Multistation AI cuts weather-profile errors in field test
A new deep-learning system called CTDformer improved regional temperature and humidity retrievals from ground-based microwave radiometers across four stations in Xi’an. The study, published in Journal of Remote Sensing, suggests the approach could strengthen severe-weather monitoring and short-term forecasting where station coverage is sparse.
Why it matters: - Accurate atmospheric temperature and humidity profiles are central to weather prediction, climate simulation, convective assessment, pollution transport modeling and climate research. - Sparse and uneven radiometer coverage can leave gaps in regional profile retrievals. - Better multistation retrieval can improve numerical weather prediction initialization and earlier warnings for severe convection.
What happened: - Researchers at Wuhan University and collaborators developed CTDformer, a geoscience-informed deep-learning framework that combines Convolutional Neural Network, Transformer and DualBranchMLP components. - The study was published in Journal of Remote Sensing on 28 August 2026 and is identified by DOI 10.34133/remotesensing.1063. - CTDformer used multistation microwave brightness temperatures and spatial information to retrieve regional vertical atmospheric profiles. - The test network included four Xi’an-area stations: Weiyang, Chang’an, Lintong and Yangling, with Jinghe as the target station.
The details: - CTDformer combined CNN, Transformer and DualBranchMLP base learners with an XGBoost meta-learner. - The model explicitly encoded station identity and geographic relationships instead of treating each site as independent. - Inputs included 22-channel brightness temperatures, station one-hot encoding, latitude, longitude and distance to the target station. - Radiosonde profiles served as reference labels. - After temporal matching and quality control, the study used 529 samples across 89 vertical layers. - The data were split 7:1:2, normalized to [0.1,0.9] and trained with 10-fold cross-validation. - Rain-affected samples were removed using near-surface humidity thresholds, and channel-wise IQR filtering removed outliers. - Spatial priors were appended to the input features. - In the four-station test, CTDformer reduced temperature profile error by 7.64% and humidity profile error by 5.31% versus the individual models. - Adding spatial information lowered errors by more than 7%. - Networked training improved accuracy by 17.39% for temperature and 5.69% for humidity compared with single-site retrieval. - The model reached temperature correlation of about 0.99 and humidity correlation of about 0.75. - Before the 15 May 2021 rainfall event, CTDformer captured stronger convective available potential energy buildup than ERA5 and single-station methods. - CTDformer estimated pre-rain CAPE above 1,500 J/kg, while ERA5 showed only hundreds of joules per kilogram. - The strongest instability signal appeared 1 to 3 hours before rainfall.
Between the lines: - The study addresses a common weakness in atmospheric retrievals: models that learn one station at a time often miss spatial relationships across nearby instruments. - By fusing station metadata with multistation observations, CTDformer aims to preserve both numerical accuracy and physical consistency. - The results suggest that adding geographic context can help deep-learning systems generalize better in sparse or heterogeneous radiometer networks.
What's next: - The research team plans to integrate radiometers, radar, radiosondes and satellite products. - Future validation will focus on complex terrain and harder operational conditions. - The team also pointed to cross-attention, graph-based fusion, physics-informed losses and thermodynamic constraints as likely upgrades for stronger extreme-weather performance.
The bottom line: - CTDformer shows that multistation, geography-aware AI can improve atmospheric profile retrievals and deliver earlier severe-weather signals than single-station approaches.
Disclaimer: This article was produced by AGP Wire with the assistance of artificial intelligence based on original source content and has been refined to improve clarity, structure, and readability. This content is provided on an “as is” basis. While care has been taken in its preparation, it may contain inaccuracies or omissions, and readers should consult the original source and independently verify key information where appropriate. This content is for informational purposes only and does not constitute legal, financial, investment, or other professional advice.
Sign up for:
Energy News Watch
The daily local news briefing you can trust. Every day. Subscribe now.
Check Your Email!
We sent a one-time activation link to: .
Confirm it's you by clicking the email link.
If the email is not in your inbox, check spam or try again.
Welcome back!
is already signed up. Check your inbox for updates.