The exact text we saved

This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.

TRUVACE RECORD VERSION
record: TRV-2026-2539
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-11-12T07:50:25.975481Z
status: published
lens: g_space
sector: climate
headline: Satellite and Atmospheric Data Fusion for Improved Characterization of Thermodynamic Profiles
dek: Abstract This paper presents Total Atmospheric Characterization–Thermodynamics (TAC-Thermo), a global artificial intelligence (AI) data fusion system to improve satellite-retrieved thermodynamic profiles with a focus on data-sparse regions. TAC-Thermo uses available satellite retrievals and numerical weather prediction model output to upscale satellite-retrieved temperature and moisture profiles that can capture thermodynamic features (up to ∼12 km) on sigma vertical coordinate levels (i.e., 1.0–0.2). TAC-Thermo…
gain_title: In a validation against radiosonde soundings, the TAC-Thermo AI system had smaller dewpoint temperature errors than NUCAPS and the GFS at every sigma level, and filled in sounding data where radiosondes are absent.
problem_title: (none)
trace_subject: (none)
gain_reading: In a validation against radiosonde soundings, the TAC-Thermo AI system had smaller dewpoint temperature errors than NUCAPS and the GFS at every sigma level, and filled in sounding data where radiosondes are absent.
gain_evidence: TAC-Thermo had lower dewpoint temperature standard deviation errors for all sigma levels within the analysis compared to NUCAPS and GFS | they provide sounding information in the absence of radiosondes
problem_reading: (none)
problem_evidence: (none)
quick_read: Researchers built TAC-Thermo, an AI system that fuses satellite retrievals with numerical weather prediction output to sharpen temperature and moisture profiles of the atmosphere up to about 12 km, with a focus on regions that have few weather balloon launches. It was trained on 68,887 global radiosonde and collocated satellite profiles from a two-year period.

Checked against radiosonde soundings in all climate regimes, its dewpoint errors were smaller than those of NOAA's NUCAPS and the Global Forecast System at every sigma level analysed. For temperature, its largest standard deviation error (1.97°C) was close to the best values from NUCAPS (1.76°C) and the GFS (2.28°C).

The profiles can feed weather-model data assimilation and, as 0.25° gridded maps every 3 hours, help forecasters spot temperature and moisture inversions and cloud layers. The authors say the soundings could also give product developers and mission planners a global record of the planetary boundary layer.
limitation: The gridded product was examined for only a 24-hour period, and its temperature errors were close to, not clearly below, the best NUCAPS and GFS values.
tag: Evidence-backed gain
key_points: TAC-Thermo was trained on 68,887 global radiosonde and collocated satellite profiles (microwave, hyperspectral infrared and radio occultation) over two years. | Its largest temperature standard deviation error was 1.97°C at the 0.96 sigma level, compared with a minimum of 1.76°C for NUCAPS and 2.28°C for the GFS. | Dewpoint standard deviation errors were lower than NUCAPS and the GFS at all sigma levels in the analysis. | In a 24-hour test of the gridded product against radiosondes, 98% of sigma levels had an absolute median temperature error under 1.0°C. | For dewpoint temperature, 85.3% of sigma levels had an absolute median error under 1.0°C in the same test. | Gridded products give global feature maps at 0.25° resolution every 3 hours.
rundown: TAC-Thermo combines available satellite retrievals with numerical weather prediction output to upscale temperature and moisture profiles on sigma vertical coordinate levels from 1.0 to 0.2, capturing features up to about 12 km. It targets data-sparse regions where radiosonde launches are limited.

The system was validated against radiosonde soundings and benchmarked against NUCAPS and GFS analysis profiles in all climate regimes. Its temperature errors were comparable rather than uniformly better: its worst error sat near the best values of the other two datasets.

Gridded versions were examined over a 24-hour period against radiosonde data. The authors say the products support operational forecasting by identifying temperature and moisture inversions and cloud layers, and that the soundings support data assimilation in weather prediction models.
sources:
- peer_reviewed | Artificial Intelligence for the Earth Systems | https://doi.org/10.1175/aies-d-25-0092.1 | 2026-10-01
prev: 0000000000000000000000000000000000000000000000000000000000000000

SHA-256 hash

398d66c6c0433009d3210032d6901ffe8195c47e76e5da4663a969deecb668e1

Previous version’s hash

0000000000000000000000000000000000000000000000000000000000000000