Скачать или смотреть NASA & Harvard Low-Frequency Microwave Reconstruction for CubeSats, 37 GHz Ice-Storm Retrieval

Low-Frequency Microwave (LFM) Reconstruction is a joint Harvard University–NASA study led by Monica Klimczak that uses machine-learning to recover the 37 GHz passive-microwave brightness-temperature channel from higher-frequency radiometry, infrared imagery and Lightning Imaging Sensor data. The team trained on the 17-year archive from the Tropical Rainfall Measuring Mission (TRMM), whose Microwave Imager (TMI) provides collocated 10, 19, 37 and 85 GHz snapshots of convective storms, giving the models millions of labeled samples to learn from. https://gpm.nasa.gov/missions/trmm

Today’s Earth-science nanosatellites are shrinking to 6 U CubeSats, yet low-frequency microwave antennas demand apertures measured in metres. Without 37 GHz measurements, critical brightness-temperature depressions caused by ice scattering cannot be captured, threatening a gap in global ice-storm surveillance and inflating mission costs when only bus-sized platforms can fly such sensors.

At 37 GHz the wavelength (\~8 mm) penetrates well below the mixed-phase region, making this channel disproportionately sensitive to integrated ice water path and hail embryo density. Radiative-transfer analyses show that brightness-temperature combinations involving 19 GHz and 37 GHz explain about ten percent more variance in near-surface rainfall than the traditional 85 GHz channel, underscoring the value of low-frequency retrievals for both precipitation rate and electrification potential

The researchers fused TRMM’s higher-frequency TMI channels, the Visible and Infrared Scanner (VIRS) cloud-top temperatures, Precipitation Radar (PR) phase flags and lightning flash rates into a single multi-spectral tensor. These inputs were normalised and mapped to a 0.1° Earth-centred grid, then partitioned into three climatological regimes: sub-freezing scenes with positive PR, the full temperature span with positive PR, and an all-samples ensemble that removes hydrometeor filtering.

Four predictive frameworks were benchmarked: multilinear regression, random-forest regression with 500 trees, a shallow convolutional neural network similar to NASA’s baseline CNN, and an encoder-decoder U-Net with residual skip connections. Models were optimised against mean-squared error and evaluated on withheld TRMM orbits spanning multiple tropical seasons to ensure generalisation beyond a single storm morphology.

The U-Net reproduced azimuthal gradients and cold-core minima most faithfully, halving the root-mean-square error relative to the NASA baseline and reducing structural similarity loss by nearly 50 %. Visually, its brightness-temperature fields preserved the annular pattern of ice-rich eyewalls and the convective asymmetries in mesoscale convective complexes, while random-forest outputs tended to over-smooth and linear regression underestimated extremes.

Because the algorithm synthesises low-frequency content from lightweight sensors already flying on compact buses, a constellation of dozens to thousands of 6 U platforms could orbit in low-Earth-orbit swaths and still retrieve ice water path, graupel mass and lightning probability. The technique dovetails with upcoming missions such as the Atmosphere Observing System (AOS) and Investigation of Convective Updrafts (INCUS), where only high-frequency microwave and optical payloads are baselined

The Harvard–NASA team has delivered trained weights and reconstruction code to NASA Marshall for feasibility testing on on-orbit CubeSats. Planned next steps include lightning-constrained ensemble inference, data-fusion with Global Precipitation Measurement (GPM) radiometers, and insertion of synthetic 37 GHz fields into numerical weather-prediction assimilation pipelines to sharpen ice-storm nowcasts while keeping launch mass and power budgets at student-satellite levels.

https://www.nasa.gov/marshall-science...
https://espo.nasa.gov/sarp/content/On...

My full AI For Good 2025 video coverage is here:    • AI For Good 2025 video coverage  

This video was filmed using the DJI Pocket 3 ($669 at https://amzn.to/4aMpKIC using the dual wireless DJI Mic 2 microphones with the DJI lapel microphone https://amzn.to/3XIj3l8 ), watch all my DJI Pocket 3 videos here    • Filmed with DJI Pocket 3  

Join https://www.youtube.com/charbax/join for Early Access to my videos and to support my work, or you can Click the "Super Thanks" button below the video to send a highlighted comment under the video! Brands I film are welcome to support my work in this way 😁

Check out my video with Daylight Computer about their revolutionary Sunlight Readable Transflective LCD Display for Healthy Learning:    • Daylight Computer CEO interview: Healthy C...