the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Estimation of wet radome and rain induced attenuation in cloud radar observations
Dmitri Moisseev
Estimation of the properties of ice clouds above rain from ground-based cloud radar observations is affected by the wet radome, rain path, and melting layer attenuation. In this study, we present a method for estimating wet radome and rain attenuation using collocated disdrometer observations. To ensure that cloud radar retrievals can be performed at a relatively high temporal resolution, on the order of a minute, attenuation estimates are computed at the same temporal resolution. However, the small sampling volumes of disdrometers introduce significant uncertainty in these estimates. To address this, we developed a statistical model to calculate the uncertainty range of the estimated drop size distribution (DSD) parameters and radar variables. The computed radar variables are compared to radar observations to identify radome attenuation. As an additional check of the validity of the estimated variables, the DSD observations are used to calculate rain intensity and accumulated precipitation, which are then compared to precipitation gauge measurements. In addition to radome attenuation, rain path attenuation is estimated using specific attenuation derived from the disdrometer observations. Limitations of this approach are presented and discussed. The proposed method is demonstrated using observations of two rain events collected in Hyytiälä, Finland. These events were selected to demonstrate the effect of radome attenuation, which was rather minor in one case and as large as 10 dB in the other.
- Article
(9016 KB) - Full-text XML
- BibTeX
- EndNote
The cloud remote sensing observations carried out at the Aerosol, Clouds and Trace gases Research Infrastructure (ACTRIS) stations provide a comprehensive dataset of cloud properties that span across different climatic zones. The cloud products provided by ACTRIS cloud profiling stations (also known as Cloudnet) range from raw data to multi-sensor cloud microphysical retrievals, such as liquid and ice water contents (Hogan and O’Connor, 2004; Hogan et al., 2006b; Illingworth et al., 2007). These observations can be used to evaluate the cloud representation in numerical weather prediction models (Hogan et al., 2006a), among other things.
The availability of data from certain remote sensors and the associated measurement uncertainties could limit availability of some retrieval products. Currently, there is a significant uncertainty associated with retrievals of ice clouds properties, in ice over rain cases. Uncorrected attenuation caused by a wet radar radome, path attenuation in rain, and attenuation in the melting layer introduce significant errors that make the retrieval of ice cloud properties highly inaccurate (Hogan et al., 2006b). To improve the retrievals of the ice clouds, the above-mentioned attenuation should be estimated and compensated for.
Radome attenuation can be reduced using a radome blower in combination with hydrophobic coating as implemented in some of the newer cloud radar systems (Küchler et al., 2017). However, the hydrophobic coating degrades over time, making radome attenuation an issue, even for systems equipped with blowers. The strength of attenuation depends on the rain intensity and duration (Gorgucci et al., 2013; Kurri and Huuskonen, 2008). For radars operating at W-band wavelengths, this attenuation can be greater than 10 dB (Myagkov et al., 2020). Additionally, rain path attenuation is strongly influenced by the rain intensity (Williams, 2022; Matrosov, 2008; Lhermitte, 1990). Various methods exist to determine the attenuation in rain (Matrosov, 2007; L'Ecuyer and Stephens, 2002; Hitschfeld and Bordan, 1954). L'Ecuyer and Stephens (2002) and Hitschfeld and Bordan (1954) require knowledge of path integrated attenuation, which is not available for a single frequency cloud radar observations. Matrosov (2007) utilizes drop size distributions (DSDs) obtained from a disdrometer to compute attenuation by T-matrix calculations. The last attenuation correction needs to be applied to the melting layer. This has been studied by Li and Moisseev (2019), who presented a technique to derive melting layer attenuation based on multi-frequency radar differential attenuation measurements, as well as by Matrosov (2008), who used a relationship between rain rate and melting layer attenuation. The derived melting layer attenuation parameterizations Matrosov (2008) and Li and Moisseev (2019) can be used to estimate this attenuation, but they need rain rate and unattenuated reflectivity values from just below the melting layer.
The standard equipment of the ACTRIS cloud profiling stations consists of a cloud radar, ceilometer, microwave radiometer, disdrometer, precipitation gauge and weather station. Disdrometer observations of DSDs are frequently used in combination with cloud radars, in order to derive radar-based methods to estimate rainfall and precipitation properties (Giangrande et al., 2019; Tridon et al., 2017), or to determine radar calibration offsets (Deng et al., 2025; Myagkov et al., 2020; Kollias et al., 2019; Williams et al., 2005; Lamer et al., 2021). When validating radar-derived rain retrieval methods, disdrometers provide essential data on precipitation characteristics. For radar calibration, the disdrometer serves as a reference, allowing for monitoring differences between the disdrometer-derived reflectivity measurements and the radar observations.
The objective of this study is to estimate radome and rain path attenuation at temporal scales of a minute. Since such estimates can be rather uncertain, the validity interval of the estimates should also be computed. The challenge in this type of analysis is the inherent uncertainty of computed radar retrievals from disdrometer observations arising from the relatively small sampling volume. Specifically, the observed DSDs from the disdrometer are used to estimate radar-observable quantities, including the reflectivity factor (Z), rainfall rate (R), and specific attenuation (Ai). A disdrometer sampling volume is in the order of 10−1 m3, whereas a cloud radar observation volume usually exceeds 103 m3. Disdrometer observations of DSD are typically truncated on smaller and larger diameter sides. Due to smaller sampling volumes, disdrometers tend not to observe larger droplets (Ulbrich and Atlas, 1998). Because the sensitivity of the disdrometer sensors, smaller droplets, typically with diameters less than 0.4 mm, are also not detected (Moisseev and Chandrasekar, 2007).
Beyond estimating radar retrieval uncertainties, a further challenge lies in comparing disdrometer and cloud radar measurements. This is caused by the difference in the measurement altitudes. As droplets fall from the radar volume to the surface, droplets are affected by processes such as evaporation and collision-coalescence. These processes modify the DSD affecting the comparison between radar retrievals. In addition, near-field effects, parallax errors, and residual pointing errors may influence the lowest radar range gates. The impact of these processes on the radar–disdrometer comparison is also evaluated in this study.
In this study, DSD observations from disdrometer are used to calculate radar reflectivity values, to estimate radome attenuation and rain path attenuation. The uncertainty range of disdrometer based estimates of radar variables are also computed. To compute the uncertainty range, a forward model of the disdrometer observations is used. This model takes into account the sampling characteristics of a disdrometer that is used for DSD observations. It is proposed that the rain path attenuation can be mitigated by using specific attenuation computed from DSD observations. This approach is constrained by factors such as rain evaporation and increased precipitation intensity due to embedded clouds. A method for diagnosing such cases is also presented. The method is illustrated on data collected at the ACTRIS cloud remote sensing station in Hyytiälä, during rain events recorded in the years 2022 and 2023. Two case studies are presented: one characterized by minimal radome attenuation and another exhibiting strong radome attenuation.
This study is based on measurements collected at the University of Helsinki Station for Measuring Ecosystem-Atmosphere Relations (SMEARII) located in Hyytiälä, southern Finland (61.844° N, 24.287° E) (Hari and Kulmala, 2005). As an ACTRIS cloud profiling station, SMEARII is equipped with a set of collocated standard instruments consisting of a disdrometer, ceilometer, 94 GHz cloud radar, two precipitation gauges, weather station, and microwave radiometer. The cloud profiling instrumentation is located on the lake Kuivajärvi shore, about 30 m (150 m above mean sea level, a m.s.l.) below the base of the SMEAR II mast, where the majority of the SMEARII observations are collected. Meteorological data used in this study were collected at the radar field at an altitude of 1.5 m above ground level (a.g.l.) or 151.5 m a m.s.l., and at multiple heights on the SMEAR II mast. Two rain events that occurred on 26 May 2022 (Moisseev et al., 2025a) and 28–29 July 2023 (Moisseev et al., 2025b) are used to illustrate the presented methodology. In addition, disdrometer observations have been used to present a climatology of DSD parameters over the period 2014–2024 (Moisseev and Petäjä, 2025).
2.1 Radar
A zenith pointing 94 GHz dual-polarization frequency modulated continuous-wave Doppler cloud radar (HYytiälä Doppler RAdar, HYDRA-W) (Küchler et al., 2017) is used in this study. The radar is equipped with two antennas that have an aperture of 500 mm and a beam width of 0.50°. HYDRA-W uses three chirps, but only the two lowest chirps are utilized since the proposed method applies solely to rain below the melting layer, which these two chirps cover. The first chirp is used for measurements below 996 m with a range resolution of 25.5 m, a Nyquist velocity of 10.7 m s−1, and a Doppler velocity resolution of 0.021 m s−1. The second chirp covers ranges from 996 m to 3577 m with the same range resolution, a Nyquist velocity of 5.1 m s−1, and a Doppler velocity resolution of 0.020 m s−1. The temporal resolution of the radar observations is 3.35 s.
