Vertical motions in clouds from EarthCare satellite and a global storm-resolving modeling Authors: Woosub Roh, Masaki Satoh, Shuhei Matsugishi, Shunsuke Aoki, Takuji Kubota, Hajime Okamoto Published: 2025 ABSTRACT AbstractUnderstanding vertical motions in clouds is essential for improving weather and climate models but remains challenging due to limited observations. The newly launched EarthCARE satellite with a Doppler radar now provides global measurements of how fast particles fall through clouds, offering a way to estimate vertical air movements. In this study, we provide an initial assessment of EarthCARE’s Doppler radar observations by comparing them with high-resolution global simulations using NICAM and a satellite simulator. We analyze two representative cases of frontal and convective precipitation to examine how vertical motions influence the relationship between radar reflectivity and the fall speed of ice hydrometeors, and to identify potential discrepancies between observations and simulations. The results show that the simulations reproduce the overall vertical structures observed by EarthCARE, while differences in Doppler velocity, particularly in convective regions, indicate biases in simulations and uncertainties in the observations. Nevertheless, these complementary perspectives reveal EarthCARE’s capability to provide physically consistent information on hydrometeor vertical motions and demonstrate its potential to constrain and refine the representation of cloud microphysics and dynamics in global weather and climate simulations.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-32256-8.Subject terms: Climate sciences, Environmental sciences Introduction Clouds and their associated vertical motions remain among the most ambiguous and challenging components in climate models, contributing substantially to uncertainties in both climate forcing and climate sensitivity1. Reducing such uncertainties requires advances in observational technologies, improved understanding of convective vertical velocities, and more physically based numerical simulations. The successful launch of the Earth Cloud Aerosol and Radiation Explorer (EarthCARE) satellite2,3 in May 2024 marked a major milestone. Equipped with the Cloud Profiling Radar (CPR), EarthCARE provides unprecedented global observations of radar reflectivity and Doppler velocity, enabling detailed analyses of cloud particle sizes, falling speeds, and vertical air motions that are essential for understanding extreme precipitation and climate sensitivity. These observations are used to evaluate and improve global storm-resolving models (GSRMs) such as NICAM by identifying and reducing systematic biases, thereby contributing to reduced uncertainties in weather and climate forecasts. In parallel, global storm-resolving models (GSRMs) have become essential tools in atmospheric science4–6. With horizontal mesh sizes near a few km, they can resolve mesoscale convective systems (MCSs) and some of their internal dynamics7. A key advantage of GSRMs is their ability to reproduce vertical air motions without relying on deep cumulus parameterization, while also incorporating more detailed representations of cloud microphysical processes compared to conventional global circulation models. For this study, we use the Nonhydrostatic Icosahedral Atmospheric Model (NICAM)8–10 with a horizontal resolution of 870 m. To complement EarthCARE, NICAM simulations at this resolution were performed globally for EarthCARE’s initial observational period (June 17–18, 2024). These simulations provide a more detailed representation of cloud systems compared to conventional GCMs, although further refinement is needed to fully resolve vertical air motions, revealing current GSRM limitations and future development needs11. Measuring Doppler velocity from space involves challenges due to satellite motion (~ 7 km s− 1), vibrations, and other uncertainties such as multiple scattering, non-uniform beam filling effects, and positioning errors12–15. Differentiating between hydrometeor terminal velocities and vertical air motions in Doppler data is difficult. While ground-based profilers have successfully separated these components16–18, applying similar techniques from space is more challenging. Thus, validating EarthCARE Doppler data using complementary observations and simulations is crucial. Numerical models provide comprehensive insights into cloud structures and can fill gaps in satellite measurements. A global view of precipitation at 03 UTC on June 18, 2024, is presented in Fig. 1 using the GSMaP satellite observations and the NICAM simulation (see Methodology). The NICAM simulation captures synoptic-scale precipitation patterns worldwide, including MCSs in the tropical region and extratropical cyclone systems in the mid-latitudes. Both GSMaP and NICAM present the