The raindrop size distribution, or DSD, is one of the most fundamental quantities in cloud and precipitation physics. It describes how many raindrops exist in each size interval within a volume of air, and it underpins everything from the microphysics schemes inside numerical weather models to weather modification activities and studies of cloud-climate interactions. Because precipitation particles vary enormously in shape, density, size, orientation and spectral character, a radar operating at a single frequency simply cannot pin down these properties with confidence. The reflected signal conflates the effects of many drops with the effects of a few large ones, leaving scientists to guess at the underlying distribution.

Dual-frequency radars were invented to break that deadlock. The technique exploits a well-known divide in electromagnetic scattering: at longer wavelengths, such as the X band, radar waves scatter off raindrops in the Rayleigh regime, where the returned power depends only weakly on drop size. At shorter wavelengths, such as the Ka band, sufficiently large drops trigger Mie scattering, a resonance-sensitive process that responds strongly to particle diameter. The difference between the reflectivities measured at the two frequencies, expressed in decibels, is called the dual-frequency ratio, or DFR. As drops grow larger, the gap between the two scattering regimes widens and the DFR climbs, making it a sensitive diagnostic of the characteristic drop size.

The elegance of the DFR lies in what it ignores. Under the standard assumption of a normalized gamma size distribution with a fixed shape factor, the DFR depends only on the mass-weighted mean diameter, usually written as Dm, and is independent of the normalized intercept parameter Nw, which controls how many drops there are overall. That independence makes the DFR a remarkably clean probe of drop size. Previous work by Meneghini and colleagues showed that when the shape factor is set to 3, the retrieved distribution matches in situ observations closely, and most DFR-based retrieval research has centered on the Ku/Ka-band radar aboard the Global Precipitation Measurement satellite.

But there is a catch, and it is the central problem the new study attacks. When the DFR falls to zero or below, the mathematical relationship between the DFR and Dm folds back on itself: a single measured DFR value corresponds to two entirely different mean drop diameters. This dual-solution ambiguity is worst exactly where scientists most need clarity, in light precipitation dominated by small drops. The two conventional workarounds are both unsatisfying. One simply assumes the larger of the two possible diameters is correct, which introduces systematic errors in drizzle and light rain. The other discards all data with non-positive DFR values, throwing away a large fraction of the measurements. Liao and Meneghini proposed a generalized dual-frequency ratio that sidesteps the ambiguity, but it sacrifices the DFR’s prized independence from Nw. Other approaches based on the differential form of the radar equation help mainly in heavy precipitation with strong attenuation. Faced with these limitations, the GPM mission currently relies on a rain-rate-to-Dm relationship rather than the DFR itself.

Feng’s team proposed a deceptively simple fix: bring in a third frequency. Their framework combines X/Ka-band radar observations with a constraint from a hypothetical Ka/W-band pair, where the W band sits at even shorter wavelengths, around three millimeters. Because the Ka/W dual-frequency ratio responds to drop size in a different scattering regime, its own ambiguity zone is far narrower than that of the X/Ka pair. When the X/Ka ratio is ambiguous, the Ka/W ratio can usually tell which of the two candidate diameters is real. The team built the theoretical foundation using T-matrix scattering simulations, a rigorous method for computing electromagnetic scattering by nonspherical particles, applied to real raindrop size distributions measured by a two-dimensional video disdrometer at a meteorological site in Yangjiang, Guangdong Province.

The observational data came from an X/Ka dual-frequency radar developed by the 38th Research Institute of China Electronics Technology Group Corporation, operating in vertically pointing mode alongside the disdrometer across four precipitation events spanning 2022 to 2025. These included stratiform rain, convective storms, and extreme events with hourly rain rates exceeding 100 millimeters. Before any retrieval could happen, the team had to correct both radar channels for rain-induced attenuation, which is modest at X band but severe at Ka band. They used the differential attenuation between the two frequencies, constrained by theoretical DFR values from the scattering simulations, to prevent the overestimation that occurs when large drops produce Mie scattering at Ka band. After correction, the Ka-band reflectivities agreed closely with theoretical values computed from the ground-based disdrometer measurements, and the melting-layer bright band, nearly invisible in the attenuated raw data, reappeared clearly in the corrected fields.

The results are striking. Relying on the X/Ka dual-frequency ratio alone, only 35.38 percent of the dataset yielded a unique solution for the mean drop diameter. When the Ka/W constraint was incorporated, that proportion surged to 87.5 percent. The Ka/W ambiguity zone is confined to Ka-band reflectivities between roughly minus 20 and 20 dBZ, corresponding to diameters between 0.3 and 0.8 millimeters, whereas the X/Ka ambiguity stretches across reflectivities up to 45 dBZ and diameters up to 1.6 millimeters. The combined approach accurately retrieved drops larger than 0.5 millimeters, and the residual errors for smaller drops turn out to be nearly irrelevant, because rain rates stay below 0.6 millimeters per hour whenever the mean diameter is that small. Compared against two conventional methods, the new framework retrieved the mean diameter accurately across the entire size spectrum, while the first conventional method failed entirely for drops smaller than 1.3 millimeters and the second systematically overestimated small drops, which in turn corrupted the retrieved number concentrations through the inverse relationship between Dm and Nw.

