• Adirosi, E., L. Baldini, F. Lombardo, F. Russo, F. Napolitano, E. Volpi, and A. Tokay, 2015: Comparison of different fittings of drop spectra for rainfall retrievals. Adv. Water Resour., 83, 5567, https://doi.org/10.1016/j.advwatres.2015.05.009.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Aires, F., A. Chédin, N. Scott, and W. B. Rossow, 2002: A regularized neural network approach for retrieval of atmospheric and surface temperatures with the IASI instrument. J. Appl. Meteor., 41, 144159, https://doi.org/10.1175/1520-0450(2002)041<0144:ARNNAF>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Andsager, K., K. V. Beard, and N. F. Laird, 1999: Laboratory measurements of axis ratios for large rain drops. J. Atmos. Sci., 56, 26732683, https://doi.org/10.1175/1520-0469(1999)056<2673:LMOARF>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Battaglia, A., S. Tanelli, G. M. Heymsfield, and L. Tian, 2014: The dual wavelength ratio knee: A signature of multiple scattering in airborne Ku–Ka observations. J. Appl. Meteor. Climatol., 53, 17901808, https://doi.org/10.1175/JAMC-D-13-0341.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Beard, K. V., and C. Chuang, 1987: A new model for the equilibrium shape of raindrops. J. Atmos. Sci., 44, 15091524, https://doi.org/10.1175/1520-0469(1987)044<1509:ANMFTE>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bech, J., B. Codina, J. Lorente, and D. Bebbington, 2003: The sensitivity of single polarization weather radar beam blockage correction to variability in the vertical refractivity gradient. J. Atmos. Oceanic Technol., 20, 845855, https://doi.org/10.1175/1520-0426(2003)020<0845:TSOSPW>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Brandes, E., G. Zhang, and J. Vivekanandan, 2002: Experiments in rainfall estimation with a polarimetric radar in a subtropical environment. J. Appl. Meteor., 41, 674685, https://doi.org/10.1175/1520-0450(2002)041<0674:EIREWA>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bringi, V. N., V. Chandrasekar, N. Balakrishnan, and D. S. Zrnić, 1990: An examination of propagation effects in rainfall on radar measurements at microwave frequencies. J. Atmos. Oceanic Technol., 7, 829840, https://doi.org/10.1175/1520-0426(1990)007<0829:AEOPEI>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cao, Q., G. Zhang, E. Brandes, and T. Schuur, 2010: Polarimetric radar rain estimation through retrieval of drop size distribution using a Bayesian approach. J. Appl. Meteor. Climatol., 49, 973990, https://doi.org/10.1175/2009JAMC2227.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cao, Q., G. Zhang, and M. Xue, 2013: A variational approach for retrieving raindrop size distribution from polarimetric radar measurements in the presence of attenuation. J. Appl. Meteor. Climatol., 52, 169185, https://doi.org/10.1175/JAMC-D-12-0101.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • D’Adderio, L. P., F. Porcu, and A. Tokay, 2015: Raindrop size distribution in the presence of break-up. J. Atmos. Sci., 72, 34043416, https://doi.org/10.1175/JAS-D-14-0304.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Drobinski, P., and et al. , 2014: HyMeX: A 10-year multidisciplinary program on the Mediterranean water cycle. Bull. Amer. Meteor. Soc., 95, 10631082, https://doi.org/10.1175/BAMS-D-12-00242.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Grecu, M., W. S. Olson, S. J. Munchak, S. Ringerud, L. Liao, Z. Haddad, B. L. Kelley, and S. F. McLaughlin, 2016: The GPM combined algorithm. J. Atmos. Oceanic Technol., 33, 22252245, https://doi.org/10.1175/JTECH-D-16-0019.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hou, A. Y., and et al. , 2014: The Global Precipitation Measurement Mission. Bull. Amer. Meteor. Soc., 95, 701722, https://doi.org/10.1175/BAMS-D-13-00164.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Houze, R., Jr., and et al. , 2017: The Olympic Mountains Experiment (OLYMPEX). Bull. Amer. Meteor. Soc., 98, 21672188, https://doi.org/10.1175/BAMS-D-16-0182.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Iguchi, T., and et al. , 2016: Precipitation rates estimated with GPM’s dual-frequency radar. 2016 IEEE Int. Geoscience and Remote Sensing Symp., Beijing, China, IEEE, 3918, https://doi.org/10.1109/IGARSS.2016.7730017.

    • Crossref
    • Export Citation
  • Iguchi, T., S. Seto, R. Meneghini, N. Yoshida, J. Awaka, M. Le, V. Chandrasekar, and T. Kubota, 2017: GPM/DPR level-2. Algorithm Theoretical Basis Doc., 81 pp., http://www.eorc.jaxa.jp/GPM/do AU6 c/algorithm/ATBD_DPR_201708_whole_1.pdf.

  • Kruger, A., and W. F. Krajewski, 2002: Two-dimensional video disdrometer: A description. J. Atmos. Oceanic Technol., 19, 602617, https://doi.org/10.1175/1520-0426(2002)019<0602:TDVDAD>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kubota, T., T. Iguchi, M. Kojima, L. Liao, T. Masaki, H. Hanado, R. M. Meneghini, and R. Oki, 2016: A statistical method for reducing sidelobe clutter for the Ku-band precipitation radar onboard the GPM Core Observatory. J. Atmos. Oceanic Technol., 33, 14131428, https://doi.org/10.1175/JTECH-D-15-0202.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lee, G. W., and I. Zawadzki, 2005: Variability of drop size distributions: Time-scale dependence of the variability and its effects on rain estimation. J. Appl. Meteor., 44, 634652, https://doi.org/10.1175/JAM2222.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Meneghini, R., H. Kim, L. Liao, J. A. Jones, and J. M. Kwiatkowski, 2015: An initial assessment of the surface reference technique applied to data from the Dual-Frequency Precipitation Radar (DPR) on the GPM satellite. J. Atmos. Oceanic Technol., 32, 22812296, https://doi.org/10.1175/JTECH-D-15-0044.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mishchenko, M. I., 2000: Calculation of the amplitude matrix for a nonspherical particle in a fixed orientation. Appl. Opt., 39, 10261031, https://doi.org/10.1364/AO.39.001026.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mroz, K., A. Battaglia, T. J. Lang, D. J. Cecil, S. Tanelli, and F. Tridon, 2017: Hail-detection algorithm for the GPM Core Observatory satellite sensors. J. Appl. Meteor. Climatol., 56, 19391957, https://doi.org/10.1175/JAMC-D-16-0368.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ni, X., C. Liu, D. J. Cecil, and Q. Zhang, 2017: On the detection of hail using satellite passive microwave radiometers and precipitation radar. J. Appl. Meteor. Climatol., 56, 26932709, https://doi.org/10.1175/JAMC-D-17-0065.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Nurmi, P., 2003: Recommendations on the verification of local weather forecasts. ECMWF Tech. Memo. 430, 19 pp., https://www.ecmwf.int/en/elibrary/11401-recommendations-verification-local-weather-forecasts.

