Hao Huang

Single-transmit, dual-receive millimeter-wave radars can be attractive for compact and scanning precipitation observations, but they do not necessarily provide the conventional horizontal/vertical dual-polarization variables used to constrain raindrop size distributions. This study evaluates whether a fixed slant-linear basis can turn the weak backscattering anisotropy of oblate raindrops into a useful liquid-rain observable. In the proposed geometry, a +45° transmitted polarization and co-/cross-slant receive channels project the difference between horizontal and vertical co-polar backscattering amplitudes into an orthogonal-slant return. We use explicit T-matrix scattering calculations, normalized-gamma and observed drop size distributions, and controlled neural-regression diagnostics to quantify the information content and observability of the resulting slant-linear depolarization ratio (SLDR). For the baseline Ka-band gamma ensemble, the reflectivity (Z₊)–SLDR space separates characteristic drop size and concentration that are ambiguous under reflectivity alone. With 1 dB SLDR uncertainty, adding SLDR reduces the independent-test RMSE of mass-weighted mean diameter (D_m) from 0.672 to 0.243 mm and reduces the RMSE of log-transformed rain rate (log₁₀R) from 0.163 to 0.055. The same sign of improvement persists for Nanjing 2DVD, EPFL HyMeX, NASA IFLOODS, and variable-shape gamma checks. Sensitivity tests show that random SLDR uncertainty degrades retrievals gradually, whereas relative co-/cross-slant gain bias and polarization leakage define stronger practical constraints. Through the controlled forward-simulation information-content tests, SLDR is identified as a physically distinct measurement coordinate for liquid-rain microphysics. The calibration and detectability conditions are also defined for a field test with suitable single-transmit, dual-receive radars.

Single-transmit, dual-receive millimeter-wave radars can be attractive for compact and scanning precipitation observations, but they do not necessarily provide the conventional horizontal/vertical dual-polarization variables used to constrain raindrop size distributions. This study evaluates whether a fixed slant-linear basis can turn the weak backscattering anisotropy of oblate raindrops into a useful liquid-rain observable. In the proposed geometry, a +45° transmitted polarization and co-/cross-slant receive channels project the difference between horizontal and vertical co-polar backscattering amplitudes into an orthogonal-slant return. We use explicit T-matrix scattering calculations, normalized-gamma and observed drop size distributions, and controlled neural-regression diagnostics to quantify the information content and observability of the resulting slant-linear depolarization ratio (SLDR). For the baseline Ka-band gamma ensemble, the reflectivity (Z₊)–SLDR space separates characteristic drop size and concentration that are ambiguous under reflectivity alone. With 1 dB SLDR uncertainty, adding SLDR reduces the independent-test RMSE of mass-weighted mean diameter (D_m) from 0.672 to 0.243 mm and reduces the RMSE of log-transformed rain rate (log₁₀R) from 0.163 to 0.055. The same sign of improvement persists for Nanjing 2DVD, EPFL HyMeX, NASA IFLOODS, and variable-shape gamma checks. Sensitivity tests show that random SLDR uncertainty degrades retrievals gradually, whereas relative co-/cross-slant gain bias and polarization leakage define stronger practical constraints. Through the controlled forward-simulation information-content tests, SLDR is identified as a physically distinct measurement coordinate for liquid-rain microphysics. The calibration and detectability conditions are also defined for a field test with suitable single-transmit, dual-receive radars.

Radar data assimilation (DA) alone remains insufficient for precipitation forecasts due to the lack of thermodynamic information in the pre‐storm environment. This study investigates the assimilation of dense automatic weather station (AWS) observations from China's mesoscale network alongside radar during an extreme rainfall event caused by the remnant vortex of typhoon Haikui. Optimal sequence and scale representation of dense AWS observations within the operational two‐step DA framework are examined. Results demonstrate that adding regional AWS data to radar and national‐level observations significantly refines surface wind, temperature, and humidity analyses, directly enhancing precipitation forecasts. Notably, assimilating synoptic‐scale information followed by radar outperforms the reverse approach, as initial large‐scale increments stabilize subsequent convective‐scale updates, including stronger maritime airflow, enhanced low‐level convergence and atmospheric instability. Furthermore, despite ∼10 km spacing, AWS data function optimally as synoptic‐scale observations rather than convective‐scale, as the single‐layer temperature and humidity increments with very small decorrelation scales have limited impact in NWP models.

