EarthCARE/CPR is the W-band radar with the capability of the Doppler velocity measurement. The science data processing chain in the EarthCARE mission was summarized in Eisinger et al. (2024). In this study, EarthCARE L1B CPR one-sensor products (JAXA, 2024a) from 1 August 2024 to 30 June 2025 was utilized, providing radar reflectivity factor and Doppler velocity products. Although the native footprint diameter is about 750 m, the product provides data at 500 m intervals along the track. The vertical resolution is 500 m with 100 m vertical grid spacing. In this study, 5 km along-track integration is applied for each CPR grid point using neighboring 10 grid points to mitigate the effects of footprint differences between CPR and DPR. This horizontal integration also helps reduce the errors that contaminate the Doppler velocity, as described in Sect. 2.2. The integrated data retains the original 500 m spacing, rather than being resampled at 5 km intervals. The CPR was designed to achieve a sensitivity of approximately -35 dBZ with 10 km integration. The EarthCARE spacecraft operates in a sun-synchronous polar orbit with an inclination angle of 97.05°, crossing over the equator at local times of 02:00 and 14:00. In Sect. 3.4, the vertical air motion provided by the EarthCARE L2a CPR one-sensor cloud products (CPR_CLP) (Sato et al., 2025) Version Bb (JAXA, 2024b) was utilized, moving-averaged horizontally over 5 km to match the DPR footprint.

To ensure the precision of Doppler velocity measurements, EarthCARE/CPR operates at a higher pulse repetition frequency (PRF) than CloudSat/CPR. As a result, second-trip echoes due to mirror images and multiple scattering more frequently appear at higher altitudes. Following the approach of Battaglia (2021), we estimated the echo power of second-trip returns and applied corresponding masks to remove these artifacts, as described in Aoki et al. (2025a). Pre-launch studies suggested that the EarthCARE/CPR would be less affected by surface clutter than the CloudSat/CPR and that, over flat surfaces, clutter would not extend above 600 m (Roh et al., 2023). Observations are consistent with this expectation, although the altitude affected by clutter is higher over mountainous terrain. Taking these factors into account, we excluded data within five range bins (∼500 m) above the ocean surface and ten range bins (∼1000 m) above land to avoid potential contamination from ground clutter. Interpretations of Doppler velocities are described in Sect. 2.2. GPM/DPR consists of two radars: the Ku-band radar (KuPR) and the Ka-band radar (KaPR). Both radars operate with 49 beams in the cross-track direction, providing a swath width of approximately 250 km and a horizontal footprint of about 5 km in diameter (Kojima et al., 2012). The KuPR has a vertical resolution of 250 m and a minimum detectable reflectivity of 15.71 dBZ (Masaki et al., 2020). The KaPR supports two observation modes: High Sensitivity (HS) and Matched Scan (MS). In this study, only data from the HS mode are used in the statistical analysis from Sect. 3.2 onward, which offers a vertical resolution of 500 m and a minimum sensitivity of 13.71 dBZ (Masaki et al., 2020). The altitude of the GPM Core Observatory was raised from 407 to 442 km in November 2023 (Kubota et al., 2024). GPM Core Observatory observations that intersect with the EarthCARE satellite are only be available after the altitude change.

We used the GPM/DPR L2 Precipitation (hereafter, referred to as “2A.DPR”) (JAXA, 2014) that corresponds to the same observation period as the CPR. In 2A.DPR algorithm, the measured radar reflectivity is corrected for attenuation using the method of Seto et al. (2021). For the convective-stratiform classification in Sect. 3, we adopted the “typePrecip” flag provided in the 2A.DPR product (Awaka et al., 2021), which labels each radar footprint as either stratiform, convective, or others. Temperature data were provided by the 2A.DPR product, which were calculated from global analysis data of the Japan Meteorological Agency by interpolating to fit the footprint of the DPR (Kubota et al., 2020b).

The GPM satellite flies in a non-sun-synchronous orbit with an inclination angle of 65°, enabling observations of the low- and mid-latitude regions at a wide range of local times. Because of this unique orbital configuration, GPM ground tracks intersect with those of many sun-synchronous satellites, including EarthCARE, at multiple locations across the globe.