  • Panegrossi, G., and et al. , 2016: Use of the GPM constellation for monitoring heavy precipitation events over the Mediterranean region. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 9, 27332753, https://doi.org/10.1109/JSTARS.2016.2520660.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Petracca, M., L. P. D’Adderio, F. Porcù, G. Vulpiani, S. Sebastianelli, and S. Puca, 2018: Validation of GPM Dual-Frequency Precipitation Radar (DPR) rainfall products over Italy. J. Hydrometeor., 19, 907925, https://doi.org/10.1175/JHM-D-17-0144.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Seto, S., T. Iguchi, and T. Oki, 2013: The basic performance of a precipitation retrieval algorithm for the Global Precipitation Measurement mission’s single/dual frequency radar measurements. IEEE Trans. Geosci. Remote Sens., 51, 52395251, https://doi.org/10.1109/TGRS.2012.2231686.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Seto, S., T. Shimozumal, T. Iguchi, and T. Oki, 2016: Spatial and temporal variations of mass-weighted mean diameter estimated by GPM/DPR. IEEE Int. Geoscience and Remote Sensing Symp., Beijing, China, IEEE, 3938–3940, https://doi.org/10.1109/IGARSS.2016.7730023.

    • Crossref
    • Export Citation
  • Speirs, P., M. Gabella, and A. Berne, 2017: A comparison between the GPM Dual-Frequency Precipitation Radar and ground-based radar precipitation rate estimates in the Swiss Alps and Plateau. J. Hydrometeor., 18, 12471269, https://doi.org/10.1175/JHM-D-16-0085.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tabary, P., 2007: The new French operational radar rainfall product. Part I: Methodology. Wea. Forecasting, 22, 393408, https://doi.org/10.1175/WAF1004.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Thurai, M., and V. N. Bringi, 2005: Drop axis ratios from 2D video disdrometer. J. Atmos. Oceanic Technol., 22, 966978, https://doi.org/10.1175/JTECH1767.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Thurai, M., G. Huang, V. Bringi, W. Randeu, and M. Schönhuber, 2007: Drop shapes, model comparisons, and calculations of polarimetric radar parameters in rain. J. Atmos. Oceanic Technol., 24, 10191032, https://doi.org/10.1175/JTECH2051.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tokay, A., L. P. D’Adderio, D. B. Wolff, and W. A. Petersen, 2016: A field study of pixel scale variability of raindrop size distribution in the Mid-Atlantic region. J. Hydrometeor., 17, 18551868, https://doi.org/10.1175/JHM-D-15-0159.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tokay, A., L. P. D’Adderio, F. Porcù, D. B. Wolff, and W. A. Petersen, 2017: A Field study of footprint-scale variability of raindrop size distribution. J. Hydrometeor., 18, 31653179, https://doi.org/10.1175/JHM-D-17-0003.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vulpiani, G., F. S. Marzano, V. Chandrasekar, A. Berne, and R. Uijlenhoet, 2006: Polarimetric weather radar retrieval of raindrop size distribution by means of a regularized artificial neural network. IEEE Trans. Geosci. Remote Sens., 44, 32623275, https://doi.org/10.1109/TGRS.2006.878438.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vulpiani, G., S. Giangrande, and F. S. Marzano, 2009: Rainfall estimation from polarimetric S-band radar measurements: Validation of a neural network approach. J. Appl. Meteor. Climatol., 48, 20222036, https://doi.org/10.1175/2009JAMC2172.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vulpiani, G., M. Montopoli, L. Delli Passeri, A. Gioia, P. Giordano, and F. S. Marzano, 2012: On the use of dual-polarized C-band radar for operational rainfall retrieval in mountainous areas. J. Appl. Meteor. Climatol., 51, 405425, https://doi.org/10.1175/JAMC-D-10-05024.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vulpiani, G., L. Baldini, and N. Roberto, 2015: Characterization of Mediterranean hail-bearing storms using an operational polarimetric X-band radar. Atmos. Meas. Tech., 8, 46814698, https://doi.org/10.5194/amt-8-4681-2015.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • View in gallery

    Example of three LBM reflectivities as measured by GR overlapped by corresponding GPM overpasses for (a) Il Monte, (b) Serano, and (c) Lauro sites. The gray area represents the center of Ku-band footprints, while the black area represents the center of Ka-band footprints. The red plus sign indicates the location of the GR.

  • View in gallery

    The DmZDR relationships derived for each GV dataset and combining all together (ALL).

  • View in gallery

    Comparison between Dm measured by 2DVD and Dm estimated by applying the (a) NN approach and (b) DmZDR approach to the 2DVD data.

  • View in gallery

    Comparison between the Dm estimated by NN and DmZDR when the two approaches are applied to (a) 2DVD and (b) GR data.

  • View in gallery

    Relationship by simulated reflectivity at (a) C and Ku band and at (b) C and Ka band.

  • View in gallery

    Comparison between reflectivity measured by GR at C band (corrected by ground clutter, noise, etc.) and reflectivity measured by GPM radars scaled at C band for stratiform precipitation over land after being corrected by attenuation using (a) Ku-only information, (b) Ka-only information, and dual-frequency information for (c) Ka and (d) Ku band.