internal precipitation structure of MCSs, representing precipitation associated with convective cores. However, the precise location of precipitation does not exactly match, and NICAM exhibits a higher abundance of fine-scale precipitation compared to GSMaP. Since GSMaP relies on precipitation radar and passive microwave channels, it is less sensitive to weak precipitation. By incorporating EarthCARE observations, it becomes possible to distinguish whether the discrepancies arise from model biases or from observational sensitivity limitations. Horizontal distribution of precipitation from observations with GSMaP (a) and the NICAM simulation (b). The NICAM data have been coarsened to 0.1° resolution to match the GSMaP grid. Contours are precipitation rates in mm hr-1. The maps were generated using Python Matplotlib (version 3.8.4; https://matplotlib.org/) and Cartopy (version 0.23.0; https://cartopy.readthedocs.io/stable/), and the two panels were combined using Inkscape (version 1.4.2; https://inkscape.org/).” In this study, we explore the capabilities of the newly launched EarthCARE satellite, particularly its CPR, which provides global observations of radar reflectivity and Doppler velocity. These measurements offer new opportunities to infer the terminal velocities of hydrometeors and to estimate vertical air motions within clouds. Using NICAM, a high-resolution global model, and a satellite simulator19,20 (Joint Simulator for Satellite Sensors, J-Sim; see Methodology), we generate simulated radar signals to directly compare with EarthCARE/CPR observations. This comparison enables a detailed evaluation of vertical structures in convective and frontal cloud systems. The simulations broadly reproduce key observed features, including strong upward motion in tropical convection. However, differences in Doppler velocity suggest systematic biases, such as an overestimation of hydrometeor fall speeds in the model. These findings highlight the value of integrating satellite data and high-resolution models to improve the representation of cloud dynamics and vertical transport in numerical simulations. Results A cold front case The left panel of Fig. 2 shows cross sections of EarthCARE/CPR observations and CPR-like signals from the NICAM simulation using J-Sim for 1:08–1:20 UTC on June 18, 2024 (Case 1). Observed and simulated radar reflectivity (Fig. 2a, c) and Doppler velocity (Fig. 2b, d) are compared across a cold front system of an extratropical cyclone. The radar reflectivity shows stronger signals in the middle level of convective clouds near 50°S. The Doppler velocity distinctly illustrates a faster falling speed in the lower levels and a slower downward motion above. Note that in this study, we use a sign convention where positive Doppler velocity indicates upward motion from the ground, while negative values represent downward motion. A cold front forms when a mass of relatively colder, drier air advances into a region occupied by warmer, moister air. This interaction creates a steeply sloping boundary beneath the warmer air, forcing it to ascend as the front progresses. In this case, the boundary near 52°S separates precipitation associated with cold air, characterized by snow in the higher latitudes, from precipitation linked to warm air, characterized by rain in the lower latitudes. The radar reflectivity reveals a stronger bright band signal at approximately 2–3 km altitude in the latitude range of 46–50°S, which extends closer to the ground around 52°S, consistent with the cold front structure. Correspondingly, the Doppler velocity cross-section clearly distinguishes between regions of fast falling hydrometeors (blue, primarily rain) and slow falling particles (green, primarily snow or ice), reflecting differences in terminal velocity. The boundary between faster falling speeds and slower downward motion approximately corresponds to the 0 °C isotherm, showing the vertical and horizontal thermal structure of the front. Radar reflectivity (a, c) and Doppler velocity (b, d) from EarthCARE (a, b) and NICAM (c, d) for Case 1 (left) and Case 2 (right). The unit of radar reflectivity is dBZ and the unit of Doppler velocity is m s-1. NICAM successfully reproduces a realistic cold front system similar to the observations, with a boundary location between warm and cold air consistent with the observed system. The NICAM simulation also captured a similar structure in both the radar reflectivity’s melting layer and the Doppler velocity distribution. However, NICAM tends to overestimate radar reflectivity near the melting layer and also underestimates Doppler velocity above 8 km (EarthCARE shows this region in yellow, whereas NICAM appears in green) due to an overestimation of the terminal velocity ice hydrometeors with small sizes. In this case, convection associated with the cold front is relatively