The radar is protected with an oval-shaped radome constructed from a foam-like material with a water-repellent outer coating. The radome protects the antenna from environmental conditions while minimizing signal losses under dry conditions. During rainfall, however, water can accumulate on the surface and penetrate the radome material, resulting in additional attenuation of the radar signal. More detailed information on the radome material properties is not available, as these specifications are not provided by the manufacturer.
2.2 Disdrometer
An OTT ParSiVel2 laser disdrometer (hereafter Parsivel, Tokay et al., 2014; Löffler-Mang and Joss, 2000) is used for DSD observations. The drops are detected using a laser with a wavelength of 650 nm that covers an area of 180×30 mm (length × width), resulting in a measurement area of 54 cm2. As droplets cross the laser beam, both their size and fall speed are recorded. These measurements are subsequently categorized into one of the 32 predefined size and velocity classes. The classification of drop sizes depends on whether the type of precipitation is in a liquid or solid state. The lowest and highest measurable range of drop sizes measured for liquid precipitation ranges from 0.2 to 8.0 mm, while for solid precipitation the range is from 0.2 to 25.0 mm. However, the instrument's minimum detectable drop size is 0.312 mm (class 3). Consequently, although size classes are defined down to 0.2 mm, the two smallest classes lie below the detection threshold and are therefore not effectively observed. Thus, the smallest reliably detected particles correspond to class 3. The lowest and highest classes for the fall speed of drops range from 0.2 to 20.0 m s−1. In both the size and velocity classes, the bin width incrementally increases every 5 classes starting at class 10. The bin width begins at 0.125 mm for the smallest diameters, and subsequently doubles every 5 classes, reaching up to a width of 2.0 mm. The width of the two largest classes is 3.0 mm (OTT HydroMet GmbH, 2017). Furthermore, this study uses the categorization of precipitation type by precipitation codes, the number of detected particles, the number density, the drop velocity distribution and the raw data (i.e., the original, unprocessed size–velocity matrix).
Accurate estimation of rainfall DSD parameters from Parsivel observations requires enough data to ensure statistical reliability and representativeness. The proposed method applies only to rain and drizzle cases. Consequently, observations are filtered using SYNOP precipitation codes, codes 51–53 for drizzle; 57–58 for drizzle with rain; and 61–63 for rain, as defined in Table 4680 (OTT HydroMet GmbH, 2017, Appendix D). To ensure accurate DSD parameter estimation, a minimum threshold for the number of detected particles is set. This guarantees that a sufficient number of droplets is present across as many drop size classes as possible. In this study, the threshold is set at 100 particles. For the same purpose, droplets should be present in at least five consecutive size classes without falling more than two classes beyond this continuity. Size classes not meeting this criterion are seen as outliers and are removed. As Parsivel disdrometers are less sensitive to smaller droplets, size classes with median diameters below 0.4 mm are omitted. A closer examination of the remaining data revealed certain features that are not consistent with those of rain, such as insects or side fallers, which deviate in size and fall velocity from typical drops. Raindrops follow a specific velocity-size relationship, as defined by Atlas et al. (1973). A limit is set at the 0.5–1.5 percentile of this velocity-size curve for retaining size and velocity classes, with any falling outside being discarded.
2.3 Precipitation gauge
The OTT Pluvio2 weighing precipitation gauge with an opening diameter of 400 mm was used for precipitation observations during the two analyzed events. The gauge uses a double alter wind shield to minimize the impact of wind on the observations. The gauge records measurements of the bucket weight at 1 min intervals in both real-time and non-real-time. The non-real-time product is derived using a 5 min integration period, resulting in higher measurement accuracy (OTT HydroMet GmbH, 2014). This study uses the non-real-time 1 min data product. From these records, precipitation intensity and accumulation are computed.
2.4 Temperature and humidity observations
Temperature and relative humidity observations were collected at the radar field (1.5 m a.g.l.), and multiple heights at the SMEARII mast, namely at 16.8, 35.0, 67.2, and 125.0 m a.g.l. Given the difference in ground levels, the highest altitude temperature and humidity measurements were taken about 151 m above the radar (mast 125 m + 26 m elevation). This aligns with the third range gate of the HYDRA-W.
3.1 Difference in heights of radar and disdrometer sampling volumes
In this study we are comparing radar variables computed from disdrometer observations to radar measurements. While the instruments are collocated, there is a vertical distance between the sampling volumes of the instruments. The lowest radar gate, in our case, is at 102 m, while the disdrometer measures at ground level. In this section we address the choice of the lowest range gate used for the study and what factors affect comparison of disdrometer and radar observations. There are several factors that should be taken into account before disdrometer and radar observations can be compared. Such as near-field effect (Sekelsky, 2002) and parallax errors (Sekelsky and Clothiaux, 2002), which can affect reflectivity measurements at close ranges. Additionally, droplet evaporation can change DSDs between radar and disdrometer observation volumes.
It is expected that the radar reflectivity values in radar, expressed in dBZ scale, will decrease linearly with height. The slope of this line should be equal to the rain specific attenuation. Analysis of the radar reflectivity profile shows that this is not the case for lowest three range gates (102, 127 and 153 m). Instead, reflectivity increases from the first to the third gate by approximately 1 dB, after which it begins to decrease with height. This feature appears consistently in most, if not all, radar reflectivity profiles obtained from HYDRA‑W measurements in the entire observational dataset. Myagkov et al. (2020) have reported a similar feature in other W-band RPG FM-CW cloud radars. A possible explanations could be that the observations are not corrected for the near-field effect (Sekelsky, 2002), parallax errors (Sekelsky and Clothiaux, 2002), or evaporation (Myagkov et al., 2020; Tridon et al., 2017).
The radar software computes compensation for the parallax error, i.e. error caused by not a complete overlap of receiver and transmitter antenna beams, by taking into account the distance between the antennas and their beam widths. Sekelsky and Clothiaux (2002) derived a formula to calculate the parallax error for a dual-antenna radar. Under nominal alignment, this correction is about 3 dB for the lowest gate and about 1.3 dB for the third gate. However, the correction is particularly sensitive to antenna pointing accuracy at short ranges, where the beam overlap changes rapidly with distance. For example, a residual pointing offset of 0.05°, chosen here as a plausible example, increases the correction factor to approximately 4 dB for the lowest gate and 2 dB for the third gate. Although the radar software automatically computes the parallax correction, small residual pointing errors may still affect the result. As demonstrated by the example above, even minor pointing inaccuracies can lead to changes in the correction factor at the lowest range gates and consequently introduce noticeable uncertainties in the corrected reflectivity. For range gates at and above the third gate, the correction is comparatively small and less sensitive to pointing errors. Therefore, while residual parallax errors may contribute to the anomalous reflectivity behaviour observed in the lowest gates, their impact is expected to be limited at greater ranges. Based on this analysis, the subsequent analysis is restricted to the third range gate and above.
For the HYDRA-W system, W-band radar with 0.5 m antenna reflector, the near-field region extends up to 156 m from the radar (Sekelsky, 2002). So the lowest three gates are located within the near-field. In the near-field the observed reflectivity factor is underestimated. In Sekelsky (2002) equations for computing the reflectivity bias caused by antenna near-field are presented. Using these equations the correction factors were computed. These are 0.14 dB for the first gate and 0.05 dB for the third gate and are too small to explain the observations.
A final explanation for the decrease in reflectivity below the third gate is rain evaporation. To investigate how evaporation affects reflectivity, we extrapolated the DSD measured by Parsivel at ground level to the height of the third gate. This is done by applying the droplet growth equation (Lamb and Verlinde, 2011) to observed temperature and relative humidity profiles, which are measured by weather stations on the mast and radar field.
The impact of evaporation on DSDs can be derived from Maxwell's mass growth law (Lamb and Verlinde, 2011). The mass of a droplet is related to its radius through , where md and rd are the mass and radius of the droplet, respectively, and ρl is the mass density of liquid water. Hence, Maxwell's law for mass growth can be reformulated in terms of the droplets' radius:
where fv is the ventilation coefficient, G acts as a diffusivity with units m s−1 and s along with sk are the ambient supersaturation and the Köhler function supersaturation, respectively. sk is assumed to be zero for raindrops. Subsequently, Eq. (1) can be rewritten in terms of the rate of change in radius. The last step involves the derivation of an expression for dt, using , where v is the terminal velocity of a droplet in m s−1 and dH is the height interval in m. The height interval is set to steps of 10 m, from which the closest height with available temperature and relative humidity data are found. The terminal velocity is assumed to follow the form presented in Atlas et al. (1973) (Eq. 2), as the droplet is expected to reach its equilibrium terminal velocity near the ground.
where v is the terminal velocity in m s−1 and D is the diameter of the droplet in mm. The Parsivel bin sizes are adjusted every minute to account for evaporation, which consequently changes the bin sizes and causes a shift in the DSD. It is assumed that evaporation affect only the bin sizes themselves, while the number of particles within each size bin remains constant.