To jointly utilize three-frequency radar reflectivity and W-band Doppler velocity observations, we constructed the EarthCARE – GPM coincidence dataset, following a similar approach to that used for the 2B.CSATGPM dataset (Turk et al., 2021). Note that CloudSat was in Daylight-Only Operations while the GPM Core Observatory was in operation. The EarthCARE-GPM coincidence dataset allows for day and night analysis, which was impossible in the CSATGPM dataset. All cases in which the ground tracks of EarthCARE and GPM intersected within a 15 min time window were identified over the period from August 2024 to June 2025. For each coincidence event, the nearest DPR footprint was matched to every CPR horizontal grid points with 500 m spacing.

Figure 1 shows the latitudinal distribution of coincident observation footprints between EarthCARE/CPR and GPM/DPR. The number of coincidences increases with latitude and peaks around 65°, corresponding to the edge of the DPR's observation area. While a larger allowable time difference increases the number of samples, it also raises the risk of mismatches due to the movement and evolution of precipitation systems. In this study, a 15 min threshold is adopted, consistent with the 2B.CSATGPM dataset, to ensure a balance between minimizing observation mismatch and maintaining sufficient number of samples.

Given the higher sensitivity of CPR compared to both KuPR and KaPR, the echo top height (ETH) observed by CPR is, in principle, expected to be higher. However, in some cases, KuPR or KaPR report higher ETHs than CPR, likely due to temporal mismatches between sensors or spatial mismatches caused by differences in footprint size. In the statistical analyses, such cases were excluded. Additionally, in instances where multiple cloud layers were detected by CPR (e.g., upper-level anvils overlying shallow clouds), only the cloud layer that overlapped with the DPR echo region was extracted for analysis.

The Doppler velocity measured by CPR (Vd) can be expressed as 1Vd=Vair+Vtz+ε, where Vair is the vertical air motion, Vtz is the reflectivity-weighted terminal velocity of hydrometeors, and ε represents the measurement error. In this paper, positive values of Vd are defined as upward direction. Therefore, positive Vair means upward air motion and Vtz is always negative. The reflectivity-weighted terminal velocity Vtz is further formulated as 2Vtz=∫vtDNDσbDdD∫NDσbDdD, where vtD,ND,σbD are the terminal velocity, particle size distribution function, and backscattering cross-section for each particle diameter D, respectively. The denominator of Eq. (2), when multiplied by λ4/π5Kw2, where λ is the radar wavelength and Kw is the normalizing dielectric factor, corresponds to the radar reflectivity factor (Z). The error term ε includes several components: random noise due to reduced decorrelation arising from the fast-moving speed of the satellite-based sensor (εrandom) (Doviak and Zrnic, 1993; Hagihara et al., 2023); uncertainty caused by pointing due to perturbation in satellite attitude and antenna thermal distortion (εpointing) (Tanelli et al., 2005); uncertainty due to velocity aliasing or folding beyond the Nyquist limit, non-uniform beam filling (NUBF) (Sy et al., 2014), and multiple scattering (Battaglia et al., 2011). In Eq. (2), ε is expressed as an additive term, but it includes not only systematic biases, but also random uncertainties mentioned above that can only be mitigated by adaptive filtering or along-track integration.

To correct for εpointing, we applied a bias correction based on the assumption that the 100 km horizontally averaged Doppler velocity at the surface should be zero. In the averaging, only horizontal grid points whose normalized surface cross section (σ0) satisfies -15<σ0<27.5 dB are used to exclude cases where the surface echo is either too weak due to cloud attenuation or too strong, causing Vd to be severely affected by surface property. This approach compensates for periodic biases of approximately 0 to 0.5 m s⁻¹, which are thought to result from thermal distortion of the antenna leading to slight variations in pointing angle with orbital position. Horizontal 5 km-integration introduced in this study suppresses εrandom to a level that does not hinder analysis and also reduces the impact of NUBF. To mitigate contamination from multiple scattering and heavily attenuated cases, we followed the method of Battaglia et al. (2011) and excluded range bins where the cumulative reflectivity from the top of the atmosphere exceeds a specified threshold. Finally, an aliasing correction was applied to Vd following the method of Hagihara et al. (2023) and was further refined using the surrounding Z and Vd profiles as a reference.