weak compared to terminal velocity. As evidenced by the relatively smaller downward Doppler velocity inside the cold air between 55°S and 50°S in both the observations and simulations, we will expect upward vertical air motions near the cold front, as will be retrieved in the following section. A tropical convection case The right panel of Fig. 2 presents cross-sections near the Caribbean Sea in the Northern Hemisphere for 07:30–07:42 UTC on June 18, 2024 (Case 2), illustrating precipitation linked to a tropical storm. Strong convective systems are observed near 7°N and 20°N in the EarthCARE data (Fig. 2b, d), with radar reflectivity reaching up to 10 km. A faster falling speed is generally observed below 5 km, while slower downward motion appears at higher altitudes. However, compared to Case 1, the structure becomes more complex, with areas of enhanced downward motion near 7°N and a mix of strong upward and downward motions near 20°N. This profile suggests the presence of significant vertical air motions within the convective clouds. Figure 2b, d also shows low clouds in the southern hemisphere off the coast of South America, characterized by a Doppler velocity of approximately 2 m s− 1. Convective cores near 7°N and 20°N are evident. The 7°N core is narrower and more intense with strong downward Doppler velocities due to graupel/hail. The 20°N system, an MCS, includes both convective and stratiform regions, showing upward motion in convective areas and uniform downward motion in stratiform regions. NICAM reproduces cloud and precipitation structures in both hemispheres, capturing the 20°N core but missing the narrower 7°N one. Radar reflectivity and terminal velocity are well represented. However, NICAM tends to overestimate reflectivity and underestimate Doppler velocity. Doppler velocity is systematically underestimated for ice clouds, suggesting potential biases in the microphysics scheme, such as the representation of ice particle size distributions and fall speeds. We suspect that a major source of these biases is an inaccurate representation of the relationship between snow particle size and terminal velocity in NICAM, where relatively small snow particles are assigned unrealistically high terminal velocities. These biases are consistent with previous evaluations of NICAM against ground-based W-band vertical-pointing Doppler radar observations over Japan21. Furthermore, the strong radar reflectivity observed above the melting layer may be related to the snow size distribution22, which increases snow particle sizes with temperature. This assumption could lead to an overestimation of snow sizes in warmer conditions. For low clouds in the Southern Hemisphere, Doppler velocity is also systematically underestimated, further indicating the need for improvements in the model’s microphysics parameterizations. Terminal velocity of ice hydrometeors When the vertical air velocity is relatively small in a spatially and temporally averaged sense, the terminal velocity of hydrometeors in clouds can be inferred from Doppler velocity through its relationship with radar reflectivity. The approach using the relationship between radar reflectivity and Doppler velocity to estimate the terminal velocity has been widely used in ground-based observations23. In this study, we apply the same assumption across different cases and compute corresponding power-law relationships. The Doppler velocity from space is defined as where is the radar reflectivity-weighted terminal velocity of all hydrometeors, is vertical air velocity, and accounts for errors, which arise from random noise due to the spread of Doppler velocity within the radar beam widths, multiple scattering effects, non-uniform beam filling (NUBF), and positioning errors12–14. In this study, we do not take into account the effect of these errors in the Doppler velocity simulations using the NICAM data. The radar reflectivity-weighted terminal velocity () is expressed as where is the particle diameter, is the particle size distribution (number concentration per unit volume per unit diameter), is the terminal velocity of a particle of diameter , and is the backscattering cross-section of a particle of diameter . Under the assumption that vertical air motions are relatively small compared to the terminal velocity, the effect of the vertical air velocity is neglected. Consequently, we can estimate using a power-law relationship: where and are fitting parameters, and represents the radar reflectivity factor. We acknowledge this method has inherent limitations due to its sensitivity to vertical air motion variations, temperature, particle types, and convective/stratiform cloud categorization. Among them, we use the NICAM data to examine how vertical air motions influence the relationship described by Eq. (3), since this cannot be directly verified from