Maxwell's law indicates a preference for growth among smaller droplets, which makes them more susceptible to evaporation. Under relatively dry conditions (75 % relative humidity at 10 °C), droplets falling from the third radar range gate (153 m a.g.l.) to the surface decrease in diameter by roughly 0.05 mm for the smallest measurable droplets and in the order of 10−3 mm for droplets with diameters of 10 mm. These changes correspond to reflectivity reductions of about 0.2–0.7 dB at respectively 10 and 15 dBZ, depending on the shape of the DSD. However, such dry conditions are typically limited to the onset of precipitation. As rainfall develops, relative humidity increases rapidly (e.g., to >95 %), which suppresses evaporation and reduces its impact on reflectivity to approximately 0.1 dB. These results indicate that evaporation generally causes a slight decrease in reflectivity toward the surface. However, its magnitude is small and primarily limited to the early stages of rain events. Therefore, evaporation explains only a minor fraction of the observed reflectivity variations in the lowest radar range gates.
For the events presented in this study, evaporation effects are negligible, and the near-field effect is too small to explain the observations in the lowest three gates. A more plausible explanation is a small antenna pointing misalignment, which increases the parallax error correction. This effect is strongest at short ranges, where beam overlap is incomplete, and decreases with increasing height as the overlap improves. To quantify the sensitivity, a slightly larger pointing offset of about 0.07° would increase the correction at the third range gate (153 m) from approximately 1.3 to about 2.3 dB, corresponding to a 1 dB increase in reflectivity. Even in this case, the resulting uncertainty at the third range gate remains on the order of 1 dB. Therefore, in this study, radar measurements from the third range gate are used.
3.2 Computation of DSD parameters and radar variables
To characterize DSDs, the normalized gamma function (Testud et al., 2001; Illingworth and Blackman, 2002; Bringi and Chandrasekar, 2001) given in the following form is used:
where Nw is the intercept parameter of a normalized gamma distribution expressed in , while D and D0 represent the equivalent volume diameter and the median volume diameter, respectively, both measured in mm. The last parameter is μ which is a unitless shape parameter of the DSD. The three parameters that characterize the DSD, Nw, D0, and μ (Eq. 6), are calculated using the third and fourth moments (Eq. 4) of the distribution (Bringi and Chandrasekar, 2001).
where Dm is the mass-weighted mean diameter, σm is the standard deviation of the mass-weighted mean diameter which is a measure of the DSD width.
Computations of radar variables are performed using the PyTMatrix python package (Leinonen, 2014; Mishchenko and Travis, 1994; Mishchenko et al., 1996; Wielaard et al., 1997), using the DSD parameters as input. The DSD integration is performed for D ranging from 0 to 2.5D0 (Carey and Petersen, 2015). In addition to reflectivity, the mean Doppler velocity is also computed. The computed and observed mean Doppler velocity were used as an additional check of our computations and validity of the presented disdrometer-radar comparisons. While this comparison is not presented, the agreement between computed and observed mean Doppler velocity is excellent. The mean Doppler velocity is not directly computed by PyTMatrix. It was calculated using the following equation:
where “mdv” is the mean Doppler velocity, v(D) is the terminal velocity in m s−1 of a droplet estimated by Eq. (2) taken from Atlas et al. (1973), σ is the radar cross section m2, and θ is the elevation angle in radians. The integration limits Dmin and Dmax are 0 and 2.5D0 respectively. The radar cross sections were computed by PyTMatrix and saved as lookup table.
3.3 Forward model of disdrometer measurements
In this study, the method for computing DSD parameters, radar variables and associated uncertainty range is proposed. This problem can be formulated as follows: for a given DSD observation, what is the corresponding range of true, intrinsic, DSDs. For simplicity sake, we will assume that DSDs follow a Gamma functional form and therefore can be described using three parameters, as presented in the previous section. This assumption is supported by Gatidis et al. (2020), who evaluated the adequacy of the Gamma model for fitting DSDs. Their results show that, at a temporal resolution of 60 s, a substantial fraction of observed DSDs can be well represented by a Gamma distribution, although deviations may occur, particularly during transitional rainfall periods. The intrinsic DSD parameters are “true” parameters describing rainfall that we aim to measure. Because of measurement uncertainties, one intrinsic DSD can correspond to an ensemble of observed DSDs. And vice versa, when we go from the observation space to the intrinsic DSD space, one observed DSD can be linked to an ensemble of intrinsic DSDs.
The forward model that describes disdrometer observations by combining statistical uncertainties of DSD measurements and potential hardware limitations links intrinsic and observation DSD spaces. Within an integration time, the number of observed raindrops is defined by a sampling volume and number concentration of raindrops. The sampling volume is determined by disdrometer technical characteristics. For example, for a Parsivel disdrometer the sampling area is 180×30 mm. The sampling volume therefore is equal to the sampling area times raindrop velocity times the integration time. Because the raindrop velocity depends on droplet size, the sampling volume is size dependent. To obtain an analytical expression for this dependence, we use the velocity–diameter power-law relation of Atlas and Ulbrich (1977), v(D)=3.78D0.67. This formulation is convenient for use in combination with Eq. (3). The expected number of observed droplets during an integration time, Δtint, for a disdrometer with a sampling area, A, can be estimated as
where Nw, D0 and μ are parameters of an intrinsic DSD. The number of observed raindrops in a volume is a random variable following a Poisson distribution with the expected value Ntv (Chandrasekar and Bringi, 1987; Wong and Chidambaram, 1985). Given the number of observed droplets, a simulated DSD is derived by sampling from a Gamma probability density function defined by the intrinsic DSD parameters.
As the intrinsic DSD space should be able to capture as many as possible observed DSDs, the forward model must be applied to a range of DSD parameters. This range should mimic local climatology or cover a wide enough range of DSDs. In this study, observations of DSD parameters collected in Hyytiälä between 2014 and 2024 are used to define this range. The observed distributions of the DSD parameters, namely D0, μ and log10(Nw) (see Fig. 1) resemble those derived over a 5-year period at another location in Finland (Leinonen et al., 2012). Note that a different filtering method is applied in this study. When comparing the cumulative density function (CDF) at these two locations, a CDF of 0.8 is reached at around the same values for the DSD parameters. In Hyytiälä, this corresponds to a D0, μ and log10(Nw) of 1.1 mm, 10 and 3.8×103 , compared to the values reported by Leinonen et al. (2012), which are 1.1 mm, 12 and 3.6×103 , respectively.
Figure 1The probability density function (PDF) and the cumulative density function (CDF) of (a) the median volume diameter (D0), (b) the shape parameter (μ), and (c) the normalized intercept parameter (Nw). Data from Hyytiälä spanning the years 2014 to 2024 are used. Similar as in Figs. 4, 5 and 6 in Leinonen et al. (2012), but with a different filtering method. The mean, standard deviation and skewness for each DSD parameter is shown at the top right.
Differences emerge only in D0 values below 0.7 mm, together with a minor increase in values of log10(Nw). The probability density functions shown in Fig. 1 are used as inputs in the forward model. To avoid that the intrinsic DSDs are sampled from a climatology containing measurement uncertainties, the distributions are slightly expanded. This also ensures that rarely occurring DSD parameter combinations are included in the simulations. However, given the strong attenuation of W-band radar in heavy rainfall, the sampling of DSDs is primarily focused for DSDs associated weak to moderate precipitation.
A skewed normal distribution is fitted to the parameters D0 and μ, as this best represents the shapes of the curves in Fig. 1. For D0, the fitted distribution has a mean of 0.65 mm, a standard deviation of 0.5 mm, and a skewness of 4, with values ranging from 0 to about 4 mm. For mu, the mean, standard deviation, and skewness are −1, 14 and 10 respectively, with values ranging from −2 to about 60. In contrast, for log10 (Nw), a uniform distribution is fitted in between the range 2 to 5 log10 (mm−1 m−3). This choice is motivated by the need to sample the full observed range of droplet number concentrations, which spans approximately three orders of magnitude. Using a uniform distribution ensures that both low and high values are equally represented, whereas a normal distribution would under-sample the tails. Furthermore, physical constraints are established for a reflectivity of <55 dBZ and rain rate of <300 mm h−1 (Chandrasekar and Bringi, 1987). Intrinsic DSDs are formed by drawing a random sample of all three parameters.
The forward model procedure can be described by the following steps:
-
Step 1: Select intrinsic D0, μ and Nw by taking a random sample of the observed distributions. Check that the combination of the DSD parameters is reasonable, namely that the reflectivity is less than 55 dBZ and the rain rate is less than 300 mm h−1 (Chandrasekar and Bringi, 1987; Moisseev and Chandrasekar, 2007).
-
Step 2: Compute expected Ntv and sample the Poisson distribution to obtain the number of droplets that will be used to simulated DSD.
-
Step 3: Using the simulated number of droplets, derive their diameters by sampling a Gamma pdf defined by the intrinsic D0, μ and Nw.
-
Step 4: Apply filtering to the simulated DSD, i.e. data quality filtering that is applied to the disdrometer observations. Apply small diameter threshold, defined by the disdrometer sensitivity. A Parsivel disdrometer will not record droplets smaller than 0.3–0.4 mm in diameter. To simulate this, we reject simulated DSD values for diameters less than 0.4 mm.