  • View in gallery

    Comparison between Dm estimated over land by GR and by (a) DPR NS stratiform, (b) DPR MS stratiform, (c) DPR HS stratiform, (d) DPR NS convective, (e) DPR MS convective, and (f) DPR HS convective.

  • View in gallery

    Comparison between Dm estimated over land for (top) stratiform and (bottom) convective precipitation by GR and by the (a),(c) DPR–GMI NS and (b),(d) DPR–GMI MS algorithms.

  • View in gallery

    Distribution of PE for the (a) DPR algorithm and (b) DPR–GMI algorithm. The blue and red lines refer to stratiform and convective classification, respectively. The different line styles refer to the DPR scan type.

  • View in gallery

    Mean ± standard deviation of PE (expressed as difference between GPM and GR Dm estimations) with respect to the distance of the matching point from GR for (a) stratiform and (b) convective precipitation. The line colors are blue for DPR NS, red for DPR MS, black for DPR HS, green for DPR–GMI NS, and magenta for DPR–GMI MS.

  • View in gallery

    Dependence of PE (expressed as difference between GPM and GR Dm estimations) with respect to the number of no-rain GR pixels and their standard deviation within a DPR footprint for DPR (a) NS, (b) MS, and (c) HS algorithm and for DPR–GMI (d) NS and (e) MS.

  • View in gallery

    Reflectivity measured by (a) DPR ZKu and (b) DPR ZKa and GR (ZC) for the cases where the DPR algorithm reported Dm as missed value.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 140 140 18
PDF Downloads 115 115 17

Comparison of GPM Core Observatory and Ground-Based Radar Retrieval of Mass-Weighted Mean Raindrop Diameter at Midlatitude

View More View Less
  • 1 Department of Physics and Earth Science, University of Ferrara, Ferrara, and Institute of Atmospheric Sciences and Climate, National Research Council, Rome, Italy
  • | 2 Department of Civil Protection, Presidency of the Council of Ministers, Rome, Italy
  • | 3 Department of Physics and Astronomy, University of Bologna, Bologna, Italy
  • | 4 Joint Center for Earth Systems Technology, University of Maryland, Baltimore County, Baltimore, and NASA Goddard Space Flight Center, Greenbelt, Maryland
  • | 5 NASA Goddard Space Flight Center, Greenbelt, Maryland
© Get Permissions
Full access

Abstract

One of the main goals of the National Aeronautics and Space Administration (NASA) Global Precipitation Measurement (GPM) mission is to retrieve parameters of the raindrop size distribution (DSD) globally. As a standard product of the Dual-Frequency Precipitation Radar (DPR) on board the GPM Core Observatory satellite, the mass-weighted mean diameter Dm and the normalized intercept parameter Nw are estimated in three dimensions at the resolution of the radar. These are two parameters of the three-parameter gamma model DSD adopted by the GPM algorithms. This study investigates the accuracy of the Dm retrieval through a comparative study of C-band ground radars (GRs) and GPM products over Italy. The reliability of the ground reference is tested by using two different approaches to estimate Dm. The results show good agreement between the ground-based and spaceborne-derived Dm, with an absolute bias being generally lower than 0.5 mm over land in stratiform precipitation for the DPR algorithm and the combined DPR–GMI algorithm. For the DPR–GMI algorithm, the good agreement extends to convective precipitation as well. Estimates of Dm from the DPR high-sensitivity (HS) Ka-band data show slightly worse results. A sensitivity study indicates that the accuracy of the Dm estimation is independent of the height above surface (not shown) and the distance from the ground radar. On the other hand, a nonuniform precipitation pattern (interpreted both as high variability and as a patchy spatial distribution) within the DPR footprint is usually associated with a significant error in the DPR-derived estimate of Dm.

© 2018 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Leo Pio D’Adderio, leopio.dadderio@artov.isac.cnr.it

Abstract

One of the main goals of the National Aeronautics and Space Administration (NASA) Global Precipitation Measurement (GPM) mission is to retrieve parameters of the raindrop size distribution (DSD) globally. As a standard product of the Dual-Frequency Precipitation Radar (DPR) on board the GPM Core Observatory satellite, the mass-weighted mean diameter Dm and the normalized intercept parameter Nw are estimated in three dimensions at the resolution of the radar. These are two parameters of the three-parameter gamma model DSD adopted by the GPM algorithms. This study investigates the accuracy of the Dm retrieval through a comparative study of C-band ground radars (GRs) and GPM products over Italy. The reliability of the ground reference is tested by using two different approaches to estimate Dm. The results show good agreement between the ground-based and spaceborne-derived Dm, with an absolute bias being generally lower than 0.5 mm over land in stratiform precipitation for the DPR algorithm and the combined DPR–GMI algorithm. For the DPR–GMI algorithm, the good agreement extends to convective precipitation as well. Estimates of Dm from the DPR high-sensitivity (HS) Ka-band data show slightly worse results. A sensitivity study indicates that the accuracy of the Dm estimation is independent of the height above surface (not shown) and the distance from the ground radar. On the other hand, a nonuniform precipitation pattern (interpreted both as high variability and as a patchy spatial distribution) within the DPR footprint is usually associated with a significant error in the DPR-derived estimate of Dm.

© 2018 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Leo Pio D’Adderio, leopio.dadderio@artov.isac.cnr.it

1. Introduction

One of the goals of the National Aeronautics and Space Administration (NASA) Global Precipitation Measurement (GPM) mission is to demonstrate that a radar–radiometer space-based measuring system can provide frequent and accurate precipitation measurements globally. As such, this set of sensors can serve as the flagship space mission for hydrology-related regional and urban climate research and applications programs. The GPM Core Observatory, the leading satellite in global precipitation mapping, is equipped with a Dual-Frequency Precipitation Radar (DPR) and a high-resolution multichannel passive microwave radiometer called the GPM Microwave Imager (GMI; Hou et al. 2014).

The main role of the DPR and the DPR–GMI is to retrieve the instantaneous three-dimensional precipitation rate and drop size distribution (DSD), through the estimation of the DSD parameters, by using Ka- (35 GHz) and Ku-band (13.6 GHz) radar measurements along with multifrequency brightness temperature measurements from the GMI.

