Huan Li, PhD

Air motion and fall speed in EarthCARE Doppler observations

A ground-based dual-Doppler reference along the satellite track, used to test three satellite-only decompositions

2025 to present, independent research

Reference NEXRAD dual-Doppler winds along the EarthCARE track VALIDATED ESA C-CD, JAXA CPR_CLP and DPR-based fall speeds

EarthCARE's cloud radar is the first Doppler radar in space. Its nadir beam measures air motion and hydrometeor fall speed together, and cannot separate them, and none of the three published satellite-only fall speed (ESA C-CD, JAXA CPR_CLP, DPR-PSD-assumed decomposition) had been tested against an independent measurement.

I built a ground-based dual-Doppler reference for vertical air motion along eight CPR tracks (PyDDA variational and explicit CEDRIC-type methods) from pairs of NEXRAD radars, cross-checked by two independent solvers: accurate to 0.65 m s⁻¹ in stratiform cloud and 1–2.3 m s⁻¹ in convection. And used it to evaluate the ESA C-CD, JAXA CPR_CLP and DPR-PSD-based fall-speed decompositions.

The ESA C-CD sedimentation velocity agrees with the reference-corrected fall speed to 0.45 m s⁻¹ where |w| < 0.5 m s⁻¹ and retains the air motion up to |w| ≈ 1.5 m s⁻¹. The JAXA CPR_CLP product assigns cloud-particle fall speeds 3 m s⁻¹ too small in rain and 0.5 m s⁻¹ in ice, so its air velocity is 1.1 m s⁻¹ too high and uncorrelated with the reference. The PSD-assumed decomposition is unbiased inside its stated snow range but biased by +0.7 to +2 m s⁻¹ above it and, on weaker evidence, by +3 to +4 m s⁻¹ in light rain at the DPR sensitivity floor. A disdrometer-derived W-band fall speed at the ARM Southern Great Plains site agrees with the C-CD sedimentation velocity to 0.3 m s⁻¹.

Li, H., 2026: Ground-based dual-Doppler validation of vertical air motion and hydrometeor fall speed in EarthCARE CPR Doppler observations. in preparation for Atmospheric Measurement Techniques.

Marine warm-cloud drizzle life cycle

An EarthCARE Doppler view, validated at ARM ENA: generation → sub-cloud evaporation → surface fallout or virga

2025 to present, College Park, MD

Reference ARM ENA cloud radar, radiometer and 1,585 radiosondes Tested EarthCARE CPR overpasses

EarthCARE carries the first Doppler cloud radar in space, but it cannot see the lowest ~0.5 km above the sea, where most marine drizzle evaporates. With 201 overpasses of the ARM Eastern North Atlantic site (Dec 2024–May 2026), checked against the ground radar and 1,585 radiosondes, and sorted overpasses into four cloud regimes by an ω₅₀₀–cold-air-outbreak–EIS decision tree.

I show that the satellite and the ground site see different halves of the drizzle life cycle. Cloud-top height and liquid water path, both observable from orbit, control how much drizzle forms; cloud-base drop size and sub-cloud depth, visible only from the ground, decide whether it reaches the sea. Satellite-only drizzle climatologies therefore map where drizzle forms, not where it lands. In warm liquid cloud, EarthCARE's air-motion-corrected fall speeds from all four weather regimes collapse onto a single power law, V_t = 1.58 Z_e^0.205.

Drizzle generation turns out to be set by cloud-top height and LWP (β = 0.51, 0.34; R² = 0.38), both visible from space; its fate, mostly virga, by cloud-base drop size and sub-cloud depth below the CPR clutter floor, mostly evaporation before reaching the sea, is decided below the satellite's clutter floor, where only the ground site sees, so that satellite and ground site together close a drizzle life cycle neither observes alone.

First-author manuscript submitted to Atmospheric Chemistry and Physics.