Accurate identification of non-precipitation echoes (NPEs) in weather radar observations requires effective use of polarimetric signatures together with spatiotemporal structure. Here we present a unified deep-learning framework to quantify the independent and synergistic contributions of model architecture, dual-polarization variables, and short-term temporal evolution to NPE identification. Using data from the Guangzhou S-band dual-polarization radar, we conduct controlled comparative experiments with two representative architectures: a pointwise multilayer perceptron (MLP) and a Transformer-based Swin U-Net that explicitly learns spatial context. We further perform ablation experiments across single- versus dual-polarization inputs and single-volume versus two-volume inputs. Results show that architecture-driven spatial-context learning is the dominant factor: Swin U-Net consistently outperforms the pointwise MLP under all input settings. On a high-confidence test subset, for example, the Critical Success Index (CSI) increases from 0.887 for the dual-polarization MLP to 0.950 for the dual-polarization Swin U-Net. Dual-polarization variables provide essential microphysical constraints and substantially improve class separability, particularly for pointwise classifiers. Incorporating two consecutive volumes further improves performance by capturing short-term echo evolution, with larger gains for the MLP than for Swin U-Net. The best-performing configuration, combining Swin U-Net with dual-polarization and two-volume inputs, achieves a CSI of 0.953 on the high-confidence test subset. Notably, the Swin U-Net using only the reflectivity factor (Z_H) as input retains strong skill (CSI = 0.927), indicating that spatial-context learning can partially compensate for missing polarimetry and thus providing a practical pathway for quality control of legacy single-polarization archives.

Accurate identification of non-precipitation echoes (NPEs) in weather radar observations requires effective use of polarimetric signatures together with spatiotemporal structure. Here we present a unified deep-learning framework to quantify the independent and synergistic contributions of model architecture, dual-polarization variables, and short-term temporal evolution to NPE identification. Using data from the Guangzhou S-band dual-polarization radar, we conduct controlled comparative experiments with two representative architectures: a pointwise multilayer perceptron (MLP) and a Transformer-based Swin U-Net that explicitly learns spatial context. We further perform ablation experiments across single- versus dual-polarization inputs and single-volume versus two-volume inputs. Results show that architecture-driven spatial-context learning is the dominant factor: Swin U-Net consistently outperforms the pointwise MLP under all input settings. On a high-confidence test subset, for example, the Critical Success Index (CSI) increases from 0.887 for the dual-polarization MLP to 0.950 for the dual-polarization Swin U-Net. Dual-polarization variables provide essential microphysical constraints and substantially improve class separability, particularly for pointwise classifiers. Incorporating two consecutive volumes further improves performance by capturing short-term echo evolution, with larger gains for the MLP than for Swin U-Net. The best-performing configuration, combining Swin U-Net with dual-polarization and two-volume inputs, achieves a CSI of 0.953 on the high-confidence test subset. Notably, the Swin U-Net using only the reflectivity factor (Z_H) as input retains strong skill (CSI = 0.927), indicating that spatial-context learning can partially compensate for missing polarimetry and thus providing a practical pathway for quality control of legacy single-polarization archives.

Accurate identification of non-precipitation echoes (NPEs) in weather radar observations requires effective use of polarimetric signatures together with spatiotemporal structure. Here we present a unified deep-learning framework to quantify the independent and synergistic contributions of model architecture, dual-polarization variables, and short-term temporal evolution to NPE identification. Using data from the Guangzhou S-band dual-polarization radar, we conduct controlled comparative experiments with two representative architectures: a pointwise multilayer perceptron (MLP) and a Transformer-based Swin U-Net that explicitly learns spatial context. We further perform ablation experiments across single- versus dual-polarization inputs and single-volume versus two-volume inputs. Results show that architecture-driven spatial-context learning is the dominant factor: Swin U-Net consistently outperforms the pointwise MLP under all input settings. On a high-confidence test subset, for example, the Critical Success Index (CSI) increases from 0.887 for the dual-polarization MLP to 0.950 for the dual-polarization Swin U-Net. Dual-polarization variables provide essential microphysical constraints and substantially improve class separability, particularly for pointwise classifiers. Incorporating two consecutive volumes further improves performance by capturing short-term echo evolution, with larger gains for the MLP than for Swin U-Net. The best-performing configuration, combining Swin U-Net with dual-polarization and two-volume inputs, achieves a CSI of 0.953 on the high-confidence test subset. Notably, the Swin U-Net using only the reflectivity factor (Z_H) as input retains strong skill (CSI = 0.927), indicating that spatial-context learning can partially compensate for missing polarimetry and thus providing a practical pathway for quality control of legacy single-polarization archives.