The mass-weighted mean diameter Dm is one of the two DSD parameters given in the DPR and the DPR–GMI standard products, together with the normalized intercept parameter Nw. The GPM adopted a three-parameter Gamma distribution consisting of both Dm and Nw as well as a shape parameter that is set to 3 in the DPR (Seto et al. 2013) and 2 in the DPR–GMI combined algorithm (Grecu et al. 2016). Quantification of the uncertainties in the DSD and precipitation retrievals is the main responsibility of the GPM Ground Validation (GV) program and can be assessed through comparative studies with the ground-based estimates. Both spaceborne and ground-based estimates of DSD parameters are subject to different sources of error, and it is important to understand and quantify them. The goal of characterizing the errors in the DPR products from extensive and reliable ground-based references takes on greater importance as the DPR estimates are used as the primary validation tool for the GMI global precipitation products. The relatively recent availability of GPM data (the satellite was launched on 27 February 2014) together with the need for sufficient high-quality, dual-polarized, ground-based radar data over different climatologies have limited the number of studies focused on GPM validation of DSD parameters. Among the few studies that have been made, Seto et al. (2016) analyzed, at global scale, the variation of Dm estimated by version 04 (V04) of the DPR algorithm at given rainfall rate bands. They found that Dm is generally larger over land than over ocean. While over land there is not an appreciable difference in Dm between mid- and tropical latitudes, over ocean Dm is larger in midlatitudes than in intertropical zone.

The importance of ground-based validation of GPM products is evidenced by the fact that several authors have examined Core Observatory rainfall rate products in different climatological and terrain conditions (Petracca et al. 2018; Speirs et al. 2017; Iguchi et al. 2016) as well as in different precipitation systems (Panegrossi et al. 2016; Mroz et al. 2017; Ni et al. 2017). Furthermore, the latest version (V05) of the DPR algorithm assumes constraint relationships between rainfall rate R and Dm, which are different for stratiform and convective precipitation. This again highlights the importance of reliable estimates of Dm and the need to validate the RDm relationships used in the DPR algorithm by the use of ground-based radars and disdrometers.

DPR measurements are affected by several sources of error, including the effects of attenuation (especially at Ka band; Meneghini et al. 2015); ground clutter (Kubota et al. 2016), which interferes with near-surface rain observations; nonuniform beam filling (NUBF), which is related to the variability of precipitation within the DPR footprint (Tokay et al. 2017); and multiple scattering (Battaglia et al. 2014). Although the DPR and DPR–GMI algorithms attempt to correct for most of these effects, the effectiveness of the correction and the accuracy of the retrievals in general must be evaluated by direct comparisons to ground-based data.

The present work fits into this general framework by using ground-based polarimetric radar data (GR) managed by the Italian Department of Civil Protection (DPC). These data provide comparisons of Dm estimated by both DPR and combined DPR–GMI algorithms with GR estimates, which are taken as reference over Italy (Vulpiani et al. 2006, 2009). The paper aims, on one hand, to investigate the reliability of both DPR and DPR–GMI algorithms in estimating Dm and, on the other, to identify conditions that affect the goodness of the estimation. As highlighted previously, the importance of a reliable Dm estimate is increased by the fact that DPR algorithm uses an RDm relationship as a constraint. The region and the seasons (summer and early fall) analyzed are challenging for the precipitation measurement because of the complexity of the terrain and the development of isolated, mesoscale organized and/or embedded systems in stratiform cloud structures.

The paper is organized as follows. Section 2 briefly describes the spaceborne and ground-based instrumentation. Section 3 summarizes the methodologies used to estimate Dm. In sections 4 and 5, respectively, the reflectivities and Dm values estimated by the DPR and GR are compared. Section 6 provides a sensitivity analysis on the possible source of error, while the conclusions are given in section 7.

2. Data and instrumentation

Fourteen GPM overpasses over Italy, collected mainly in the summer and early fall seasons (from June to the beginning of October) in the 2015–17 time interval, have been analyzed in this work. The 2A-DPR, which includes both single- and dual-frequency algorithms, and the 2B-CMB (combined DPR–GMI algorithm) level products have been coupled and compared with dual-polarized data collected from the DPC GR. Figure 1 shows, as an example, the lowest beam map (LBM) reflectivity as measured by the GR overlapped by the GPM overpass. The gray dots represent the centers of the Ku-band footprints, while the black dots represent the centers of the Ka-band footprints. The red plus sign marks the location of the GR. Because of the different sizes of the DPR and GR resolution volumes, the GR-based Dm estimates within a DPR footprint have been averaged before performing the comparisons. At the same time, the 1° beamwidth of the GR antennas generates an horizontal and vertical extent of the radar volume which increases with the distance from GR (e.g., at 100 km from GR it is about 1.75 km). Because of the high spatial vertical resolution of DPR (125 or 250 m depending on the scan type), all GPM-based Dm estimated within the GR pixel have been averaged.

Fig. 1.
Fig. 1.

Example of three LBM reflectivities as measured by GR overlapped by corresponding GPM overpasses for (a) Il Monte, (b) Serano, and (c) Lauro sites. The gray area represents the center of Ku-band footprints, while the black area represents the center of Ka-band footprints. The red plus sign indicates the location of the GR.

Citation: Journal of Hydrometeorology 19, 10; 10.1175/JHM-D-18-0002.1

a. DPR properties and products

The DPR is the first dual-frequency spaceborne radar with operation at Ka (35 GHz) and Ku band (13.6 GHz). The DPR employs three scanning modes: matched scan (MS), normal scan (NS), and high-sensitivity scan (HS). The Ka- and Ku-band radar matched (MS) footprints of the inner swath consist of 25 angle bins with range sampling at 125 m while the Ku-band radar (NS) covers the full swath with 49 angle bins with a range sampling of 125 m. The HS Ka-band radar footprints are interlaced with matched Ku-/Ka-band footprints and consist of 24 angle bins with range sampling at 250 m. The swath widths of Ka- and Ku-band radars are 120 and 245 km, respectively, while both Ka- and Ku-band footprints at nadir are approximately 5 km in diameter.