A drifting cloud radar: calibrating the ARM ENA Ka-band radar with ground disdrometers

2018 to 2023,Visiting Assistant Research Scientist, ESSIC, University of Maryland

DATAARM Calibrated Ka-band Radar

The ARM Eastern North Atlantic radar has recorded marine low clouds and drizzle since 2015, but its long-term reflectivity files are labeled "not yet calibrated." Before using the record for drizzle science, I checked how far off the radar is and whether the error changes. I compared the radar's reflectivity at ~240 m with the Ka-band reflectivity computed from raindrops counted by two independent surface disdrometers, using light rain (0–20 dBZ) over 1,629 days (~380,000 one-minute pairs).

The radar's reading drifted steadily low, by about 2 dB per year: near 0 dB in late 2015 and 9–10 dB low by the end of 2019. The two disdrometers agree month by month, so the drift is in the radar. The estimates reproduce the published calibrations of Kollias et al. (2019) to within 1–2 dB and cover April 2018 to December 2019, a period with no published calibration.

Warm-rain detection from CloudSat and VIIRS

2018 to 2023,Visiting Assistant Research Scientist, ESSIC, University of Maryland

DATACloudSat, VIIRSDetect Warm Rain

Warm rain dominates precipitation occurrence over the subtropical oceans but is largely missed by infrared and passive-microwave retrievals. I built a pipeline that collocates CloudSat radar with MODIS and VIIRS (~10⁵ rays matched within ±5 min）and tested whether retrieved cloud properties from visible and near-infrared imagery can detect warm rain.

Revisiting Chen et al. (2011) with MODIS, I found liquid water path (LWP) to be the best single predictor of warm rain (HSS 0.61, threshold ≈ 0.22 mm), outperforming cloud-top temperature. Adding effective radius (r_e) helps further: at the same LWP, clouds with larger droplets rain more readily (Figure).

I then tested whether this approach carries over to VIIRS. A rain/no-rain rule in LWP–r_e space learned from 2008 MODIS transfers to 2016–17 VIIRS without refitting: HSS 0.52, compared with 0.49 for MODIS on the same collocated rays. Applied pixel by pixel, the rule produces full-swath VIIRS warm-rain maps, extending CloudSat's narrow curtain to the full imager swath.

Retrieval of atmospheric water vapor from SEVIRI infrared

2017, Visiting Scholar, CIMSS, University of Wisconsin–Madison

Retrieval Atmospheric Water Vapor Radiative Transfer PFAAST

I contributed to adapting the GOES Sounder atmospheric profile algorithm to the SEVIRI imager to retrieve atmospheric water vapor: a regression trained on a database of atmospheric profiles, then a physical retrieval with the PFAAST radiative transfer model.

Modeled and observed brightness temperatures differ slightly, so the work included a bias correction for five infrared bands derived from matched profiles, the same matchup analysis that underlies radiometer intercalibration.

Error structure of satellite precipitation over Tibetan Plateau in Winter

2015 to 2016, Postdoctoral Fellow, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences; revisited in 2026 with the current product versions

PI Grant NSFCTested IMERG, TMPA, CMORPH and GSMaP Case the snow-covered winter Plateau

PI Grant China Postdoctoral Science Foundation Merged Radar-satellite-gauge precipitation

The Tibetan Plateau in winter is a snow-covered, gauge-sparse surface where the retrievals are known to struggle. Four gauge-adjusted products (IMERG V07 Final, TMPA 3B43, CMORPH CDR, GSMaP Gauge) are put on one 0.25° grid and compared with their multi-product mean, GPCC gauges, PERSIANN-CDR(infrared-only)and their own satellite-only versions. Cells above 3000 m are split by ERA5-Land snow cover into snow-free, intermittent and stable-snow classes that sit at the same mean elevation, so any difference between classes is a surface effect.