Accurate identification of non-precipitation echoes (NPEs) in weather radar observations requires effective use of polarimetric signatures together with spatiotemporal structure. Here we present a unified deep-learning framework to quantify the independent and synergistic contributions of model architecture, dual-polarization variables, and short-term temporal evolution to NPE identification. Using data from the Guangzhou S-band dual-polarization radar, we conduct controlled comparative experiments with two representative architectures: a pointwise multilayer perceptron (MLP) and a Transformer-based Swin U-Net that explicitly learns spatial context. We further perform ablation experiments across single- versus dual-polarization inputs and single-volume versus two-volume inputs. Results show that architecture-driven spatial-context learning is the dominant factor: Swin U-Net consistently outperforms the pointwise MLP under all input settings. On a high-confidence test subset, for example, the Critical Success Index (CSI) increases from 0.887 for the dual-polarization MLP to 0.950 for the dual-polarization Swin U-Net. Dual-polarization variables provide essential microphysical constraints and substantially improve class separability, particularly for pointwise classifiers. Incorporating two consecutive volumes further improves performance by capturing short-term echo evolution, with larger gains for the MLP than for Swin U-Net. The best-performing configuration, combining Swin U-Net with dual-polarization and two-volume inputs, achieves a CSI of 0.953 on the high-confidence test subset. Notably, the Swin U-Net using only the reflectivity factor (Z_H) as input retains strong skill (CSI = 0.927), indicating that spatial-context learning can partially compensate for missing polarimetry and thus providing a practical pathway for quality control of legacy single-polarization archives.

Single-transmit, dual-receive millimeter-wave radars can be attractive for compact and scanning precipitation observations, but they do not necessarily provide the conventional horizontal/vertical dual-polarization variables used to constrain raindrop size distributions. This study evaluates whether a fixed slant-linear basis can turn the weak backscattering anisotropy of oblate raindrops into a useful liquid-rain observable. In the proposed geometry, a +45° transmitted polarization and co-/cross-slant receive channels project the difference between horizontal and vertical co-polar backscattering amplitudes into an orthogonal-slant return. We use explicit T-matrix scattering calculations, normalized-gamma and observed drop size distributions, and controlled neural-regression diagnostics to quantify the information content and observability of the resulting slant-linear depolarization ratio (SLDR). For the baseline Ka-band gamma ensemble, the reflectivity (Z+)–SLDR space separates characteristic drop size and concentration that are ambiguous under reflectivity alone. With 1 dB SLDR uncertainty, adding SLDR reduces the independent-test RMSE of mass-weighted mean diameter (Dm) from 0.672 to 0.243 mm and reduces the RMSE of log-transformed rain rate (log10R) from 0.163 to 0.055. The same sign of improvement persists for Nanjing 2DVD, EPFL HyMeX, NASA IFLOODS, and variable-shape gamma checks. Sensitivity tests show that random SLDR uncertainty degrades retrievals gradually, whereas relative co-/cross-slant gain bias and polarization leakage define stronger practical constraints. Through the controlled forward-simulation information-content tests, SLDR is identified as a physically distinct measurement coordinate for liquid-rain microphysics. The calibration and detectability conditions are also defined for a field test with suitable single-transmit, dual-receive radars.

Shear‐relative, quadrant‐scale differences in precipitation microphysics between intensifying tropical cyclones (ITCs) and weakening tropical cyclones (WTCs) are investigated using Ku/Ka‐band observations from the Global Precipitation Measurement (GPM) Dual‐frequency Precipitation Radar (DPR) over the western North Pacific during 2014–2024. The results indicate that ITCs exhibit a higher fraction of convective pixels and statistically significantly larger near‐surface drop sizes compared with WTCs, despite their weaker instantaneous storm intensities. Notably, the downshear right (DR) quadrant of ITCs acts as a convective initiation zone, exhibiting the highest convective fraction yet the lowest mean storm echo top height, with microphysical signatures of more active drop breakup. In contrast, the downshear left (DL) quadrant emerges as particularly critical for intensification, showing the greatest enhancement in precipitation properties, especially in rain rate and storm echo top height, while the corresponding enhancement for WTCs is comparatively subdued. Additionally, the presence of significantly greater Z _e and larger drop sizes observed above the melting layer in the DL quadrant of ITCs indicates vigorous ice‐phase processes (e.g., riming and aggregation), while in WTCs, rapid Z _e increases are confined to a shallow layer, consistent with weaker updrafts that limit vertical ice growth. These results suggest that monitoring storm echo top heights and drop‐size evolution, particularly within the DR and DL quadrants, likely provide valuable insight towards early‐warning indicators of TC intensification and offer observational constraints for improving TC intensity forecasting.