For this work, the V05 level 2 products from both DPR and DPR–GMI algorithm have been used. The 2A-DPR products provide three scan types: 2A-DPR NS, 2A-DPR HS, and 2A-DPR MS. The 2A-DPR HS (hereafter DPR HS) output parameters are derived from Ka-only HS measurements (even though some ancillary dual-frequency information may be used). The inner swath (footprints 13–37 as labeled in the full-swath numbering system from 1 to 49) of 2A-DPR NS as well as the 2A-DPR MS (hereafter DPR MS) are processed by using the combined information from both the frequencies, while the output parameters of the outer swath (footprints 1–12 and 38–49) of 2A-DPR NS are derived from Ku-only measurements (Iguchi et al. 2017). On the other hand, the 2A-Ku NS products are derived only from Ku measurements (the products of the outer swath of 2A-DPR NS and 2A-Ku NS are identical) over the full swath. Consequently, in the following analysis the so-called DPR NS refers to the 2A-Ku NS products in order to test the performance of the whole swath using only the Ku-band data. The Dm is a standard output of all three DPR scan types.

On the other hand, the 2B-CMB level products combine the information from the active sensor DPR and passive sensor GMI and provide two scan types: 2B-CMB NS and 2B-CMB MS. As for the DPR algorithm, the DPR–GMI algorithm provides Dm as standard output, by using the single- and dual-frequency information for 2B-CMB NS and 2B-CMB MS, respectively.

The DPR and DPR–GMI outputs have been used to categorize and select the samples. The precipitation has been divided into stratiform, convective, and other (for those cases which are classified as neither stratiform nor convective) according to the DPR classification. The classification is based on analysis of the vertical profile of the measured dual-frequency ratio (DFRm) when the dual-frequency data are available; otherwise, it is based on the presence of the bright band (BB) together with a horizontal (H) or vertical (V) method (Iguchi et al. 2017). Furthermore, the GR pixels and DPR footprints used in the analysis are only those reporting liquid precipitation (as identified by DPR) below the BB, when it is clearly detected, or the freezing level both reported as outputs by DPR/DPR–GMI algorithm. In addition, the samples have been divided according to the surface type (land and sea) as classified by the DPR.

b. DPC ground radars properties

The DPC currently manages seven C-band and two X-band radar systems, all with dual-polarization capabilities. For this work, two C-band radars located in the center of Italy at Il Monte (41.9401°N, 14.6208°E) and Serano (42.8666°N, 12.8002°E), and one C-band radar located in the south Italy at Lauro (37.1130°N, 14.8355°E), at the altitudes of 700, 1500, and 980 m MSL, respectively, have been used. Of the 14 cases considered, only one case has been recorded by the Lauro GR, while two cases have been simultaneously observed by the Il Monte and Serano radars.

The scan strategy for the operational DPC GR involves eleven plane position indicator (PPI) inverse scans at inclination angles ranging between 16° and 90°, depending on the starting time of the scan, and 0.5° with uneven angle bin steps. The data have a time resolution of 5 min, while the 1° beamwidth and the 150-m range size ensure a quite good spatial resolution.

The data processing begins with the identification and compensation for nonweather returns based on the use of the fuzzy logic approach proposed in Vulpiani et al. (2012). The partial beam blockage is evaluated as proposed by Bech et al. (2003) and compensated up to 70% (Tabary 2007). The differential phase ΦDP is processed through the iterative finite difference (IFD) scheme proposed in Vulpiani et al. (2012, 2015) using a 3-km size moving window. The resulting ΦDP, which is immune to any potential system offset and unfolding, is used to correct for rain path attenuation (Bringi et al. 1990), whereas specific differential phase KDP is used jointly with reflectivity factor Z and differential reflectivity ZDR to estimate the DSD parameters, as described in section 3a.

Given the topography of the region (both the Il Monte and Serano radars are surrounded by the Apennines), the LBM of the GR has been spatiotemporally matched with the DPR measurements. The spatiotemporal matching was considered valid only when the DPR identified the phase of precipitation as liquid.

c. Disdrometer data

The data used in this study include the latest version of the two-dimensional video (2DVD; Kruger and Krajewski 2002) disdrometer, which has been deployed during GV field campaigns and at radar sites both before and after the GPM launch. This instrument is considered among the new standards for disdrometer measurements. The 2DVD measures the size, fall velocity, and shape of each hydrometeor falling within its measurement volume so that the DSD is available with a temporal resolution of 1 min.

In particular, the 2DVD observations of rain events from six different midlatitude GPM-GV filed campaigns that are used in this study include Iowa Flood Studies (IFloodS; 41.6°N, 91.5°W); Midlatitude Continental Convective Clouds Experiment (MC3E; 36.7°N, 97.1°W); Wallops Island, Virginia (Wallops; 37.9°N, 75.5°W); Huntsville, Alabama (Alabama; 34.7°N, 86.6°W); Integrated Precipitation and Hydrology Experiment (IPHEx; 35.5°N, 82.5°W; D’Adderio et al. 2015); and the Olympic Mountains Experiment (OLYMPEx; 47.5°N, 123.5°W; Houze et al. 2017). The data collected in Italy during the Hydrological Cycle in the Mediterranean Experiment (HyMeX; 41.9°N, 12.5°E) have been used as well (Drobinski et al. 2014).

The disdrometer data have been used for multiple purposes ranging from the derivation of a relationship to estimate Dm to the simulation of the radar variables. The GR retrieval approach adopted here is based on a neural network (NN) inversion technique to derive Dm. To test the reliability of NN estimations, a different approach based on a polynomial relationship between Dm and ZDR has been considered. The DmZDR relationship has been derived by fitting the 2DVD-based measurements (i.e., Dm) and simulations (i.e., ZDR). Both the NN and DmZDR approach have been applied to the 2DVD data and the estimated sets of Dm have been compared. Other than Dm, the reflectivity measured by GR and DPR has been compared in this study. Since the DPR operates at Ka- and Ku-band frequencies, the reflectivities have been adjusted for comparison to C band. This conversion is discussed in section 4.