satellite observations alone. To fit the power-law, data with reflectivity > − 20 dBZ were used, focusing on ice clouds above the melting layer (3 km for Case 1; 5 km for Case 2). Doppler velocities < 1 m s− 1 were used to reduce turbulence and error effects. Since attenuation is significant for liquid hydrometeors, especially when applying Eq. (3), attenuation correction would be required. Therefore, this study focuses on ice hydrometeors above the melting layer to avoid such complications. The resulting fits from NICAM were consistent under weak motion in Case 1 (Supplementary Information). Figure 3 shows joint histograms of radar reflectivity and Doppler velocity from EarthCARE and NICAM. For Case 1, EarthCARE yielded coefficients of a = -0.812 and b = 0.089, consistent with previous findings24. A smaller value of a indicates that smaller ice particles exhibit relatively larger Doppler velocities, suggesting that NICAM overestimates the terminal velocity of small ice particles. A larger b implies a steeper dependence of terminal velocity on particle size. In Case 2, EarthCARE yielded a = -0.853 and b = 0.182, corresponding to higher reflectivity (>5 dBZ) and stronger downward motion ( < − 2 m s⁻¹), indicative of rimed particles such as graupel, whereas Case 1 was dominated by slower-falling dry ice. Normalized joint histograms of radar reflectivity factor (Z) and Doppler velocity (DV) for EarthCARE (top) and NICAM (bottom) in Case 1 (left) and Case 2 (right). The contour unit is normalized frequency. The bin size is 2 dBZ for radar reflectivity and 0.2 m s- 1 for Doppler velocity. The unit of Z is mm- 6 m- 3. The black line represents a fitted power-law relationship between radar reflectivity and Doppler velocity. For NICAM, histograms were broadly similar but the coefficients showed systematic differences: in Case 1, NICAM underestimated a = -1.093 and b = 0.030, differing by nearly a factor of three from EarthCARE. These results suggest that NICAM may misrepresent the terminal velocity of smaller ice particles and underestimate the steepness of the size–velocity relationship. In Case 2, NICAM still showed these patterns like coefficients (a = -1.387, b = 0.062). These findings imply that improving the representation of ice particle size and terminal velocity in NICAM’s microphysics schemes could reduce such discrepancies. These relationships allow quantitative assessment of cloud microphysics. In Case 1, weak updrafts and absence of rimed particles made estimates reliable. In Case 2, the influence of vertical air velocity on the power-law relationship in Eq. (3) becomes more significant, and the attenuation of radar reflectivity increases the uncertainty. Nevertheless, the presence of upward Doppler velocity clearly indicates vertical air motions in this case.The power-law fit enables the derivation of terminal velocity as a function of particle diameter, demonstrating EarthCARE’s potential to improve microphysical representations in models. Vertical air motions inside ice clouds Vertical air motions inside clouds can be retrieved from Doppler velocity. Several retrieval approaches have been developed that combine various assumptions about particle size distributions and lidar information25. In this study, we propose a simpler and more intuitive method for estimating vertical air velocity based on a power-law relationship between terminal velocity and radar reflectivity, as defined in Eq. (1). observations. The analysis focuses on the atmospheric layer above the melting layer, where only ice hydrometeors are included because the terminal velocity is derived for ice particles. To ensure sufficient signal-to-noise ratio in Doppler velocity, the analysis is limited to regions with radar reflectivity greater than − 20 dBZ. To assess the validity of this assumption, we used NICAM data to examine the relationship between vertical air velocity and radar reflectivity. The results indicate that the assumption holds well in Case 1. In Case 2, however, notable errors appear when radar reflectivity exceeds 15 dBZ (see Supplementary Information). The left panel of Fig. 4 shows retrieved vertical velocities from EarthCARE/CPR and NICAM CPR-like data for Case 1, compared with NICAM simulations. Both datasets capture upward motions near the cold front at 52°S, validating the method. Near the cloud top (~ 10 km), EarthCARE shows upward motion, possibly due to retrieval bias; NICAM lacks such features (Fig. 4c) and serves as reference. EarthCARE also shows more frequent downward motion than NICAM, suggesting uncertainty. NICAM cold front updrafts are mostly < 0.5 m s− 1. EarthCARE captures weak, frequent motions with greater variance, likely due to unresolved turbulence or Doppler error. Retrieved vertical air velocity from EarthCARE (a) and NICAM (b), and simulated NICAM vertical air velocity (c) with radar