The advantages were most pronounced in weak echoes below 20 dBZ, precisely the regime where conventional methods collapse and where light rain, drizzle and the early stages of precipitation formation live. The retrieved profiles also revealed physically meaningful structure: convective clouds showed larger and more variable drop diameters with strong vertical inhomogeneity, signaling intense coalescence and breakup within updrafts and downdrafts, while stratiform rain displayed uniform profiles punctuated by peaks near the melting layer. An independent validation on a precipitation event from 14 June 2025, not used in building the retrieval relationships, produced accuracy comparable to the training evaluation, with correlation coefficients exceeding 0.94 for diameter and 0.89 for the intercept parameter, suggesting the framework generalizes beyond the events it was built on.

The authors are careful about the caveats. No actual W-band radar was flown; the third-frequency reflectivities came from simulations, and the sensitivity experiments show the method tolerates a combined W-band reflectivity uncertainty of only about plus or minus 1 dB, a level the team argues is achievable with current ground-based W-band calibration practice and standard attenuation correction. The validation rests on a single site and a handful of events, which may represent an upper bound on real-world performance, and the vertically pointing geometry limits the use of polarimetric information. Still, as a proof of concept, the study charts a clear path toward three-frequency radar retrieval of raindrop size distributions. If future systems pair X, Ka and W band radars with the accuracy this work demands, forecasters and climate scientists could gain an unprecedented view of the smallest, most elusive drops, the ones where every rainstorm begins.

Subject of Research: Three-frequency radar retrieval of raindrop size distributions using combined dual-frequency ratio constraints

Article Title: The proof-of-concept for raindrop size distribution retrieval using combined dual-frequency radar DFR

Article References: Feng, L., Yang, J., Guo, Z., Sun, J., Sun, Y., Yang, H., & Hu, X. (2026). The proof-of-concept for raindrop size distribution retrieval using combined dual-frequency radar DFR. Atmospheric Measurement Techniques, 19(19), 6395-6414. https://doi.org/10.5194/amt-19-6395-2026

Image Credits: AI Generated

DOI: 10.5194/amt-19-6395-2026

Keywords: radar meteorology, raindrop size distribution, dual-frequency ratio, W-band radar, Mie scattering, T-matrix simulation, precipitation microphysics, attenuation correction, GPM mission, light rain, disdrometer, atmospheric measurement

This long-standing gap is the target of a new study published in Medical & Biological Engineering & Computing by Seyede Sajede Mousavi Eshkiki and Alireza Sadremomtaz of the University of Guilan in Iran. Rather than proposing yet another neural-network architecture, the pair set out to build and rigorously evaluate a complete, two-stage workflow that takes ordinary T1-weighted MRI of the brain and produces both a CT-like image and a fully attenuation-corrected PET scan. Crucially, they wrapped the entire pipeline in a graphical user interface, so that the deep learning machinery can be operated through inference without requiring users to write code. The work is an exercise in systems engineering as much as machine learning: the question was not whether a clever model could be designed, but whether a sequence of established components could be assembled into a workflow that behaves reliably enough to matter clinically.

The researchers drew their data from the CERMEP-IDB-MRXFDG cohort, a publicly available database of thirty-seven healthy adults who underwent fluorodeoxyglucose PET, T1-weighted and FLAIR MRI, and CT imaging of the brain. Because every subject had a real CT scan, the team could train their networks against genuine ground truth and then measure how closely the synthetic versions matched reality. They used subject-level five-fold cross-validation, meaning that in each round of testing the network had never seen the individual whose images it was correcting. This patient-level separation matters, because it prevents the model from effectively memorizing a person’s anatomy and inflating its apparent accuracy.

The first stage of the workflow is a two-dimensional U-Net, the workhorse convolutional network of biomedical image analysis, first introduced in 2015 and still favored for its efficiency and strong performance on relatively small datasets. Fed only a T1-weighted MRI volume, the network predicts a pseudo-CT: a synthetic image in Hounsfield units that approximates what a CT scanner would have measured for the same head. Once generated, this pseudo-CT is converted into an approximate linear attenuation coefficient map, the so-called mu-map that describes how strongly tissue attenuates the 511-keV photons emitted during positron annihilation. That mu-map, paired with the non-attenuation-corrected PET image, then serves as the input to a second U-Net, which synthesizes a predicted attenuation-corrected PET scan directly.

