technology and Brown University pioneer technology for next-generation lightsails in space exploration
Researchers at Delft University of Technology (The Netherlands) and Brown University have developed scalable nanotechnology-based lightsails that could support future advances in space exploration and experimental physics
Delft University of Technology
image:
Comparison of produced nanomaterial with largest lightsail shown at Starshot's announcemet in 2016.
view moreCredit: Richard Norte
Researchers at TU Delft and Brown University have developed scalable nanotechnology-based lightsails that could support future advances in space exploration and experimental physics. Their research, published in Nature Communications, introduces new materials and production methods to create the thinnest large-scale reflectors ever made.
Lightsails are ultra-thin, reflective structures that use laser-driven radiation pressure to propel spacecraft at high speeds. Unlike conventional nanotechnology, which miniaturizes devices in all dimensions, lightsails follow a different approach. They are nanoscale in thickness—about 1/1000th the thickness of a human hair—but can extend to sheets with large dimensions.
Fabricating a lightsail as envisioned for the Starshot Breakthrough Initiative* would traditionally take 15 years, mainly because it is covered in billions of nanoscale holes. Using advanced techniques, the team, including first author and PhD student Lucas Norder, has reduced this process to a single day.
A new type of nanotechnology
‘This is not just another step in making things smaller; it’s an entirely new way of thinking about nanotechnology,’ explains Dr. Richard Norte, associate professor at TU Delft. ‘We’re creating high-aspect-ratio devices that are thinner than anything previously engineered but span dimensions akin to massive structures.’ The current prototype measures 60mm x 60mm and is 200 nanometres thick, covered in billions of nanosized holes. This represents a significant step forward in large-scale lightsail fabrication.
‘Other recent advancements in the field, such as from Caltech, have demonstrated nanoscale control over sail structures at micrometer scales, whereas our approach scales to centimeter-sized structures while maintaining nanoscale precision manufacturing.’ If scaled up, the lightsail made by Norte and colleagues would extend over the length of seven football fields with a thickness of only a millimetre. ‘It’s not just its high aspect ratio that makes this material special; it’s the simultaneous combination of large scale and nanoscale in the same material that makes it lightweight and reflective,’ says Norte.
The team combined state-of-the-art neural topology optimization techniques with cutting-edge fabrication methods to achieve this. ‘We have developed a new gas-based etch that allows us to delicately remove the material under the sails, leaving only the sail,’ Norte explains. ‘If the sails break, it’s most likely during manufacturing. Once the sails are suspended, they are actually quite robust. These techniques have been uniquely developed at TU Delft.’
‘Our work combines the latest advancements in optimisation to explore new ways to find unintuitive designs,’ says Dr. Miguel Bessa from Brown University. ‘By blending neural networks with topology optimization, we’ve created designs that push the boundaries of what’s possible in both nanophotonics and large-scale manufacturing.’
From picometers to centimetres to lightyears
The proposed lightsails leverage laser-driven radiation pressure to accelerate to astonishing velocities, enabling rapid interplanetary travel. For instance, probes propelled with the developed lightsails could, in theory, reach Mars in the time it takes for international mail to arrive. While such vast distances remain a goal for the future, recent studies have demonstrated that similar lightsails can currently be propelled over distances as small as picometers. Norte and his team are now preparing experiments to push the new membrane sails across distances measured in centimetres against Earth’s gravity. ‘It might not sound like a lot, but this would be 10 billion times farther than anything pushed with lasers so far.’
A universe of possibilities
Beyond space exploration, these materials open new possibilities for experimental physics. The ability to accelerate masses to high velocities offers unprecedented opportunities to study light-matter interactions and relativistic physics at macroscopic scales.