Over the winter Plateau (cells above 3000 m, December 2014 to February 2015) the four products lie between 31 % below and 32 % above their common mean, with a median spread of 56 % against 26 % in summer. Each departs for a different reason. IMERG sees almost nothing (0.07 mm/day, 1 % wet days) and the gauge ratio multiplies it by 3.8: the Plateau value is GPCC's. TMPA TMPA sees too much: only 12 % of its amount is microwave, the rest is infrared over cold ground; the gauges halve it and it is still 32 % high. CMORPH is the most surface-sensitive, +13 % over bare ground and −27 % over stable snow, as microwave screening over snow propagates through the morphing. GSMaP is the driest (−31 %) and effectively unadjusted, its gauge calibration acting weakly where gauges are sparse.

What this means for retrieval. The gauge adjustment equalizes the products' totals but not their errors: at constant elevation all four still over-report on cold bare ground and under-report on snow. What the retrievals lose is the microwave signal, screened or down-weighted where frozen ground and snow look like precipitation to the radiometer, and infrared morphing cannot put it back. Improving winter precipitation over land therefore has to happen inside the retrieval, with a surface-dependent prior, the high-frequency channels, and reanalysis temperature and snow depth as ancillary inputs, rather than in the adjustment afterwards.

Data: IMERG V07B, TMPA 3B43/3B42/3B42RT V7 (NASA GES DISC); CMORPH V1.0 CDR, PERSIANN-CDR, ETOPO 2022 and GHCN-Daily (NOAA NCEI); GSMaP V04B (JAXA); GPCC Full Data Monthly v2022 (DWD); ERA5-Land(Copernicus CDS).

Variational merging of hourly satellite precipitation and rain gauge over China

2010 to 2014, Ph.D. Institute of Remote Sensing Applications, Chinese Academy of Sciences

2012 to 2013, Exchage scholarship. Naional Weather Center, University of Oklahoma

I developed the variational scheme that merges hourly rain-gauge observations with NOAA CMORPH satellite estimates into a 0.1 latitude/longitude precipitation analysis over mainland China, working with Dr. Jidong Gao at NOAA's National Severe Storms Laboratory. The scheme minimizes a cost function with a recursive-filter background term that spreads gauge information to nearby grid points, with the gradient obtained by the adjoint technique and the minimization by a quasi-Newton method; I wrote the implementation in Fortran.（Li, 2015, JGR-Atmos）

Cross-validation against withheld gauges showed that the analysis improves on satellite estimates, most of all where gauges are sparse. Satellite estimates capture heavy and extreme rainfall well but underestimate light rain; the merged analysis corrects the light-rain deficit where gauges exist and keeps the satellite's extreme-event rain structure where they do not (Li, 2016, IGARSS).

Li et al. (2015), Journal of Geophysical Research: Atmospheres, doi:10.1002/2015JD023710. With Y. Hong, P. Xie (NOAA CPC), J. Gao (NOAA NSSL), Z. Niu, P. Kirstetter and B. Yong.

Li, H (2016): Merged of gauge-satellite precipitation benefits to data-sparse regions and to extreme rainfall events. Proc. 2016 IEEE Int. Geosci. Remote Sens. Symp. (IGARSS), Beijing, doi:10.1109/IGARSS.2016.7729149.

Gauge-based daily rainfall products in China

2011, Research Intern, National Meteorological Information Center, China Meteorological Administration

Gauge-based daily rainfall products over China were constructed by optimal interpolation of 2,419 gauges on a 0.25 latitude/longitude grid for 1951–2010.

The stations measure rainfall with siphon or tipping-bucket gauges, record automatically, and pass quality control. The precipitation climatology used by the interpolation is adjusted with the Parameter-Elevation Regressions on Independent Slopes Model (PRISM) to correct the bias caused by orographic effects.

The analysis captures daily precipitation well over eastern China, where stations are relatively dense, while precipitation patterns may be distorted over western China, where analyzed values are interpolated from reports of distant stations.