-
Step 5: Compute DSD parameters from the simulated DSD
-
Step 6: Repeat Steps 2–5 100 times. We have decided to repeat this calculations 100 times to obtain enough statistics linking the intrinsic DSD and simulated DSD space.
The forward model simulations produce, for each intrinsic DSD, a corresponding ensemble of simulated DSD parameters, as schematically depicted in Fig. 2a. In our study, each intrinsic DSD defines an ensemble of 100 simulated DSD parameter sets. For each ensemble, we compute an enclosing ellipsoid (dashed curve in Fig. 2a), whose center is defined by the median values of the simulated DSD parameters, as shown by the circular marker in the figure. Because the forward model's parameters are chosen to match disdrometer observations, the simulated DSDs are designed to reproduce the disdrometer's measurement and sampling uncertainty.
Figure 2Schematic illustrating the steps taken in the forward (a) and inverse (b) models. The forward model relates a single intrinsic DSD to an ensemble of simulated DSDs (blue stars). The inverse model links the disdrometer observed DSD (black dot) to an ensemble of intrinsic DSD via the medians of the simulated ensemble (purple dots). The DSD parameter radius (Eq. 9) is indicated with orange arrows and the mapping is represented by the gold arrow.
3.4 Inverse problem; from observations to an ensemble of intrinsic DSDs
The forward model establishes a relationship between an intrinsic DSD and a DSD that mimics observations. The purpose of the inversion is to link an observed DSD with a set of intrinsic DSDs that, given the measurement uncertainty, could have produced that observation. At the forward stage, for every intrinsic DSD we define a three dimensional parameter space (D0, μ, Nw), described by the corresponding ensemble of simulated DSD parameters. This space is represented as an ellipsoid, characterized by a center, the median of the simulated DSD parameter values, and semi-axis lengths given by the standard deviation of each parameter.
Because the DSD parameters span very different ranges, each is rescaled to the interval [0, 1] using min-max scaling, . The minimum and maximum for each DSD parameter are taken across the simulated DSDs for all intrinsic DSDs. This results in a normalized ensemble of simulated DSDs, from which the median and standard deviation are computed. The same normalization is applied to the observed DSDs. In our case, the minimum and maximum values found for D0, μ, and Nw are 0.48 to 5.45, −2.59 to 125.5, and 25.68 to 1.07×105, respectively.
The spread within each ensemble of simulated DSDs, its standard deviation, depends on the number of droplets (Fig. 3). To capture this relationship, a second degree polynomial is fitted between the standard deviation of each DSD parameter and the number of droplets Ntv, calculated using Eq. (8). The resulting relations are given below:
where x denotes the number of observed droplets.
Figure 3Drop size distribution parameter ellipsoid radii estimation for the number of drops in log-scale. Radii estimation is computed from the normalized standard deviations of simulated DSDs for (a) D0, (b) μ, (c) Nw. In (a)–(c), the dotted line represents the fitted second degree polynomial and the colors show the number of simulations. The DSD parameters polynomials are shown together in (d), where D0 (blue line), μ (orange line), and Nw (green line) are displayed.
Because the forward model includes statistical uncertainties, DSD truncation, and filtering, the inverse problem can be used to estimate the statistical uncertainties and possible biases of the resulting DSD parameters and radar variables. The inverse problem is formulated in three steps:
-
Step 1: For the observed DSD, with its associated number of drops, an expected uncertainty is defined using the standard deviations given by Eq. (9). These standard deviations define the semi-axes of a three-dimensional ellipsoid, centered on the observed DSD parameters, in the (D0, μ, Nw) space.
-
Step 2: The medians of all simulated DSD ensembles are checked against this ellipsoid. Since each median corresponds to one intrinsic DSD, the medians that fall within the ellipsoid identify the intrinsic DSDs consistent with the observation, yielding an ensemble of intrinsic DSDs associated with the observed DSD.
-
Step 3: The statistical properties (e.g. median and spread) and possible biases of the computed DSD parameters and radar variables are estimated from this ensemble of intrinsic DSDs.
This procedure is applied independently to each observed DSD. As a result, a single intrinsic DSD can be linked to multiple observations. A schematic representation of this methodology is provided in Fig. 2b.
Figure 4 presents an example of an observed DSD, red line, fitted DSD using DSD moments, green line, and estimated DSD using median values of ensembles of intrinsic DSD parameters, blue line. The boxplots and histograms represent the distribution of intrinsic DSD parameters in this ensemble and define the associated uncertainties. It should be noted that the green and blue curves fit the observations well, the main differences appear in the small diameter range, D<0.5 mm. Because of the disdrometer sensitivity, there are no measurements below 0.3 mm, as per manufacturer specifications, see (Moisseev and Chandrasekar, 2007), for example, for discussion on how it would affect DSD parameter estimation. The under sampling of smaller droplets can also be seen in the ensembles of estimated intrinsic DSD parameters. The median volume diameter computed using DSD moments is larger than most intrinsic D0, (4d). Because intrinsic D0 values are smaller and computed reflectivity factor and rainfall rate values, (Fig. 4b–c), are not too different from the values estimated from the directly fitted DSD, the intrinsic Nw values are somewhat larger Fig. (4f). It should also be noted that the estimated intrinsic μ is smaller than the one directly computed from the observed DSD (Fig. 4e). Using the ensemble of intrinsic DSDs, biases in DSD parameters and radar variables are estimated by comparing the ensemble statistics (e.g. median) with the values derived directly from the observed DSD.
Figure 4An example of an observed DSD (red line) with its fitted DSD using DSD moments (green line) and corresponding ensemble of intrinsic DSDs (black boxes) and ensemble median (blue line) in (a). Histograms (b)–(f) show the ensemble of intrinsic DSDs (gray) with its ensemble median (blue line) and the DSD fit using DSD moments (green line). From (b) to (f) are respectively computed reflectivity, rain rate, D0, μ, and Nw. The ensemble size of intrinsic DSDs corresponding to the observed DSD is displayed on the bottom right.
As shown in Fig. 4b–c, ensembles of intrinsic reflectivity and rain rate can also be obtained. Reflectivity and rain rate are computed individually for each DSD in the intrinsic ensemble using PyTMatrix (Leinonen, 2014). The resulting distributions are then used to quantify the associated uncertainties (e.g. spread). The number of simulations shown in Fig. 4 correspond to the number of intrinsic DSDs in the ensemble that satisfy the selection criteria described stepwise above, i.e. those consistent with the observed DSD within its uncertainty bounds. This number provides an indication of whether the ensemble is sufficiently large to yield robust uncertainty estimates. For measurement stations equipped with precipitation gauges, both precipitation rate and accumulation can be used as an additional verification step for the DSD-derived quantities. Because the uncertainties associated with the ensemble-based rain rates are temporally correlated, the variances of accumulated precipitation cannot be obtained by simple summation over time.
The propagation of uncertainty is therefore computed using the quadratic form
where x represents the precipitation rate time series, a is a row vector defining the linear accumulation operator and ∑x is the variance–covariance matrix of x (Mardia et al., 1979). For the case of accumulated precipitation over t timesteps, a consists of ones, representing the summation of precipitation rates over time. In this study, the diagonal elements of ∑x are estimated from the ensemble-based variance at each timestep, while the off-diagonal elements represent temporal covariances. Because the ensemble data are generated independently at each timestep, they do not contain information on temporal dependence. Therefore, the off-diagonal elements are constructed using a lag-dependent autocorrelation function derived from the precipitation time series.
For accumulated precipitation, the quadratic form can be written equivalently as
where is the variance of the accumulated precipitation over t time steps, σi and σj are the standard deviations of rain rate at time steps i and j, and is the autocorrelation coefficient between these timesteps, accounting for temporal dependence in rainfall. This formulation separates the contributions from individual variances and the covariance terms arising from temporal dependence.
Two distinct attenuation mechanisms are considered in this study. The first is wet-radome attenuation, caused by water accumulation on and within the radar radome. The second is rain-path attenuation, caused by attenuation of the radar signal as it propagates through precipitation. Wet-radome attenuation is estimated independently of rain-path attenuation and is corrected first, after which the rain-path attenuation retrieval is applied. Estimation of the radome attenuation is based on a comparison of computed and measured reflectivity factors, at the third radar range gate (153 m). If the difference is larger than the expected uncertainty, radome attenuation is identified and computed. Small negative estimates of radome attenuation may occasionally occur when the measured reflectivity slightly exceeds the computed reflectivity.
For the estimation of rain path attenuation, the specific attenuation is computed from the intrinsic DSDs using PyTMatrix calculations. Under ideal conditions, namely stratiform rainfall below the melting layer, precipitation properties vary only gradually with height and the specific attenuation can be assumed to be approximately constant. The proposed retrieval is therefore expected to perform best under stratiform precipitation conditions, where the surface-derived DSD remains representative of the precipitation column. In such cases, attenuation is expected to be the dominant contributor to the decrease in reflectivity with range, and the slope of the reflectivity profile should therefore be equal to the specific attenuation derived from the disdrometer observations. The path-integrated attenuation is then obtained by integrating the specific attenuation along the radar propagation path below the melting layer. The method may become less reliable in the presence of embedded cloud layers associated with seeder–feeder processes or other situations where significant microphysical changes occur with height.