reflectivity greater than − 20 dBZ for Case 1 (left) and Case 2 (right). In Case 2, convective cores near 13°N and 20°N are captured in both datasets, though sizes differ. Retrieved velocities tend to be weaker with more downward motion (Fig. 4, right panel). To evaluate accuracy, upward/downward velocities were averaged per frame. In Case 1, retrieved and simulated profiles between 2 and 9 km are similar (Fig. 5a), showing reliable application. EarthCARE shows peak motion near 4 km absent in NICAM. Observed vertical velocities are stronger, which may be attributed to turbulence and Doppler retrieval errors, but could also indicate an underestimation of vertical motions in NICAM. In Case 2, stronger vertical motions are observed. Retrieved and simulated values align but diverge above 14 km (Fig. 5b). Below 8 km, fast-falling hydrometeors like graupel cause strong downward motion, with increasing NICAM-retrieval discrepancies. Vertical profiles of averaged upward and downward air velocities for retrieved air velocities from EarthCARE and NICAM, and the vertical air velocity of the NICAM simulation for Case 1 (a) and Case 2 (b). Quantitative interpretation of vertical velocity must account for multiple factors. In weak convection, unresolved turbulence or Doppler velocity errors may cause overestimation. In strong convection, large vertical motions reduce Doppler error influence, but attenuation and rimed particles like hail introduce retrieval bias. Numerical simulations help assess such errors. This study emphasizes the synergy between satellite observations and high-resolution simulations. The high-resolution simulations explicitly represent convection and microphysical processes, allowing detailed comparisons with observations and improving the understanding of retrieval limitations and observational uncertainties. Conversely, satellite observations provide essential constraints for validating and refining model representations of cloud microphysics and dynamics. Together, these complementary approaches advance our understanding of cloud systems and contribute to improved atmospheric modeling and prediction. Summary and perspectives This study explores the capabilities of the newly launched EarthCARE satellite, particularly its CPR. EarthCARE/CPR provides global measurements of radar reflectivity and Doppler velocity, offering insights into terminal velocities of hydrometeors and vertical air motions within clouds. These motions drive extreme precipitation and influence climate sensitivity through cloud dynamics and microphysics. We conducted the global sub-kilometer-resolution atmospheric simulations using NICAM, generating EarthCARE-like radar signals with a satellite simulator. The simulations reproduce the observed convective structures reasonably well in selected cases, providing useful insights into cloud dynamics. We used the observed relationship between Doppler velocity and radar reflectivity to estimate the terminal velocities of ice hydrometeors, noting variability by precipitation type and riming intensity. Vertical air velocities in ice clouds were retrieved from EarthCARE/CPR data, showing consistent upward motion in tropical convection and cold fronts. Comparisons with NICAM revealed both agreement and systematic biases. We suspect that a major source of these biases is an inaccurate representation of the relationship between snow particle size and terminal velocity in NICAM, where relatively small snow particles are assigned unrealistically high terminal velocities. By analyzing the Doppler velocity–radar reflectivity relationship, our approach provides insights into these biases and suggests how to improve the model’s microphysical parameterizations. This study is intended as a starting point, demonstrating the potential of combining EarthCARE observations with high-resolution modeling for understanding cloud microphysics and vertical motions. EarthCARE data are currently being validated through extensive ground-based and airborne campaigns by various international institutions, and advanced retrieval algorithms for vertical air motion are will be introduced. Future work will build upon calibrated EarthCARE datasets and improved retrieval techniques, expanding the analysis to longer time periods to enhance the robustness and completeness of our findings. The first-ever global Doppler velocity measurements from EarthCARE thus represent a step forward in atmospheric science, enhancing our understanding of cloud processes, convection, and microphysics globally. Ongoing validation efforts using ground-based and airborne observations are essential for refining these measurements. Ultimately, EarthCARE’s unique global observations will significantly contribute to improving our knowledge of key atmospheric processes that govern cloud lifetimes, radiation budgets, precipitation formation, and