Previous studies using various types of radar have attempted to derive Vair by subtracting the estimated Vtz from the observed Doppler velocity (Kollias et al., 2003; Okamoto et al., 2003; Yamamoto et al., 2008; Sato et al., 2009, 2010). Their methodology provides Vair by estimating the particle type and size distribution from the reflectivity profile and path-integrated attenuation, and then subtracting the corresponding Vtz. This may yield reasonable estimates of Vair in ice particle regions. However, in rain regions where attenuation is strong, it is difficult to accurately correct radar reflectivity for attenuation, which makes the retrieval of the particle size distribution challenging.

In contrast, Vd is retrieved from the pulse-to-pulse phase difference rather than from signal amplitude (Eisinger et al., 2024), and is therefore intrinsically less affected by attenuation of returned power (Doviak and Zrnic, 1993). In practice, pulse-to-pulse phase correlation is maintained under moderate attenuation, allowing the velocity retrievals to remain stable (Tian et al., 2007; Kollias et al., 2014). The main limitation arises when severe attenuation and multiple scattering associated with heavy rain or ice precipitation substantially degrades the pulse-to-pulse phase correlation, leading to large errors (Matrosov et al., 2008; Battaglia et al., 2011). In this study, cases containing rain, wet snow, and graupel were retained, while severe attenuation were excluded by applying screening following Battaglia et al. (2011), leading to preserve physically meaningful velocity information in these hydrometeor regimes.

In the current study, we propose a method for estimating Vair by subtracting the reflectivity-weighted terminal fall velocity Vtz which is estimated from the particle size distribution derived from DPR measurements, from the observed Vd from collocated CPR measurements. In 2A.DPR algorithm (Seto et al., 2021), drop size distribution function is expressed as 3ND=NwfD;Dm=Nw6μ+4μ+444Γμ+4DDmμexp⁡-μ+4DDm, where Γ is the Gamma function with the value of μ fixed at 3, Nw is the parameter related to the number concentration of raindrops, and Dm is the mass-weighted mean diameter. Substituting Eq. (3) into Eq. (2) yields 4Vtz=∫vtDfD;DmσbDdD∫fD;DmσbDdD. From this formulation, Vtz can be uniquely determined from the DPR-estimated Dm, and thus Vair can be calculated. For rain layers, vt was computed using the empirical relationship proposed in Atlas and Ulbrich (1977), with a correction factor for air density, as given: 5vtD=-3.78D0.67⋅cρ,6cρ=ρ0ρ=ρ0RTp. Here, the unit of D is mm, ρ denotes the ambient air density, ρ0 is the standard air density (set to 1.225 kg m⁻³), R is the specific gas constant for dry air (287 J kg⁻¹ K⁻¹), and p and T represent pressure and temperature obtained from auxiliary data. The backscattering cross-section σb was derived from Mie scattering calculations for spherical raindrops at W-band frequency.

For snow, σb and vt was calculated in the same manner as in the 2A.DPR algorithm, assuming homogeneous spherical particles with a density of 0.10–0.13 g cm⁻³ and a melted-equivalent diameter following the particle size distribution given by Eq. (3). The terminal fall velocity of snow was calculated following Magono and Nakamura (1965) as follows: 7vtDs=-8.80.1Dsρs0.5⋅cρ, where Ds is the unmelted snow particle diameter in mm, and ρs is the density of snow particles in g cm⁻³. On the other hand, ice particles can take various shapes, sizes, and densities, such as those of snow, graupel, and hail. Because σb and vt vary depending on these parameters, the assumptions made for snow in this study are often not valid. Although it would be ideal to account for more realistic and complex scattering and fall characteristics of ice particles (Kuo et al., 2016; Ori et al., 2021), considering such diversity is challenging because the CPR observes only in the nadir direction and therefore cannot provide information on particle asymmetry. This contrasts with ground-based dual-polarization radars, which observe the hydrometeor from the side, where particle asymmetry is more evident and can provide additional information. In addition, such information on particle diversity cannot be inferred from the current version of the 2A.DPR algorithm and is therefore left for future work.