Two rain events recorded on 26 May 2022 and 28–29 July 2023 are presented to illustrate the proposed method, see Figs. 5 and 6. The main difference between these events is the radome attenuation. On 26 May 2022 the radome attenuation is negligible, while during the event on 28–29 July 2023 the radome attenuation approaches 10 dB. In both cases rain lasted for extended periods of time.
Figure 5Time-height plot of (a) HYDRA-W radar reflectivity factor. Time series of the (b) radar reflectivity factor at 3rd range gate and of (c) rainfall rate (left side) and accumulated precipitation (right side). The time is set to capture the event between 01:00 and 20:00 UTC. In (b), the reflectivity factor is shown together with the Parsivel computed reflectivity (cyan line) and the ensemble of intrinsic DSDs median (black line) and standard deviation (SD) (±1σ darker shading and ±2σ lighter shading). (c) displays the rainfall rate and the accumulated precipitation with similar color coding as in (b), but with Pluvio (green line) instead of radar. The total precipitation of the event is shown at the top of (c). The inset axis in (b) provides a detailed look between 09:00 and 10:30 UTC. The method could not be applied at the times indicated by the magenta vertical lines in (b) and (c).
Figure 6Same as in Fig. 5, but for the event on 28–29 July 2023 between 09:00 and 17:00 UTC.
On 26 May 2022 the stratiform precipitation started around 01:00 UTC and continued past 20:00 UTC, see Fig. 5. During this event the largest fraction of precipitation accumulation occurred between 05:00 and 12:00 UTC. Figure 5b–c show DSD based computations as compared to radar and precipitation gauge observations. For the radar reflectivity factor comparison, radar observations at the 3rd range gate (153.3 m) are compared to disdrometer based estimates. Both estimates based on direct disdrometer observations and intrinsic ensembles are shown. The proposed approach allows estimation of the uncertainty range as depicted by shading and is expressed as the range corresponding to ±1 and ±2 SD. For the radar reflectivity factor the uncertainty range is relatively small with a standard deviation of around 0.3 dB. For the precipitation accumulation the uncertainties accumulate and can be seen in the increasing range as shown by the red shaded area.
The event on 28–29 July started with a brief period of convective precipitation on the 28th around 14:00 UTC, which was followed by an extended period of stratiform precipitation, with intermittent moderate rainfall, lasting until about 13:00 UTC on the 29th, see Fig. 6. The stratiform precipitation can be divided in two phases: approximately two thirds of the accumulated stratiform precipitation occurred on the 28th between 14:00 UTC and midnight, while the remaining one third occurred from midnight till 13:00 UTC on the 29th (Fig. 6c). Uncertainty in the radar reflectivity factor during this event is comparable to that observed for the 26 May event.
To allow a comparison of computed and observed radar variables, data need to be corrected for a potential time lag. This time delay is identified as the lag where a maximum of the cross correlation between radar measured and disdrometer derived reflectivity factor values is observed. For both events, this lag was found to be about 1 min.
4.1 Radome attenuation
An estimate of the wet radome attenuation is obtained by computing the difference between radar reflectivity at the lowest usable range gate (3rd gate) and the “median intrinsic reflectivity”. This is illustrated for an event with minimal radome attenuation and one with a degraded radome resulting in strong radome attenuation. The selection of events is performed to ensure that there is a prolonged period of continuous rainfall. The minimal attenuation case corresponds to 26 May 2022, when the radome had been replaced one day prior to the event and can therefore be considered to be in near-optimal condition. In contrast, the second event (28–29 July 2023) shows pronounced attenuation and occurred when the radome had been in use for approximately one year. On the 26 May the light rain (<2.5 mm h−1) starts at 05:21 UTC and ends at 12:38 UTC (Fig. 5c, left axis). Before the start of the continuous rain, some light drizzle occurred. To assess the effect of the continuous rain on the radome, we compare the radar (in orange) and intrinsic (in black) reflectivity values (Fig. 5b). These two lines match almost exactly, indicating that the radome attenuation is negligible. The computed radome attenuation is shown in Fig. 7c (right side). Throughout the rain event, the radome attenuation did not exceed 1 dB for 82 % of the time.
Figure 7Time-height plot of HYDRA-W (a) uncorrected and (b) corrected radar reflectivity factor. Time series of the (c) specific attenuation observed by disdrometer (cyan line) and median (black line) and standard deviation (SD) (±1σ darker shading and ±2σ lighter shading) of the ensemble of intrinsic DSDs, left side, and magnitude of the radome correction applied to the reflectivity factor (orange line), right side. In (d), the two methods used to verify whether attenuation correction is applicable are presented. On the left, the percentage of slope falling within ±2 SD (blue line), while on the right, the difference below the melting layer (red line) is shown. Both lines are averaged over 5 min instead of the 1 min used in the method. The time span is identical to that in Fig. 5.
In contrast, the second event demonstrates significant radome attenuation, as can be seen in Fig. 6. The rain starts on the 28th at 13:46 UTC and ends on the 29th at 13:02 UTC. At the beginning of the event, heavy rainfall (reaching 53 mm h−1) is observed, resulting in rapid accumulation of water on and within the radome. The resulting continuous water coverage and water retention within the radome material causing strong radome attenuation (Fig. 6c). This strong attenuation is also visible in Fig. 6a, as indicated by a sharp drop in reflectivity. Following the heavy rain, there was a prolonged period of light to moderate precipitation. By comparing the radar reflectivity (in orange) with the intrinsic reflectivity (in black) shown in Fig. 6b, it becomes apparent that the radome became wet at the point where the two lines begin to diverge. This difference is maintained throughout the event. The resulting radome attenuation is shown in Fig. 8c (right axis). Prior to the rain shower, the radome attenuation is minimal, but it increases significantly to a range between 7 and 8 dB once the heavy rain begins. Once the intense precipitation ceased, the rain rate diminishes, enabling the radome to gradually dry, which is reflected by a decline in radome correction. With the return of moderate rain, attenuation values increase gradually to approximately 10 dB. At the end of the rainfall, the radome dried quickly, and the attenuation values returned to levels observed at the beginning of this precipitation event. Similar radome attenuation values around 10 dB at the W-band wavelength have been reported by Myagkov et al. (2020) and Hogan et al. (2003).
Figure 8Same as in Fig. 7, but for the event on 28–29 July 2023 between 09:00 and 17:00 UTC.
The assessment of radome condition in this study is based on both maintenance history and observed attenuation behaviour. The radome was newly installed prior to the May 2022 event, whereas in July 2023 it had been in use for approximately 1 year. In practice, the radome is replaced approximately every 6 months, with periodic checks for signs of wear, such as surface degradation or damage, in the meantime. The methodology presented here provides a complementary, quantitative means of monitoring radome performance and may support condition-based replacement strategies in operational settings.
It is important to note that increased attenuation during the July 2023 event cannot be attributed solely to the radome condition. High rainfall intensities can lead to rapid wetting of the radome and thus enhanced attenuation. Under normal operation, the radome is expected to dry during periods of low or no rainfall due to the blower. However, during this event, elevated attenuation persists even after rainfall intensity decreases substantially (around 16:00 UTC), indicating prolonged wetting of the radome. This suggests that both rainfall intensity and radome condition contributed to the observed attenuation.
It should be noted that during the second event, there is also a notable difference in observations of precipitation accumulation. The weighing gauge measures about 47 mm of total precipitation accumulation during the event and disdrometer derived accumulation is only 42 mm. A closer inspection of the event (Fig. 6c) indicates that the discrepancy increase during periods with rainfall rates exceeding approximately 3 mm h−1. In contrast, during the May 2022 event (Fig. 5c), where rainfall intensities remain below this threshold, the differences are generally within the expected uncertainty range. This behaviour is likely associated with sampling limitations of the disdrometer, particularly the underrepresentation of larger droplets at higher rainfall rates. While the applied correction method accounts for truncation effects in the small-drop regime, it is less sensitive to biases related to larger drop sizes, which may contribute to the observed difference in accumulated precipitation.
4.2 Specific attenuation
In addition to the radome attenuation, the rain path attenuation can be computed if certain conditions are met. If the changes in reflectivity values with height in rain are mainly caused by attenuation, the slope of the reflectivity profile is equal to the specific attenuation. Using disdrometer observations, specific attenuation can be computed and compared to radar observations. For each DSD observation specific attenuation is computed from the intrinsic ensemble, which takes into account the uncertainties arising from the relatively small sampling volume. The median of the ensemble defines the expected value of the specific attenuation. Additionally, the standard deviation of the computed specific attenuation values is used as uncertainty range.