3. Mass-weighted mean diameter GR estimation

The mass-weighted mean diameter Dm is estimated from GR by using a retrieval approach based on an NN inversion technique (Vulpiani et al. 2006). The GR Dm estimates can be also derived through an empirical relationship between Dm and the differential reflectivity ZDR. Although the official DPC GR products adopt the NN approach, in this study a preliminary sensitivity analysis of the reliability of the Dm estimation has been carried out by applying the DmZDR relationship to the selected GR data in order to compare the Dm estimated with the two approaches. Additional different approaches are present in literature and could be considered to estimate Dm (and DSD parameters in general) such as Bayesian or variational approach (Cao et al. 2013, 2010).

a. The neural network approach

An artificial NN can be considered as a nonlinear parameterized mapping from an input x to an output y = NN(x; w; M), where w is the vector of parameters (weights and biases) relating the input x to the output y, while the functional form of the mapping (i.e., the architecture of the net) is denoted as M.

The nonlinear neuron model, representing the basic processing unit, is composed by 1) a set of synapses, each of which is characterized by a weight; 2) an adder for summing the input signals, weighted by the respective synapses of the neuron; and 3) an activation function φ for limiting the amplitude of the output of a neuron. In mathematical terms, a given neuron output yk can be written as
e1
with
e2
where xj is the input to the jth synapse, wk,j is the synaptic weight, bk = wk,0x0 (with x0 = 1) is the bias, and φ is the activation function.

The NN considered here is a multilayer perceptron architecture (MLP) with back-propagation learning algorithm. It was originally proposed by Vulpiani et al. (2006) for S band, where the retrieval uncertainties were addressed with respect to a simulated and real dataset, and adapted here for C-band radar measurements to estimate the parameters of the assumed normalized Gamma distribution (i.e., the intercept parameter NW, the median diameter D0, and the shape parameter μ).

Variables D0 and Nw are independently estimated using distinct NNs with three inputs (i.e., Z, ZDR, KDP), namely, NNNw and NND0, respectively. The shape parameter μ is estimated from ZDR and the retrieved values of D0 (as suggested in Brandes et al. 2002) using a two-input neural network NNμ.

Finally, Dm is computed from D0 and μ, that is, . We remark that the effects on the DSD parameters of considering a truncated Gamma distribution become more significant decreasing the maximum drop diameter (Adirosi et al. 2015).

The neural network architecture and regularization parameters were determined according to a heuristic monitoring of the generalization capability on test data, the root-mean-square error (RMSE) having been used as metric. According to what was suggested in Aires et al. (2002), it was found that the one-hidden-layer configuration improves the generalization capability of the NNs. The number of nodes in the hidden layer was fixed to 6 for NND0 and NNNw, whereas it was set to 12 for NNμ.

b. Dm–ZDR relationship and preliminary test

Variable Dm is the ratio of the fourth to the third moment of the DSD and is expressed as a function of the ZDR through a third-degree polynomial fit:
e3
where Dm is in millimeters and ZDR (dB) is the ratio of the radar reflectivity at horizontal (ZH) and vertical polarization (ZV). Parameter Dm is directly obtained from the DSD measurements while ZDR is obtained by simulating ZH and ZV at the operating radar frequency (i.e., C band for the Italian DPC GR) through the T-matrix scattering model (Mishchenko 2000). To simulate the radar observables, the axis ratio of the oblate spheroid is needed, and in this study the Brandes et al. (2002) model has been used, even if different models given in the literature (e.g., Beard and Chuang 1987; Andsager et al. 1999; Thurai and Bringi 2005; Thurai et al. 2007) have been tested resulting in very similar DmZDR relationships.

The GPM GV 2DVD data have been used to derive the DmZDR relationship by applying the sequential intensity filtering technique (SIFT; Lee and Zawadzki 2005). The SIFT method calculates the mean values of the retrieved variable Dm for each prescribed bin of the observed variable ZDR. For each bin, the mean ZDR is also calculated as long as the sample size is above a certain threshold (set to 10 samples in our work). The linear least squares method is then applied to the DmZDR pairs obtaining the coefficients a, b, c, and d of Eq. (3).

Figure 2 shows the DmZDR relationship derived for each GPM GV dataset and combining all the available data (ALL). The differences are negligible except for OLYMPEx, which shows lower Dm values for values of ZDR in the 1–2.5-dB range. Each DmZDR relationship is plotted up to the last ZDR bin exceeding the sample size threshold, highlighting also the different climatological characteristics of each field campaign (i.e., higher ZDR corresponds to the presence of larger drops or lower Dm indicates a higher number of small to medium drop sizes). The ALL DmZDR relationship has been taken as reference to derive Dm from the GR data and the mathematical expression is reported here:
eq1
Fig. 2.
Fig. 2.

The DmZDR relationships derived for each GV dataset and combining all together (ALL).

Citation: Journal of Hydrometeorology 19, 10; 10.1175/JHM-D-18-0002.1

Both the NN and DmZDR approaches have been applied to the 2DVD data as preliminary analysis and the estimated Dm have been compared with 2DVD measured Dm in order to test the reliability of both approaches (Fig. 3). When the NN approach is applied to the 2DVD data, the resulting Dm is highly correlated to the measured Dm (Fig. 3a), with a more marked spread of the distribution (the normalization is done with respect to the number of samples of each prescribed bin of the variable on the x axis) at larger Dm. For a given 2DVD-based Dm, the interval Dm(NN) ± 0.2 mm contains a fraction of samples ranging between 56% and 96% around 2.5 and 1 mm, respectively. If the interval Dm(NN) ± 0.3 mm is considered, the fraction of samples within it is in the range 70%–99%. Very similar results are obtained when the DmZDR relationship is applied to 2DVD data (Fig. 3b). In this case, the range of occurrence in the Dm ± 0.2 mm interval is between 65% and 96% around 2.5 and 1 mm, respectively. For 2DVD-measured Dm larger than 2.5–3 mm, the NN approach clearly outperforms the DmZDR method, since the latter method limits the estimated Dm around 3 mm.

Fig. 3.
Fig. 3.

Comparison between Dm measured by 2DVD and Dm estimated by applying the (a) NN approach and (b) DmZDR approach to the 2DVD data.