In this subsection, two cases of CPR and DPR coincidence observations are presented: one in which stratiform precipitation was dominant and one in which convective precipitation was dominant. Figure 2 shows a case over the northeastern Pacific associated with a tropical cyclone observed on 22 August 2024. The EarthCARE and GPM passed over a broad stratiform precipitation system with time difference of approximately 6 min (Fig. 2a and b). In Fig. 2c, the vertical cross-section of W-band radar reflectivity observed by CPR (ZW) reveals deep cloud structures reaching altitudes of approximately 15 km, and detailed texture within the upper-level ice and snow clouds above the 0 °C level is evident. Locally enhanced reflectivity just below the 0 °C level corresponding to the melting layer is observed. Below the melting level, significant attenuation occurs in the rain region, resulting in reduced ZW values. The Vd profile clearly distinguishes between the slowly falling ice and snow layers aloft and the faster-falling rain below (Fig. 2d). Because Vd is derived from phase information of the radar signal, it remains observable even in layers affected by attenuation.

Ku-band radar reflectivity retrieved by KuPR (ZKu) shows ETHs around 6 to 10 km, capturing signals only in areas where ZW exceeds approximately 10 dBZ in the ice phase (Fig. 2e). However, strong scattering from developed snow near the melting layer and rain below is captured effectively without attenuation. A prominent bright band, commonly observed in stratiform precipitation, is evident near the 0 °C level. Ka-band radar reflectivity (ZKa) exhibits a similar profile to ZKu, although the reflectivity values are lower due to Mie scattering effects and stronger attenuation at Ka-band frequencies (Fig. 2f).

Figure 3 presents a case of scattered convective precipitation observed over southern South America on 23 March 2025, with a time difference of approximately 4 min between the EarthCARE and GPM observations. In this event, ETHs and reflectivity maxima observed by CPR and KuPR vary across individual convective cells, and no distinct bright band signature is evident (Fig. 3c–f). The Vd exhibits considerable variability, with no clear transition near the 0 °C level (Fig. 3d). In some locations, strong scattering in the ice phase causes severe signal attenuation in ZW measurement, leading to a complete loss of surface returns (e.g., near 30 °S). In these areas, regions of high ZKu (>30 dBZ) and large downward Vd exceeding 2 m s⁻¹ extend above the 0 °C level, suggesting the presence of densely rimed particles such as graupel or hail. In contrast, shallow precipitation is characterized by weak reflectivity and small Vd values, indicating a dominance of numerous small droplets, as typically seen in drizzle.

The horizontal 5 km integration applied to the Vd field in Fig. 2d is highly effective in reducing the contribution of random noise induced by decorrelation (εrandom). However, in the convective case shown in Fig. 3d, this integration may result in over-smoothing, mixing the signatures of strong echoes within convective cores with weaker echoes at cloud edges. Because the integration is performed as a reflectivity-weighted average rather than a simple moving average, the features of the convective core are emphasized, which may in turn lead to artificially enhanced downward velocities near cloud edges where echoes are weak. In the statistical analyses presented later in this paper, such edge regions, where coincidence with DPR observations could not be ensured after averaging, were excluded from the analysis, as indicated by the black-plotted areas in Figs. 2g and 3g.

In this section, we perform a statistical analysis of the coincident profiles of Z and Vd observed by CPR and DPR to examine the characteristics of clouds and precipitation profiles with temperature. Figure 4 presents Contoured Frequency by tEmperature Diagrams (CFEDs; Hashino et al., 2013) of ZW, Vd, ZKu, and ZKa, where temperature is used as the vertical coordinate. The probability density function is normalized within each temperature bin. Only profiles where echoes are detected by both DPR and CPR are included. These profiles correspond to 5.1 % of all profiles and 10.1 % of the profiles in which echoes are detected by the CPR. Because the sensitivity of the DPR is lower than that of the CPR, the number of samples in the colder temperature range in Fig. 4c and d is considerably smaller than that for the CPR. To avoid misinterpretation, temperature ranges with fewer than 1000 samples are not shown. Using temperature instead of altitude as the reference axis allows for more direct interpretation in terms of cloud microphysical processes. To ensure a sufficient number of samples, observations from various regions and times were aggregated over the analysis period. Consequently, the results may include observations from different altitudes depending on latitude and season; however, such variations are beyond the scope of this study.