The attenuation correction consists of two main steps. After the radome attenuation is corrected, we compute the expected profile of the radar reflectivity. The profile spans from the third range gate up to the bottom of the melting layer. The melting layer boundaries are estimated using linear depolarization ratio (LDR) observations (Li and Moisseev, 2020). Although the LDR observations are not shown, the melting layer can be readily identified in Figs. 5a and 6a by the jump in reflectivity, located at approximately 2 km altitude, from higher values below the melting layer to subsequent lower values above it. The expected profile of the reflectivity expressed in dBZ units is modeled as a linear function with a slope equal to the specific attenuation. In addition to the expected reflectivity profile, the confidence region is also computed. At the second step, the computed and observed reflectivity profiles are compared to assess if the attenuation correction can be applied.
The corrected radar reflectivity shown in Figs. 7b and 8b is obtained in two steps. First, the retrieved radome attenuation is added to the observed radar reflectivity. Subsequently, the rain-path attenuation correction is applied by integrating the specific attenuation derived from the disdrometer observations from the third range gate to each range gate below the melting layer. The resulting reflectivity field represents the radar observations corrected for both radome and rain-path attenuation.
There are several reasons that can cause differences between observed and derived profiles. The radio waves are attenuated by the vertical precipitation column. Given that we observe DSD properties at the ground, we assume that the surface-derived DSD observations are representative of the entire precipitation column below the melting layer. Spatial and temporal variations in the precipitation intensity combined with wind shear could results in fall streaks which would affect representativeness of surface observations. Additionally, changes in precipitation intensity due to evaporation or collision-coalescence growth are also not captured by disdrometer observations.
In Fig. 9 the proposed attenuation correction method is illustrated based on observations collected on 26 May 2022. The figure shows observations of radar reflectivity factor and ceilometer attenuated backscatter. During this event, the specific attenuation varies between 0.5 to 3.0 dB km−1 see Fig. 7c (left axis). Four profiles observed at 06:05, 08:50 10:30 and 14:00 UTC were selected for a more detailed comparison of the observed, black line, and modeled, green dashed line, reflectivity profiles. The reflectivity factor values computed from the disdrometer observations are shown by the error bars. The median reflectivity value matches radar observations at the third gate. The uncertainty range for the specific attenuation are depicted by red and blue dashed lines, corresponding to ±2 SD. Because reflectivity and specific attenuation values are correlated, higher reflectivity values correspond to higher specific attenuation values. This effect can be seen in the difference of the slopes of the modeled profiles for the selected times. Furthermore, the slopes of the red and blue dashed lines, which correspond to lower and higher ranges of uncertainty, are different, because they originate from the upper and lower bounds of the estimated reflectivity values.
Figure 9Time-height plot of (a) HYDRA-W radar reflectivity factor and (b) ceilometer attenuated backscatter. The time span is identical to that in Fig. 5. Reflectivity profiles (c)–(f) illustrate the observations at 06:05, 08:50 10:30 and 14:00 UTC, respectively. The radome corrected reflectivity profile (black) is shown together with the ensemble of intrinsic DSDs, including the median (green dotted line) and the upper and lower ±2 SD bounds (red and blue dotted lines). The ensemble spread is additionally represented as error bars indicating the median (gold) and ±2 SD (±1σ darker shading and ±2σ lighter shading). The melting layer is indicated by the horizontal purple dotted line. Cyan and gold dots denote the reflectivity observed by the disdrometer and at the third radar gate, respectively.
We propose two methods to assess whether the attenuation correction can be applied. The first method checks if the observed reflectivity profile lies within the ±2 SD as estimated from DSD observations. If more than 75 % of the observed reflectivity profile lies between lower and upper uncertainty bounds, the two profiles are considered to follow the same slope and attenuation correction can be applied. The 75 % threshold is an ad hoc criterion selected to be sufficiently strict without being too restrictive. As can be seen in Fig. 9, the profiles observed at 08:50 and 14:00 UTC exceed the 75 % overlap threshold and therefore satisfy this criterion, whereas the profiles observed at 06:05 and 10:30 UTC do not. For the latter two profiles, large portions of the observed reflectivity profile lie outside the modeled uncertainty bounds. The difference can be attributed to the seeder-feeder process, where raindrops are growing by collision-coalescence in the embedded cloud layers. These cloud layers are identified in the ceilometer attenuated backscatter observations as regions of enhanced backscatter, for example between 03:00 and 05:00 UTC in Fig. 9b, where the signal becomes fully attenuated below the 500 m altitude. Their presence indicates liquid cloud droplets that can be collected by falling raindrops, resulting in additional precipitation growth. The simultaneous occurrence of these cloud layers and deviations between the observed and modeled reflectivity profiles supports the interpretation that the surface DSD observations are not fully representative of the precipitation column. An example of this behaviour can be seen in the 06:05 UTC reflectivity profile shown in Fig. 9c, where the observed profile (black) decreases more rapidly with height than the modeled profile (green). Figures 7d and 8d (left axis) present the overlap percentage of observed and modeled reflectivity profiles for the two events. As can be seen, for these two analyzed events only for a small fractions of observations the overlap percentage is high enough.
A second, less restrictive criterion is to compare projected and observed reflectivity values just below the melting layer. If the difference is less than 1 dB, which is taken as a representative uncertainty in the radar reflectivity factor, the attenuation correction is considered applicable. The reason behind this criterion is that, in the case of approximately constant attenuation along the propagation path, the reflectivity profile should decrease smoothly with height. Consequently, the projected reflectivity and the observed reflectivity just below the melting layer should not differ significantly. A small discrepancy therefore indicates consistency between the projected and observed profiles. In Figs. 7d and 8d (right axis) the reflectivity difference is plotted. It is less than 1 dB for a large fraction of the 26 May 2022 event. For the 28–29 July 2023 case, even this less restrictive criterion fails for a large fraction of the event. During 28–29 July 2023 rain event there was a persistent low-level cloud, with a cloud base below 1 km. Because of the cloud the surface precipitation observations are affected by the seeder-feeder process and not representative of the whole rain column. The melting layer was at about 2 km for both cases.
As part of the Cloudnet processing chain (Tukiainen et al., 2020), a rain-attenuation correction following Crane (1980) has been applied since September 2024. This correction was implemented in Cloudnetpy version 1.66 (Tukiainen et al., 2024). It is based on a parameterization that relates the rain rate, derived from disdrometer measurements, to the specific attenuation, as presented in Crane (1980). The limitations of this method are essentially the same as for the approach presented in this paper. It is assumed that surface-based observations of rainfall microphysical properties are representative of the entire atmospheric column.
Prior to Cloudnetpy version 1.66 (Tukiainen et al., 2024), no attenuation correction was applied, and consequently, ice water content (IWC) retrievals were not performed in conditions affected by uncompensated rain attenuation. Figures 10a and 11a show the IWC retrieval status reported for the two case studies before the introduction of the attenuation correction in version 1.66. Following the application of the attenuation correction described in this study, the retrieval status was revised using the second attenuation-correction criterion, which compares the projected reflectivity below the melting layer with the observed reflectivity. When the difference is less than 1 dB, the attenuation correction is considered applicable and IWC retrievals can be performed. The updated retrieval status is shown in Figs. 10b and 11b.
Figure 10Time-height plot of the ice water content retrieval status (a) without attenuation correction and (b) after applying attenuation correction using the second method. The time span is identical to that in Fig. 5. Periods with clouds are indicated in magenta, and profiles where attenuation correction is applicable are marked in blue.
Figure 11Same as in Fig. 10, but for the event on 28–29 July 2023 between 09:00 and 17:00 UTC.
To aid the interpretation of the results, the presence of embedded liquid cloud layers is explicitly indicated. These layers are identified from regions of enhanced ceilometer attenuated backscatter below the melting layer. The melting-layer base is obtained from the standard Cloudnet categorize product, which is derived from the attenuated backscatter observations. These layers are flagged starting from the detected cloud base up to the melting layer base, under the assumption that lower cloud layers exert a stronger influence on the attenuation because in such case the observed DSD is only representative of a limited fraction of a vertical profile. This hypothesis is supported by observations from these two events. It can be seen in the figures that the reflectivity matching criterion fails in cases where clouds are present. In some cases where the cloud layers appear closer to the melting layer the attenuation correction can still be applied.
In this study, the uncertainties associated with the application of radome attenuation and rain path attenuation corrections based on disdrometer observations are investigated. It is shown that statistical uncertainties arising from sampling of the drop size distribution (DSD) with disdrometers such as the Parsivel can be quantitatively characterized. This uncertainty quantification framework enables the evaluation of radome attenuation, as attenuation events can be identified when the discrepancy between measured radar reflectivity and DSD derived radar reflectivity exceeds the corresponding uncertainty range. In addition to comparing observed and computed radar reflectivity, precipitation accumulations recorded by a weighing gauge and a disdrometer were analyzed. For one of the events, the observed discrepancy could not be explained by statistical sampling uncertainty alone.