The logic of this second stage is subtle and worth unpacking. In principle, one could simply apply the mu-map mathematically to correct the PET data, as conventional reconstruction does. But the second network learns the mapping from uncorrected PET plus attenuation information to corrected PET, allowing it to absorb residual imperfections in the pseudo-CT. If the synthetic attenuation map slightly misestimates bone density, for example, the second network can still learn to produce an output consistent with properly corrected images, because it has seen many paired examples of what uncorrected and corrected PET look like for the same brain. The two networks therefore act as a chain, with the first converting MRI into attenuation-relevant anatomy and the second converting that anatomy plus raw PET into a quantitatively usable scan.

The performance figures tell a encouraging story with a familiar caveat. The pseudo-CT model achieved a whole-image mean absolute error of 72.26 plus or minus 20.11 Hounsfield units and a structural similarity index of 0.881 plus or minus 0.024 against the real CT scans. The authors note that the residual error was concentrated in bone-dominated regions, which is exactly where MRI-to-CT synthesis has always struggled. Bone produces almost no signal on conventional MRI, so the network must essentially infer cortical skeleton from the shape of the brain and surrounding tissues. Decades of attenuation-correction research, from segmentation-based methods to atlas-based approaches and zero-echo-time MRI sequences, have wrestled with the same limitation, and deep learning approaches inherit it to varying degrees.

Where the workflow truly shines is in the PET correction itself. Compared with the raw, non-attenuation-corrected PET images, the predicted attenuation-corrected PET achieved a 76.02 percent lower normalized mean absolute error and a 73.33 percent lower normalized root mean square error relative to the reference attenuation-corrected PET. In other words, the two-stage pipeline recovered roughly three quarters of the quantitative error introduced by ignoring attenuation. For brain imaging, where attenuation is less severe than in the thorax or abdomen but still clinically significant, that level of recovery brings synthetic correction into the same conversation as established vendor solutions.

The exploratory standardized uptake value analysis adds a dose of realism. Standardized uptake values are the currency of quantitative PET, and the workflow delivered a mean percentage error of 8.66 percent for SUVmean but 22.08 percent for SUVmax, with greater variability across subjects in the maximum metric. This pattern is well understood in the field: SUVmean averages over a region and tends to forgive localized errors, while SUVmax hinges on a single hottest voxel and is exquisitely sensitive to small misestimates of attenuation, particularly near bone. The authors are appropriately measured, framing these results as supporting internal feasibility on the CERMEP dataset while emphasizing that nested cross-validation and external validation on independent cohorts remain necessary before any clinical claim can be made.

What makes the study notable beyond its numbers is its orientation toward usability and transparency. The graphical user interface lowers the barrier for imaging scientists and clinicians who lack programming expertise, echoing earlier efforts such as the Deep Learning Application Engine to bring neural networks out of notebooks and into the imaging suite. Just as importantly, the team has released the source code, trained models, cross-validation fold assignments, configuration files, and evaluation outputs through the Zenodo repository, allowing other groups to reproduce the results, stress-test the workflow on their own data, and extend it. In a subfield where reproducibility has often lagged behind headline accuracy figures, that openness is a meaningful contribution in its own right.

The broader significance lies in what accessible attenuation correction could unlock. PET/MRI offers simultaneous acquisition of molecular and anatomical information with a fraction of the radiation dose of PET/CT, making it attractive for pediatric imaging, longitudinal neuroscience studies, and repeated oncological follow-up. If a workflow like this one, running on nothing more exotic than a standard T1-weighted scan, can deliver dependable attenuation correction without dedicated sequences or vendor-proprietary tools, it could lower the cost and complexity of quantitative PET/MRI for research centers worldwide. The caveats stand: thirty-seven healthy subjects, a single tracer, and no external cohort yet. But as a carefully documented proof of feasibility, the study charts a practical path from the physics problem that has shadowed PET/MRI since its inception toward a solution that any lab with a GPU and a GUI could pick up and run.

Subject of Research: Deep learning-based attenuation correction for brain PET/MRI using pseudo-CT synthesis

Article Title: A two-stage PET/MRI attenuation-correction workflow for pseudo-CT and AC PET synthesis with GUI-based inference

Article References: Eshkiki, S. S. M., & Sadremomtaz, A. (2026). A two-stage PET/MRI attenuation-correction workflow for pseudo-CT and AC PET synthesis with GUI-based inference. Medical & Biological Engineering & Computing. https://doi.org/10.1007/s11517-026-03685-y

Image Credits: AI Generated

DOI: 10.1007/s11517-026-03685-y

Keywords: PET/MRI, attenuation correction, pseudo-CT, U-Net, deep learning, mu-map, NAC PET, SUV, CERMEP-IDB-MRXFDG, medical imaging, radiotracer imaging, quantitative PET