In Fig. 4, above the freezing level, Z in all frequency bands (W, Ku, and Ka) generally increases with decreasing altitude, and Vd also increases in the downward direction. This indicates that ice particles grow and become dominated by larger sizes, resulting in an increase in their terminal fall speeds. Reflecting the scattering properties specific to each frequency band, the height ranges over which Z increases differ between the W-band, Ku-, and Ka-band observations. As shown in Fig. 4a, above approximately -20 °C, ZW increases with decreasing altitude, indicating particle growth with height. However, below -20 °C, where ZW exceeds 0 dBZ and the scattering regime likely attenuation and transitions from Rayleigh scattering to Mie scattering, the rate of increase in ZW becomes smaller. In the height range between -10 and 0 °C, the ZW PDF shows little dependence on height, indicating it difficult to detect altitude-dependent variations in microphysical properties from ZW.

In KuPR and KaPR observations (Fig. 4c and d), reflectivity values above -10 °C are close to the minimum detectable thresholds of 15 and 13 dBZ, respectively. Echoes are detected only when hydrometeors with strong reflectivity observed by CPR extend into the upper layers. According to the conversion table in Skofronick-Jackson et al. (2019), this reflectivity level corresponds to approximately 11 dBZ in the W-band. In the layer between -10 and 0 °C, ZKu and ZKa increases with decreasing altitude, indicating growth of hydrometeors such as snowflakes or graupel. This vertical evolution, which is not captured in ZW due to Mie effects and attenuation, is better resolved in the Ku- and Ka-band observations.

On the other hand, the CFED of Vd shown in Fig. 4b illustrates an increase in downward velocity with increasing temperature between -20 and 0 °C, which is consistent with the particle growth inferred from KuPR observations. This suggests that Doppler velocity measurements can provide insights into the development of intense snowfall that could not be captured by CloudSat's reflectivity observations alone.

Around the melting level, a marked change in Vd is observed (Fig. 4b). The downward velocity increases from approximately -1 to -4 m s⁻¹, corresponding to the transition from snow to rain. This is accompanied by a localized enhancement in Z of all band, likely due to the contrast in complex refractive indices between ice and water (Fig. 4a, c, and d). Below the melting level, where rain becomes dominant, ZW decreases with increasing temperature, primarily due to strong attenuation at W-band frequencies. In contrast, median Vd stays around -4 m s⁻¹ at temperatures above 5 °C, which is consistent with the terminal fall speed of the raindrops. This indicates that Vd is less affected by attenuation than ZW, and it can reflect the properties of hydrometeors even in the rain layer. Meanwhile, Ku-band observations are subject to less attenuation than W-band, enabling more reliable attenuation correction and making Z in the rain layer more representative of the actual precipitation structure.

It should be noted that although ZKu and ZKa increase with increasing temperature between 10 and 25 °C (Fig. 4c and d), this does not necessarily imply stronger precipitation lower in the column. Near the melting layer, the data include contributions from all latitudes, whereas the observations around 20 °C are dominated by low-latitude regions. Although the limited number of samples makes detailed discussion difficult at present, future work should involve analysis separated by weather systems and freezing level height to better investigate the vertical growth processes of hydrometeors.