Because radome attenuation can become substantial during intense precipitation, even for weather radars equipped with radome blowers, radar calibration monitoring procedures that rely on disdrometer data should explicitly account for this effect. The use of disdrometer observations for rain path attenuation correction is also examined. Two criteria are introduced to assess whether the disdrometer measurements are representative of the overlying rain column and therefore suitable for use in attenuation correction. Analysis of selected case studies indicates that embedded cloud layers pose a significant challenge for the applicability and reliability of this approach. These findings highlight the need to evaluate the attenuation correction procedures implemented in Cloudnetpy and to better understand their limitations.
The data used in this study are generated by the Aerosol, Clouds and Trace Gases Research Infrastructure (ACTRIS) and are available from the ACTRIS Data Centre using the following links: https://doi.org/10.60656/1fe3a02ef09b419d (Moisseev et al., 2025a), https://doi.org/10.60656/d25b530af3d04367 (Moisseev et al., 2025b) and https://doi.org/10.60656/8d288976932d417e (Moisseev and Petäjä, 2025). The code is available on GitHub at https://doi.org/10.5281/zenodo.24166600 (Mom, 2026).
BM and DM have conceptualized the study. BM developed the methodology with input from DM. BM has implemented the methodology, processed the data, and performed the analysis. Both authors have contributed to writing of the manuscript.
The contact author has declared that neither of the authors has any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. The authors bear the ultimate responsibility for providing appropriate place names. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.
We acknowledge ACTRIS and Finnish Meteorological Institute for providing the data set which is available for download from https://cloudnet.fmi.fi (last access: 12 November 2026). We acknowledge ECMWF for providing IFS model data.
This research has been supported by the HORIZON EUROPE Climate, Energy and Mobility, HORIZON EUROPE Innovative Europe (grant no. 101137680) via project CERTAINTY (Cloud-aERosol inTeractions and their impActs IN The earth sYstem) and Research Council of Finland (decision no. 369601)
This paper was edited by Leonie von Terzi and reviewed by Christine Unal and one anonymous referee.
Atlas, D., Srivastava, R. C., and Sekhon, R. S.: Doppler radar characteristics of precipitation at vertical incidence, Rev. Geophys., 11, 1–35, https://doi.org/10.1029/RG011i001p00001, 1973. a, b, c
Bringi, V. N. and Chandrasekar, V.: Polarimetric Doppler Weather Radar: Principles and Applications, Cambridge University Press, Cambridge, ISBN 978-0-521-62384-1, https://doi.org/10.1017/CBO9780511541094, 2001. a, b
Carey, L. D. and Petersen, W. A.: Sensitivity of C-Band Polarimetric Radar–Based Drop Size Estimates to Maximum Diameter, J. Appl. Meteorol. Climatol., 54, 1352–1371, https://doi.org/10.1175/JAMC-D-14-0079.1, 2015. a
Chandrasekar, V. and Bringi, V. N.: Simulation of Radar Reflectivity and Surface Measurements of Rainfall, J. Atmos. Ocean. Technol., 4, 464–478, https://doi.org/10.1175/1520-0426(1987)004<0464:SORRAS>2.0.CO;2, 1987. a, b, c
Crane, R.: Prediction of Attenuation by Rain, IEEE T. Commun., 28, 1717–1733, https://doi.org/10.1109/TCOM.1980.1094844, 1980. a, b
Deng, M., Giangrande, S. E., Jensen, M. P., Johnson, K., Williams, C. R., Comstock, J. M., Feng, Y.-C., Matthews, A., Lindenmaier, I. A., Wendler, T. G., Rocque, M., Zhou, A., Zhu, Z., Luke, E., and Wang, D.: Wet-radome attenuation in ARM cloud radars and its utilization in radar calibration using disdrometer measurements, Atmos. Meas. Tech., 18, 1641–1657, https://doi.org/10.5194/amt-18-1641-2025, 2025. a
Gatidis, C., Schleiss, M., Unal, C., and Russchenberg, H.: A Critical Evaluation of the Adequacy of the Gamma Model for Representing Raindrop Size Distributions, J. Atmos. Ocean. Technol., 37, 1765–1779, https://doi.org/10.1175/JTECH-D-19-0106.1, 2020. a
Giangrande, S. E., Wang, D., Bartholomew, M. J., Jensen, M. P., Mechem, D. B., Hardin, J. C., and Wood, R.: Midlatitude Oceanic Cloud and Precipitation Properties as Sampled by the ARM Eastern North Atlantic Observatory, J. Geophys. Res.-Atmos., 124, 4741–4760, https://doi.org/10.1029/2018JD029667, 2019. a
Gorgucci, E., Bechini, R., Baldini, L., Cremonini, R., and Chandrasekar, V.: The Influence of Antenna Radome on Weather Radar Calibration and Its Real-Time Assessment, J. Atmos. Ocean. Technol., https://doi.org/10.1175/JTECH-D-12-00071.1, 2013. a
Hari, P. and Kulmala, M.: Station for Measuring Ecosystem–Atmosphere Relations (SMEAR II), Boreal Environ. Res., 10, 315–322, 2005. a
Hitschfeld, W. and Bordan, J.: Errors inherent in the radar measurement of rainfall at attenuating wavelengths, J. Atmos. Sci., https://doi.org/10.1175/1520-0469(1954)011<0058:EIITRM>2.0.CO;2, 1954. a, b
Hogan, R. J. and O’Connor, E. J.: Facilitating cloud radar and lidar algorithms: the Cloudnet Instrument Synergy/Target Categorization product, University of Reading, https://www.met.reading.ac.uk/~swrhgnrj/publications/categorization.pdf (last access: 12 November 2026), 2004. a
Hogan, R. J., Bouniol, D., Ladd, D. N., O'Connor, E. J., and Illingworth, A. J.: Absolute Calibration of 94/95-GHz Radars Using Rain, J. Atmos. Ocean. Technol., https://doi.org/10.1175/1520-0426(2003)20<572:ACOGRU>2.0.CO;2, 2003. a
Hogan, R. J., Illingworth, A. J., O'Connor, E. J., Bouniol, D., Brook, M. E., Delanoë, J., Donovan, D. P., Eastment, J. D., Gaussiat, N., Goddard, J. W. F., Haeffelin, M., Baltink, H. K., Krasnov, O. A., Pelon, J., Piriou, J.-M., Protat, A., Russchenberg, H. W. J., Seifert, A., Tompkins, A. M., Zadelhoff, G.-J. v., Vinit, F., Westbrook, C., Willén, U., Wilson, D. R., and Wrench, C. L.: Cloudnet: Evaluation of model clouds using ground-based observations, Proceedings of the ECMWF Workshop on Parametrization of clouds in large-scale models, Reading, UK, B. Am. Meteorol. Soc., 88, 883–898, https://doi.org/10.1175/BAMS-88-6-883, 2006a. a
Hogan, R. J., Mittermaier, M. P., and Illingworth, A. J.: The Retrieval of Ice Water Content from Radar Reflectivity Factor and Temperature and Its Use in Evaluating a Mesoscale Model, J. Appl. Meteorol. Climatol., https://doi.org/10.1175/JAM2340.1, 2006b. a, b
Illingworth, A. J. and Blackman, T. M.: The Need to Represent Raindrop Size Spectra as Normalized Gamma Distributions for the Interpretation of Polarization Radar Observations, J. Appl. Meteorol. Climatol., https://doi.org/10.1175/1520-0450(2002)041<0286:TNTRRS>2.0.CO;2, 2002. a
Illingworth, A. J., Hogan, R. J., O'Connor, E. J., Bouniol, D., Brooks, M. E., Delanoé, J., Donovan, D. P., Eastment, J. D., Gaussiat, N., Goddard, J. W. F., Haeffelin, M., Baltink, H. K., Krasnov, O. A., Pelon, J., Piriou, J.-M., Protat, A., Russchenberg, H. W. J., Seifert, A., Tompkins, A. M., Zadelhoff, G.-J. v., Vinit, F., Willén, U., Wilson, D. R., and Wrench, C. L.: Cloudnet, B. Am. Meteorol. Soc., https://doi.org/10.1175/BAMS-88-6-883, 2007. a
Kollias, P., Puigdomènech Treserras, B., and Protat, A.: Calibration of the 2007–2017 record of Atmospheric Radiation Measurements cloud radar observations using CloudSat, Atmos. Meas. Tech., 12, 4949–4964, https://doi.org/10.5194/amt-12-4949-2019, 2019. a
Küchler, N., Kneifel, S., Löhnert, U., Kollias, P., Czekala, H., and Rose, T.: A W-Band Radar–Radiometer System for Accurate and Continuous Monitoring of Clouds and Precipitation, J. Atmos. Ocean. Technol., https://doi.org/10.1175/JTECH-D-17-0019.1, 2017. a, b