‘This EU-funded research places Delft at the forefront of nanoscale material science,’ Norte adds. ‘Now that we can make these lightsails as large as semiconductors can make wafers, we are exploring what we can do with today’s capabilities in nanofabrication, lasers, and design. In some ways, I think it might be just as exciting as missions beyond the solar system. What is remarkable to me is that creating these thin optical materials can open a window into fundamental questions such as; how fast can we actually accelerate an object. The nanotechnology behind this question is certain to open new avenues of interesting research.’
*Breakthrough Starshot Initiative
Currently, it would take around 10,000 years for our fastest rockets to reach even the nearest star outside the solar system. The Breakthrough Starshot Initiative, uniting thousands of researchers, seeks to reduce that journey to just 20 years. By developing ultra-light, laser-propelled spacecraft the size of microchips, the project envisions humanity's first interstellar exploration beyond the solar system. It is part of the Breakthrough Initiatives, a program funded by private investors. Starshot was launched by Yuri Milner and Stephen Hawking in 2016.
Associate Professor Richard Norte in his lab in Delft (The Netherlands)
Credit
Delft University of Technology
Journal
Nature Communications
Method of Research
Experimental study
Subject of Research
Not applicable
Article Title
“Pentagonal photonic crystal mirrors: scalable lightsails with enhanced acceleration via neural topology optimization”,
Article Publication Date
24-Mar-2025
AI reshapes how we observe the stars
Researchers use deep learning and large language models to classify stars with high accuracy
image:
A large-language-model-based system for classifying stellar light curves, integrating specialized components for text, image, and audio data processing.
view moreCredit: Yu-Yang Li et al.
AI tools are transforming how we observe the world around us — and even the stars beyond. Recently, an international team proved that deep learning techniques and large language models can help astronomers classify stars with high accuracy and efficiency. Their study, “Deep Learning and Methods Based on Large Language Models Applied to Stellar Light Curve Classification,” was published Feb. 26 in Intelligent Computing, a Science Partner Journal.
The team introduced the StarWhisper LightCurve series, a trio of AI models, and evaluated their performance alongside other state-of-the-art approaches. All models were trained to classify variable stars from their light curves with automated deep learning, which enables automatic optimization of key factors such as learning rate, batch size, and model complexity, minimizing the need for manual tuning.
The team sourced training data from NASA’s Kepler and K2 missions, focusing on five major types of variable stars. A small number of rare variable stars were also included to improve model generalization.
The comprehensive evaluation shows high classification accuracy across different AI architectures for major variable star types. Among the top-performing models, the Conv1D + BiLSTM model — a hybrid deep learning approach combining convolutional layers for feature extraction and recurrent layers for temporal patterns — achieved 94% accuracy. The Swin Transformer model, a variant of the popular transformer architecture originally developed for natural language processing, achieved 99% accuracy.
Notably, the Swin Transformer demonstrated 83% accuracy in identifying Type II Cepheid stars, a rare class of pulsating stars that make up just 0.02% of the dataset.
Although the Swin Transformer delivers impressive accuracy, it requires extra preprocessing to convert light curve data into images. In contrast, StarWhisper LightCurve achieved nearly 90% accuracy with minimal manual intervention, reducing the need for explicit feature engineering. This efficiency not only streamlines data processing but also paves the way for parallel data analysis and the advancement of multi-modal AI applications in astronomy.
The StarWhisper LightCurve series consists of three specialized large language models, each fine-tuned for a different astronomical data format:
- A large language model, built on Gemini 7B, for classifying light curves as structured time-series text.
- A multimodal large language model, built on DeepSeek-VL-7B-Chat, for processing image-based light curve representations.
- A large audio language model, built on Qwen-Audio, for converting light curves into sound waves.
The StarWhisper LightCurve series is part of the broader StarWhisper project, a large language model designed for astronomy with strong reasoning and instruction-following capabilities. More details can be found at: https://github.com/Yu-Yang-Li/StarWhisper.
Journal
Intelligent Computing
Article Title
Deep Learning and Methods Based on Large Language Models Applied to Stellar Light Curve Classification
No comments:
Post a Comment