To examine the relationship between Z and Vd in more detail across different frequencies, histograms were constructed for various temperature ranges, as shown in Fig. 5. The dashed and dotted black lines in Fig. 5 represent the theoretical Z-Vtz relationship for rain and snow, respectively, calculated under the assumptions described in Sect. 2.2. Figure 5a–d use ZW as the horizontal axis, while Fig. 5e–h use ZKu. For snow, lines corresponding to ρs of 0.05, 0.1, 0.2, and 0.3 g cm⁻³ are plotted. Such Z-Vd relationships have long been investigated using ground-based radar observations and serves as a useful metric for inclusion in weather and climate models. In the upper troposphere (T<-10 °C; Fig. 5a), Vd tends to increase with increasing ZW, indicating that larger reflectivity is associated with faster-falling particles. Assuming that the vertical air motion averages to zero over many samples, the mean Vd can be interpreted as representative of the Vtz. The ZW–Vd distribution in Fig. 5a follows the theoretical Vtz curve for ρs=0.05 g cm⁻³, showing that the downward fall speed increases with increasing ZW, which provides insight into the growth of ice particles. On the other hand, in KuPR observations (Fig. 5e), the distribution is concentrated around 15–18 dBZ, near the instrument's sensitivity limit, suggesting that only stronger signals are detected in these layers due to limited sensitivity.

In the temperature range between -10 and 0 °C, the distribution of ZW is tightly concentrated around 8–11 dBZ, making it difficult to discern any clear relationship between Z and Vd (Fig. 5b–d). Moreover, in the melting and rain layers (T>0 °C), attenuation becomes significant, making it unclear whether low Z values are due to inherently weak reflectivity or attenuation effects (Fig. 5c and d). In contrast, when focusing on ZKu, a negative relationship between ZKu and Vd is observed (Fig. 5f–h). In the range of -10 °C <T<0 °C (Fig. 5f), the distribution closely follows the theoretical curves for snow with densities of 0.05–0.1 g cm⁻³, while in the range of T>4 °C (Fig. 5h), it aligns well with the theoretical curve for rain. In theory, the slope of ZKu versus Vtz is steeper for rain than for snow (ρs∼0.1 g cm⁻³) because the fall velocity of snow exhibits a weaker dependence on particle size compared to that of rain. This trend is also evident in the observations: the ZKu–Vd slope is -0.0170 between -10 and 0 °C, -0.0320 between 0 and 4 °C, and -0.0444 for T>4 °C, indicating a steepening slope as precipitation transitions from snow to rain through melting (Fig. 5f–h). In addition, in the 0–4 °C melting layer, Vd exhibits a broad range of values because this region captures the transition from slow-falling particles with terminal velocities around 1 m s⁻¹ to faster-falling raindrops with velocities approaching 4 m s⁻¹ (Fig. 5g). This suggests that Vd may serve as a useful complementary indicator for estimating precipitation intensity in melting and rain layers by using CPR observations where ZW alone may be insufficient due to attenuation.

To investigate how in-cloud statistics of Z and Vd differ according to the characteristics of precipitation systems, a classification of coincident events was conducted, followed by a statistical analysis of the corresponding vertical profiles. The cloud top height (CTH) and precipitation top height (PTH) are key variables that characterize the developmental stage of precipitation systems (Masunaga et al., 2005, hereafter M05; Stephens and Wood, 2007; hereafter SW07; Takahashi and Luo, 2014; Kikuchi and Suzuki, 2019). M05 first categorized precipitation systems using CTH–PTH joint histograms constructed by deriving the PTH from the 18 dBZ echo top observed by the TRMM PR and the CTH from the 11 µm brightness temperature observed by the Visible Infrared Scanner (VIRS) onboard TRMM. SW07 improved upon this approach by incorporating millimeter-wavelength radar observations, which allowed them to better represent multilayer cloud structures, whereas VIRS observations can capture only the uppermost cloud layer. Following these studies, this work assumes that the ETH retrieved from CPR corresponds to the CTH, and that from KuPR corresponds to the PTH. The joint histograms of temperature at CTH and temperature at PTH for stratiform and convective precipitation determined by 2A.DPR algorithm, respectively, are shown in Fig. 6. The histograms are reminiscent of the histograms presented by M05 (their Fig. 1) and SW07 (their Fig. 8).