Citation: Journal of Hydrometeorology 19, 10; 10.1175/JHM-D-18-0002.1

The Dm estimated by applying the two approaches to the 2DVD data have been compared to each other (Fig. 4a). The agreement is excellent until Dm = 2 mm shows a very limited spread of the distribution, then for increasing Dm the DmZDR method produces a clear underestimation of the parameter. A similar behavior is shown when the Dm values estimated by applying the two approaches to the GR data are compared (Fig. 4b; the NN approach is routinely applied to the GR data). In this case, only the GR pixels matched with the DPR footprints have been considered. The agreement is still good even if the spread becomes slightly increased, while for larger Dm the underestimation of the DmZDR approach is still evident. The underestimation of large Dm with the DmZDR approach is mainly due to a couple of factors: the low number of samples in that range (only about 10% of data with Dm > 2 mm) and the higher spread (around 1.5 mm at ZDR = 3 dB) in the DmZDR space. At higher ZDR, Dm moves toward lower values. This can be explained by the presence of high concentration of small drops that contribute to lowering Dm.

Fig. 4.
Fig. 4.

Comparison between the Dm estimated by NN and DmZDR when the two approaches are applied to (a) 2DVD and (b) GR data.

Citation: Journal of Hydrometeorology 19, 10; 10.1175/JHM-D-18-0002.1

4. Reflectivity comparison

Before proceeding to comparisons of Dm, we first consider comparisons of reflectivities measured by the spaceborne and ground-based radars. The GPM radars work at Ka and Ku band, while the GR works at C band. To correctly compare the two measurements, the Ka-/Ku-band reflectivities have been scaled to C-band reflectivity through a polynomial fit by applying the SIFT methodology. To this end, the disdrometer dataset has been used to simulate the reflectivity at the different frequencies and to derive a second- and third-order polynomial relationship for Ku/C band (Fig. 5a) and Ka/C band (Fig. 5b), respectively. The magenta dots represent the mean values of ZC and ZKu/ZKa for each ZKu/ZKa bin after the SIFT methodology has been applied. The expressions of the two relationships are given by Eqs. (4) and (5):
e4
e5
where the reflectivities are in dBZ.
Fig. 5.
Fig. 5.

Relationship by simulated reflectivity at (a) C and Ku band and at (b) C and Ka band.

Citation: Journal of Hydrometeorology 19, 10; 10.1175/JHM-D-18-0002.1

At lower reflectivities (ZC < 18 dBZ), the difference between the Ka/Ku and C frequency is negligible. For larger reflectivity values, the ZKa becomes smaller than ZC even though most of the points are close to the one-to-one line. Differences between ZKu and ZC are much smaller.

To select the data, a threshold on ZC (after correction for ground clutter, attenuation, noise, etc.) has been set to 10 dBZ, while it is known that the minimum detectable signal is around 18 and 13 dBZ for DPR Ka-band and Ku-band radar, respectively. These values are obtained from an examination of the level 3 DPR histograms of Z corrected, while Ka HS minimum detectable signal is also about 13 dBZ, comparable to Ku band. Furthermore, only the DPR footprints matched with the GR pixels have been considered for the reflectivity comparison (as well as for all the following analysis), after the upscaling described in the section 2. The selected GR data can be compared with both measured and corrected (for ground clutter, attenuation, etc.; Meneghini et al. 2015) DPR reflectivity. While the DPR NS and HS measurements take into account only single-frequency information, the DPR MS measurements can be corrected by taking into account both single- (Ka only) and dual-frequency information. This aspect deserves an extensive and complete dissertation, which falls outside the main aim of this paper and can be the object of future studies. Consequently, the results shown in Fig. 6 are limited to comparisons between the scaled DPR ZKa/ZKu to ZC (hereafter ZKaC and ZKuC, respectively) and the GR ZC for stratiform precipitation over land for DPR NS, HS, and MS, considering for the latter product the corrected Ka/Ku reflectivities when the dual-frequency information is taken into account. We recall that DPR NS refers to the single-frequency information (2A-Ku NS products). Each plot also contains the reported bias (i.e., the mean difference between ZKaC/ZKuC and ZC) and the Pearson correlation coefficient (CC) between ZKaC/ZKuC and ZC.

Fig. 6.
Fig. 6.

Comparison between reflectivity measured by GR at C band (corrected by ground clutter, noise, etc.) and reflectivity measured by GPM radars scaled at C band for stratiform precipitation over land after being corrected by attenuation using (a) Ku-only information, (b) Ka-only information, and dual-frequency information for (c) Ka and (d) Ku band.

Citation: Journal of Hydrometeorology 19, 10; 10.1175/JHM-D-18-0002.1

The agreement is good for all the DPR scan types, even if there is a clear overestimation of the reflectivity ranging from 2.19 dBZ (DPR HS; Fig. 6b) to 5.57 dBZ (DPR NS; Fig. 6a). On the other hand, DPR HS has the lowest CC (0.64) while the DPR-based corrected ZKuC has the highest CC (0.70; Fig. 6c). For convective precipitation over land (not shown), the DPR-based correction works better (for both ZKaC and ZKuC) than the Ku-only correction (bias is 3.92 and 4.59 dBZ for DPR and Ku-only correction, respectively), while the DPR HS sample size is too small to make the comparison reliable. The roughly constant positive bias between DPR and GR, regardless of the reflectivity value, suggests an offset between the two measurements.

5. DPR and DPR–GMI performance

Both the DPR- and DPR–GMI–estimated Dm have been compared with the reference GR-based Dm. The selected samples have been divided by surface type (land/ocean) and by precipitation type (stratiform/convective/other). This classification is applied to all three DPR scan types (NS, MS, and HS) and to the two DPR–GMI scan types (NS and MS).