atmospheric circulation. Methods EarthCARE/CPR The Earth Cloud Aerosol and Radiation Explorer (EarthCARE) was jointly developed with the European Space Agency (ESA) and the Japan Aerospace Exploration Agency (JAXA)2,3. It was launched at 23:20 UTC on May 28, 2024, from Vandenberg Space Force Base, California. Cloud Profiling Radar (CPR) on board EarthCARE is the world’s first spaceborne Doppler radar in the W-band (94 GHz) and was jointly developed by JAXA and the National Institute of Information and Communications Technology (NICT). This study utilized the CPR Level 1b product (version vCa), which was publicly released on 14th January 2025. The product includes radar reflectivity and Doppler velocity data. There were contaminations related to second-trip echoes26, and we removed them based on the method described by Battaglia26. Additionally, the Doppler velocity values were corrected by assuming the Doppler velocity at the ground surface was zero and unfolded for values exceeding the Nyquist velocity. We used 10 km integrated data of Doppler velocity to reduce the random error contaminated in the Doppler velocity measurements14. NICAM We use the Nonhydrostatic Icosahedral Atmospheric Model NICAM8–10. The global horizontal grid interval is approximately 870 m. The model has a terrain following a grid system with 78 vertical layers. The model’s top height is 50 km, and layer thicknesses gradually increase with altitude. Applied physics schemes are the NICAM Single-Moment Water 6 cloud microphysics scheme27,28, which considers six categories of hydrometers as prognostic variables, Smagorinsky type subgrid turbulence scheme and two-stream radiative transfer with a correlated k-distribution scheme MSTRNX29. The MATSIRO land surface model30 and slab-ocean model are used for the surface process. Initial conditions of the atmosphere and the ocean are derived by linear interpolation of ERA531. The simulation started at 00 UTC on 17 June 2024. Satellite simulator: joint simulator for satellite sensors (J-Sim) The satellite simulator, known as the Joint Simulator for Satellite Sensor (J-Sim), is used to calculate EarthCARE/CPR-like signals based on the NICAM simulation data19,20,32,33. Vertical profiles of radar reflectivity and Doppler velocity are derived using vertical profiles of cloud microphysical variables (specific mass, size distribution, and density of hydrometeors categorized as cloud water, cloud ice, rain, snow, and graupel), meteorological variables (temperature, pressure, and density), and surface conditions (albedo and emissivity). For simplicity, the satellite simulator is referred to as J-Sim throughout the text. GSMaP Global Satellite Mapping of Precipitation (GSMaP) is a satellite-based global precipitation dataset provided by JAXA34–36. The resolution is 0.1° in both latitude and longitude, and the temporal resolution is 1 h. In this study, GSMaP serves as an independent reference dataset to evaluate the large-scale distribution and intensity of precipitation simulated by NICAM and observed by EarthCARE. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements The authors thank members of the JAXA EarthCARE Science Team and the Joint-Simulator project. Simulations in this study were performed on the supercomputer Fugaku (Project ID: hp240106). Computational resources were partly provided by the National Institute for Environmental Studies. Author contributions W.R. conducted the analysis of EarthCARE and NICAM data with support from M.S. and H.O. S.M. performed the NICAM simulations. S.A. and T.K. were responsible for calibrating the EarthCARE CPR data. H.O. contributed to the development of the methodology. All authors reviewed and discussed the results collaboratively. Funding Woosub Roh and Masaki Satoh were supported by the Japan Aerospace Exploration Agency (EORA3) for the EarthCARE mission (25RT000098); JSPS KAKENHI (grant 24K00703); JSPS Core-to-Core Program (grant JPJSCCA20220001). Hajime Okamoto was supported by the Japan Aerospace Exploration Agency (EORA3) for the EarthCARE mission; JSPS (KAKENHI grant JP24H00275); Collaborative Research Program of the Research Institute for Applied Mechanics, Kyushu University (Fukuoka Japan). Simulations in this study were performed on the supercomputer Fugaku (Project ID: hp240106). Shuhei Matsugishi was supported by JSPS (KAKENHI grant JP24K22898); Data availability EarthCARE CPR Level 1b product (version vCa) are available from https://gportal.jaxa.jp/gpr/. The Joint-Simulator is available from https://www.eorc.jaxa.jp/theme/Joint-Simulator/userform/js_userform.html. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement EarthCARE CPR Level 1b product (version vCa) are available from https://gportal.jaxa.jp/gpr/. The Joint-Simulator is available from https://www.eorc.jaxa.jp/theme/Joint-Simulator/userform/js_userform.html.