When the PTH is located above the 0 °C level, it is indicative of a cold-type precipitation process, in which ice particles grow into relatively large snow aggregates or graupel through aggregation and riming. In stratiform precipitation, the PTH is mostly at temperatures lower than 0 °C, indicating that most cases are associated with cold-type precipitation (Fig. 6a). In particular, a high frequency of occurrence is confined within the PTH range of -20 to -10 °C regardless of CTH. This temperature range corresponds to the layer where ice habits transition with temperature, as shown in the classical Nakaya diagram (Libbrecht, 2005). It is also referred to as the dendritic growth layer, where cloud ice particles are thought to grow into snow through depositional growth, aggregation, and potentially secondary ice processes (von Terzi et al., 2022). Such temperature-dependent microphysical processes may explain why ice particles in typical stratiform clouds become large enough to be detected by the KuPR only when they reach below the -20 °C level.

In contrast, convective precipitation shown in Fig. 6b exhibits a much wider range of PTH values, extending from 20 °C down to below -40 °C, with a sparse distribution in the CTH–PTH histogram. Focusing on deep clouds with CTH above the -20 °C level, the PTH tends to lie close to the one-to-one line between PTH and CTH, indicating that the precipitation top height is nearly as high as the cloud top. This situation can be interpreted as the result of strong updrafts within the system that lifted large hydrometeors toward the cloud top, as discussed in Takahashi and Luo (2014). In addition, in convective cases, a pronounced peak in occurrence is found where the PTH is below the 0 °C level. When the PTH is below the 0 °C level, it suggests a warm-type precipitation process, where raindrops grow through collision and coalescence of liquid water droplets. Such warm-type shallow precipitation is likely associated mainly with shallow cumulus or congestus clouds, because the DPR has limited sensitivity to detect light precipitation (∼1 mm h⁻¹ or less; Hayden and Liu, 2018), and thus does not effectively capture shallow stratus or stratocumulus.

Based on the considerations above, we performed a classification of cloud and precipitation types. Cases where the PTH is below the 0 °C level and the CTH is also below the -20 °C level were categorized as Shallow. This type is intended to extract conditions dominated by warm-type rain processes and includes both convective and stratiform precipitation. Cases with the PTH above the 0 °C level and the CTH is also above the -20 °C level were classified as Deep. In these cases, the cloud layer extends well above the freezing level, indicative of dominant cold-type processes. Furthermore, the Deep Convective category was subdivided into two subtypes: Moderate (DC-M) and Tall (DC-T). The DC-T subtype represents highly developed convective clouds in which large ice particles are transported to higher altitudes and are detectable by KuPR. In contrast, the DC-M subtype likely corresponds to less mature or dissipating convective systems.

Figure 7 presents CFEDs of ZW, Vd, and ZKu for four precipitation types: Deep Stratiform, DC-M, DC-T, and Shallow. Since the number of stratiform events is substantially larger than that of convective or shallow events, the stratiform CFED in Fig. 7 closely resembles that in Fig. 4. In terms of reflectivity, a defining feature of stratiform precipitation is the presence of a local peak near the melting level, representing a bright band (BB), which is clearly observed in Ku band. In contrast, such features are absent in convective and shallow types. This result is consistent with the classification logic used in the DPR algorithm, in which the presence of a BB is a key criterion for identifying stratiform precipitation. In convective and shallow types, ZKu tends to increase with temperature below the melting level without exhibiting a local minimum. The difference from stratiform suggests the growth of raindrops through warm rain processes such as collision and coalescence in these types.

For all three variables, convective precipitation exhibits a broader distribution than stratiform, indicating greater variability in the microphysical properties of falling hydrometeors, more heterogeneous spatial distribution, and more vigorous vertical turbulent motions. This variability is particularly pronounced in the DC-T category, where fluctuations are especially large.

In Shallow cases, ZW and ZKu are generally low, and Vd values are also smaller than in Deep cases. This is consistent with previous studies such as Dolan et al. (2018), based on disdrometer measurements, which indicate a dominance of numerous small raindrops in such conditions in shallow warm-type rainfall.

SW07 has classified cloud types using a similar approach based on ground-based Ka-band radar observations and presented comparable histograms of Z as a function of height (their Fig. 10). Although the attenuation conditions differ – since their observations are made from the ground upward, whereas the present study is based on spaceborne downward-looking radar measurements – the vertical profiles of ZW in each category (Fig. 7a–d) exhibit similar characteristics. This study extends their findings by introducing an additional perspective through the use of Vd.