Kurri, M. and Huuskonen, A.: Measurements of the Transmission Loss of a Radome at Different Rain Intensities, J. Atmos. Ocean. Technol., https://doi.org/10.1175/2008JTECHA1056.1, 2008. a
Lamb, D. and Verlinde, J.: Physics and Chemistry of Clouds, Cambridge University Press, New York, ISBN 978-0-521-89910-9, 2011. a, b
Lamer, K., Oue, M., Battaglia, A., Roy, R. J., Cooper, K. B., Dhillon, R., and Kollias, P.: Multifrequency radar observations of clouds and precipitation including the G-band, Atmos. Meas. Tech., 14, 3615–3629, https://doi.org/10.5194/amt-14-3615-2021, 2021. a
L'Ecuyer, T. S. and Stephens, G. L.: An Estimation-Based Precipitation Retrieval Algorithm for Attenuating Radars, J. Appl. Meteorol. Climatol., https://doi.org/10.1175/1520-0450(2002)041<0272:AEBPRA>2.0.CO;2, 2002. a, b
Leinonen, J.: High-level interface to T-matrix scattering calculations: architecture, capabilities and limitations, Opt. Express, 22, 1655–1660, https://doi.org/10.1364/OE.22.001655, 2014. a, b
Leinonen, J., Moisseev, D., Leskinen, M., and Petersen, W. A.: A Climatology of Disdrometer Measurements of Rainfall in Finland over Five Years with Implications for Global Radar Observations, J. Appl. Meteorol. Climatol., https://doi.org/10.1175/JAMC-D-11-056.1, 2012. a, b, c
Lhermitte, R.: Attenuation and Scattering of Millimeter Wavelength Radiation by Clouds and Precipitation, J. Atmos. Ocean. Technol., https://doi.org/10.1175/1520-0426(1990)007<0464:AASOMW>2.0.CO;2, 1990. a
Li, H. and Moisseev, D.: Melting Layer Attenuation at Ka- and W-Bands as Derived From Multifrequency Radar Doppler Spectra Observations, J. Geophys. Res.-Atmos., 124, 9520–9533, https://doi.org/10.1029/2019JD030316, 2019. a, b
Li, H. and Moisseev, D.: Two Layers of Melting Ice Particles Within a Single Radar Bright Band: Interpretation and Implications, Geophys. Res. Lett., 47, e2020GL087499, https://doi.org/10.1029/2020GL087499, 2020. a
Löffler-Mang, M. and Joss, J.: An Optical Disdrometer for Measuring Size and Velocity of Hydrometeors, J. Atmos. Ocean. Technol., https://doi.org/10.1175/1520-0426(2000)017<0130:AODFMS>2.0.CO;2, 2000. a
Mardia, K. V., Kent, J. T., and Bibby, J. M.: Multivariate Analysis, Academic Press, London, 1979. a
Matrosov, S. Y.: Potential for attenuation-based estimations of rainfall rate from CloudSat, Geophys. Res. Lett., 34, https://doi.org/10.1029/2006GL029161, 2007. a, b
Matrosov, S. Y.: Assessment of Radar Signal Attenuation Caused by the Melting Hydrometeor Layer, IEEE T. Geosci. Remote Sens., 46, 1039–1047, https://doi.org/10.1109/TGRS.2008.915757, 2008. a, b, c
Mishchenko, M. I. and Travis, L. D.: T-matrix computations of light scattering by large spheroidal particles, Opt. Commun., 109, 16–21, https://doi.org/10.1016/0030-4018(94)90731-5, 1994. a
Mishchenko, M. I., Travis, L. D., and Macke, A.: Scattering of light by polydisperse, randomly oriented, finite circular cylinders, Appl. Opt., 35, 4927–4940, https://doi.org/10.1364/AO.35.004927, 1996. a
Moisseev, D. and Petäjä, T.: Custom collection of disdrometer, and weather station data from Hyytiälä between 1 Jan 2014 and 31 Dec 2024, ACTRIS Cloud remote sensing data centre unit (CLU), Cloudnet [data set], https://doi.org/10.60656/8d288976932d417e, 2025. a, b
Moisseev, D., O'Connor, E., and Petäjä, T.: Custom collection of categorize, classification, disdrometer, ice water content, and weather station data from Hyytiälä on 26 May 2022, Cloudnet [data set], https://doi.org/10.60656/1fe3a02ef09b419d, 2025a. a, b
Moisseev, D., O'Connor, E., and Petäjä, T.: Custom collection of categorize, classification, disdrometer, ice water content, and weather station data from Hyytiälä between 28 and 29 Jul 2023, Cloudnet [data set], https://doi.org/10.60656/d25b530af3d04367, 2025b. a, b
Moisseev, D. N. and Chandrasekar, V.: Examination of the μ–Λ Relation Suggested for Drop Size Distribution Parameters, J. Atmos. Ocean. Technol., https://doi.org/10.1175/JTECH2010.1, 2007. a, b, c
Mom, Bernd : berndmom/ERAC: Project release (Version v1.0.0), Zenodo [code], https://doi.org/10.5281/zenodo.24166601, 2026. a
Myagkov, A., Kneifel, S., and Rose, T.: Evaluation of the reflectivity calibration of W-band radars based on observations in rain, Atmos. Meas. Tech., 13, 5799–5825, https://doi.org/10.5194/amt-13-5799-2020, 2020. a, b, c, d, e
OTT HydroMet GmbH: Operating instructions Precipitation gauge OTT Pluvio2 L, https://cdn.hach.com/1XMCM0ZF/at/hqb367sgkf47xb77nhs85wrs/Operating_instructions_Pluvio2_L_V3.pdf (last access: 12 November 2026), 2014. a
OTT HydroMet GmbH: Operating instructions Present Weather Sensor OTT Parsivel2, https://www.ott.com/products/meteorological-sensors-26/ott-parsivel2-laser-weather-sensor-2392/ (last access: 12 November 2026), 2017. a, b
Sekelsky, S. M.: Near-Field Reflectivity and Antenna Boresight Gain Corrections for Millimeter-Wave Atmospheric Radars, J. Atmos. Ocean. Technol., https://doi.org/10.1175/1520-0426(2002)019<0468:NFRAAB>2.0.CO;2, 2002. a, b, c, d
Sekelsky, S. M. and Clothiaux, E. E.: Parallax Errors and Corrections for Dual-Antenna Millimeter-Wave Cloud Radars, J. Atmos. Ocean. Technol., https://doi.org/10.1175/1520-0426(2002)019<0478:PEACFD>2.0.CO;2, 2002. a, b, c
Testud, J., Oury, S., Black, R. A., Amayenc, P., and Dou, X.: The Concept of “Normalized” Distribution to Describe Raindrop Spectra: A Tool for Cloud Physics and Cloud Remote Sensing, J. Appl. Meteorol. Climatol., https://doi.org/10.1175/1520-0450(2001)040<1118:TCONDT>2.0.CO;2, 2001. a
Tokay, A., Wolff, D. B., and Petersen, W. A.: Evaluation of the New Version of the Laser-Optical Disdrometer, OTT Parsivel2, J. Atmos. Ocean. Technol., https://doi.org/10.1175/JTECH-D-13-00174.1, 2014. a
Tridon, F., Battaglia, A., Luke, E., and Kollias, P.: Rain retrieval from dual-frequency radar Doppler spectra: validation and potential for a midlatitude precipitating case-study, Q. J. Roy. Meteorol. Soc., 143, 1364–1380, https://doi.org/10.1002/qj.3010 2017. a, b
Tukiainen, S., O'Connor, E., and Korpinen, A.: CloudnetPy: A Python package for processing cloud remote sensing data, J. Open Source Softw., 5, 2123, https://doi.org/10.21105/joss.02123, 2020. a
Tukiainen, S., Siipola, T., Anniina, K., and O'Connor, E.: CloudnetPy: A Python package for processing cloud remote sensing data, Zenodo, https://doi.org/10.5281/zenodo.13833101, 2024. a, b
Ulbrich, C. W. and Atlas, D.: Rainfall Microphysics and Radar Properties: Analysis Methods for Drop Size Spectra, J. Appl. Meteorol. Climatol., https://doi.org/10.1175/1520-0450(1998)037<0912:RMARPA>2.0.CO;2, 1998. a
Wielaard, D. J., Mishchenko, M. I., Macke, A., and Carlson, B. E.: Improved T-matrix computations for large, nonabsorbing and weakly absorbing nonspherical particles and comparison with geometrical-optics approximation, Appl. Opt., 36, 4305–4313, https://doi.org/10.1364/AO.36.004305, 1997. a
Williams, C. R.: How Much Attenuation Extinguishes mm-Wave Vertically Pointing Radar Return Signals?, Remote Sens., 14, 1305, https://doi.org/10.3390/rs14061305, 2022. a
Williams, C. R., Gage, K. S., Clark, W., and Kucera, P.: Monitoring the Reflectivity Calibration of a Scanning Radar Using a Profiling Radar and a Disdrometer, J. Atmos. Ocean. Technol., https://doi.org/10.1175/JTECH1759.1, 2005. a
Wong, R. K. W. and Chidambaram, N.: Gamma Size Distribution and Stochastic Sampling Errors, J. Appl. Meteorol. Climatol., 24, 568–579, https://doi.org/10.1175/1520-0450(1985)024<0568:GSDASS>2.0.CO;2, 1985. a