This section examines how the differences in Z and Vd among precipitation types identified in Sect. 3.3 can be interpreted in terms of the underlying physical processes in convective and stratiform systems. The analysis focuses on vertical structure, particularly the contributions from vertical air motion and terminal fall velocity. Figure 8 presents the mean and standard deviation of Z and Vd at each temperature for all precipitation types, as derived from the CFEDs in Fig. 7. The discussion is organized by characteristic layers within the cloud system, namely the snow and rain layers, with particular attention to the Z-Vd relationship.

CPR_CLP, one of JAXA's EarthCARE Level-2a standard products, provides its own estimates of Vair, which are derived solely from single-sensor observations by the CPR (Sato et al., 2025). Both the method implemented in CPR_CLP and the CPR-DPR combined approach presented in this study are based on the same fundamental concept of calculating Vair by subtracting the Vtz from the Vd. However, the two methods differ in how Vtz is determined: the CPR_CLP estimates it from a particle size distribution (PSD) inferred using only the CPR-measured Zw and Vd, whereas the present method uses the PSD derived from the 2A.DPR algorithm. Because of this difference in the underlying PSDs and scattering database, discrepancies between the two Vair estimates are expected. Therefore, in this section, we compare the two Vair estimates to assess the consistency and reliability of the retrieved Vair, as well as the implicit assumptions regarding Vtz and PSD within each algorithm. Since the CPR_CLP product does not provide the quantity corresponding to Vtz, the comparison focuses solely on Vair.

Figure 11 shows histograms of Vair for deep stratiform and deep convective precipitation, separately for the temperature ranges corresponding to snow (-10 °C <T<0 °C) and rain (T>4 °C). Table 2 summarizes the mean and standard deviation of the Vair histograms shown in Fig. 11 for each retrieval method. Overall, the histograms exhibit smaller standard deviation in stratiform cases (0.5–0.9 m s⁻¹) and larger standard deviation in convective cases (0.9–1.6 m s⁻¹), likely associated with turbulent motions.

For the snow layers (Fig. 11a and b), weak upward motions appear on average in both types (0.3–0.6 m s⁻¹), likely associated with latent heat release during ice particle growth. On the other hand, some discrepancies exist between the two Vair estimates, probably due to the radar frequency and assumptions about ice particle properties. The CPR_CLP algorithm accounts for scattering from ice particles with various shapes and orientations, whereas the DPR algorithm assumes simple spherical particles with a fixed bulk density of 0.10–0.13 g cm⁻³ . The peak of the Vair distribution from the DPR-based method is about 0.3 m s⁻¹ higher than that from CPR_CLP, suggesting a positive bias. This difference may indicate that the DPR-based PSD retrieval does not fully capture the contribution from smaller particles with slow Vtz compared with the W-band CPR-based retrieval. On the other hand, for convective precipitation, the DPR-derived Vair shows a larger proportion of downward motions (<-1 m s⁻¹) than the CPR_CLP Vair. This is likely because convective cases include more dense particles, such as graupel or hail, with densities exceeding 0.3 g cm⁻³, which are not represented in the current 2A.DPR algorithm.

For the rain layers (Fig. 11c and d), the DPR-based method and CPR_CLP show more similar Vair histograms than in the snow cases. For stratiform precipitation, the mean Vair is close to 0 m s⁻¹, whereas for convective precipitation, it shows stronger upward motion of about 1 m s⁻¹ with a larger variance. The two histograms agree well for strong upward motion (Vair>1 m s⁻¹); however, for Vair<1 m s⁻¹, the CPR_CLP Vair shows a distinct peak around 0 m s⁻¹, while the DPR-based Vair exhibits a smoother distribution. Despite such several differences, both methods yield comparable Vair histograms overall.

While further validation is still needed, the approach presented in this study provides useful reference information for evaluating the validity of vertical air motion retrieval, which can otherwise only be obtained by specific ground-based vertically pointing radars where direct observational data are extremely limited.