Figure 7 shows the two-dimensional density plot (normalized with the total number of samples) comparing Dm estimated by the DPR algorithm with Dm estimated by GR over land. The top row (Figs. 7a–c) shows the results for stratiform precipitation and the bottom row (Figs. 7d–f) for convective precipitation, while the columns from the left to the right correspond to DPR NS, MS, and HS, respectively. The “other” category has too few samples for all scan and surface types and is not shown. Although the results over the sea are not discussed because of the inhomogeneity of the sample sizes, the statistics are given in Table 1. All three DPR scan types show quite good agreement of Dm with the ground reference for stratiform precipitation despite a slight overestimation in the DPR estimate (Figs. 7a–c). The agreement, however, degrades significantly in convective precipitation where the overestimation in the DPR estimate is more marked (Figs. 7d–f). The overestimation can be partially explained by the positive bias between DPR and GR reflectivities (Fig. 6). Furthermore, the nonuniform beamfilling tends to be greater in convective precipitation, leading to larger errors in the attenuation correction of DPR algorithm. The DPR HS has a relatively small sample size for the convective precipitation, highlighting the difficulties of the Ka-band radar at heavy rainfall rates (Fig. 7f). Both MS and HS show some saturation of Dm at 3 mm (the upper limit in the DPR MS and HS retrieval algorithm, while this limit is not present for DPR NS). In most cases, these problems arise from overestimation in the path attenuation leading to very high Dm values, which are consequently set to the maximum value. Although the DPR team is working on mitigating these errors, the problems are quite challenging. As the size of the DPR field of view is about 5 km, significant nonuniform beamfilling can occur, particularly in convective rain. Moreover, in mountainous regions, the surface reference technique is often inapplicable because of the high spatial inhomogeneity in the surface cross sections, especially at near-nadir incidence. Although the Hitschfeld–Bordan attenuation correction can be used, this tends to become unstable in heavier rains, especially at Ka band. Despite these problems, the DPR MS shows a better agreement, relative to the DPR NS estimates, with the GR-estimated Dm for the convective precipitation.

Fig. 7.
Fig. 7.

Comparison between Dm estimated over land by GR and by (a) DPR NS stratiform, (b) DPR MS stratiform, (c) DPR HS stratiform, (d) DPR NS convective, (e) DPR MS convective, and (f) DPR HS convective.

Citation: Journal of Hydrometeorology 19, 10; 10.1175/JHM-D-18-0002.1

Table 1.

Sample size and statistics for Dm estimation of both DPR and DPR–GMI algorithm. For each indicator and for each type of precipitation (stratiform/convective) and surface (land/sea), the scan reporting the best performance has been bolded.

Table 1.

Figure 8 is the same as Fig. 7, but for the combined DPR–GMI–estimated Dm. The top (bottom) row shows the comparison between Dm estimated by the DPR–GMI algorithm with Dm estimated by GR over land for stratiform (convective) precipitation. As described in section 2a, the 2B-CMB product provides only two scan types, NS and MS. For the stratiform precipitation over land, the performance of DPR–GMI MS is better than the performance of DPR–GMI NS (Figs. 8a,b), while it is more similar for the convective precipitation (Figs. 8c,d), where the combined DPR–GMI–estimated Dm shows a smaller spread as compared to the DPR-estimated Dm. Furthermore, the systematic slight overestimation of DPR-based Dm shown in Fig. 7 is mitigated by the combination with GMI channels. Finally, the limiting value of Dm = 3 mm for DPR–GMI MS for convective precipitation is still evident.

Fig. 8.
Fig. 8.

Comparison between Dm estimated over land for (top) stratiform and (bottom) convective precipitation by GR and by the (a),(c) DPR–GMI NS and (b),(d) DPR–GMI MS algorithms.

Citation: Journal of Hydrometeorology 19, 10; 10.1175/JHM-D-18-0002.1

Table 1 presents the performances in estimating Dm in terms of absolute bias (AB), ratio bias (RB), standard deviation bias (SDB), and the fractional standard error (FSE) for DPR and DPR–GMI products, over land and over sea, and for each precipitation and scan type, as well as the sample size of each category. For each indicator and for each precipitation (stratiform/convective) and surface type (land/sea), the scan reporting the best performance has been bolded. The AB is calculated as the mean of the absolute difference between the Dm estimated by DPR/DPR–GMI and by GR. The RB is defined as the ratio between the mean Dm estimated by satellite and GR, the SDB is the standard deviation of the bias (difference between Dm estimated by satellite and GR), and the FSE is computed as the ratio between the RMSE (Nurmi 2003) and the average of the observations at the ground (expressed in percentage values), respectively.

The AB is generally low for stratiform precipitation with comparable performances for DPR NS and MS (AB does not exceed 0.31 mm both over land and over sea); for DPR HS, the AB increases both over land and over sea. For convective precipitation, DPR MS outperforms DPR NS, indicating that the dual-frequency signal gives better results during convective precipitation, even if AB increases reaching 0.79 mm for DPR NS over land. Higher AB values are reached for DPR HS, but the sample size (for both sea and land) is too small to make the results significant. Over sea, the categories with a sufficient number of samples (DPR NS and HS for stratiform precipitation) show very similar or even better (convective cases) AB values to the corresponding categories over land. This could be also related to the fact that the matching point over sea is relatively higher than over land and generally smaller particles are observed.

As reported previously, the AB for DPR MS and HS are affected by saturated Dm samples, but DPR MS outperforms DPR NS, indicating that the dual-frequency signal gives better results during convective precipitation. We recall that DPR NS is obtained from 2A-Ku NS products, which use the single-frequency information. Currently, there is no way to know the samples where the saturation issue is present and, consequently, there is no way to test the performance of DPR MS and HS with these saturated data removed. To get an idea of how much AB could be reduced, cases where Dm(DPR) = 3 mm (subjective choice) were discarded. As a consequence of this filtering, a marked improvement is achieved by DPR HS, where the AB decreases to 0.41 and 0.69 mm over land for stratiform and convective precipitation, respectively, while decreasing by more than half over sea. On the other hand, the corresponding number of samples decreases to 1962 (60) and 590 (19) for stratiform (convective) precipitation over land and over sea, respectively. The improvement of AB for DPR MS is around 0.04–0.05 mm for convective precipitation, while it is negligible for the stratiform precipitation. At the same time, the percentage of lost samples does not exceed 8%. These results show that removal of the saturated data improves the HS scan data much more than MS data.

The error in estimating Dm over land is lower for DPR–GMI MS for all types of precipitation (0.23 and 0.40 mm for stratiform and convective rain, respectively), while over sea the performances of DPR–GMI NS and DPR–GMI MS are comparable, even if the former shows better results for the convective precipitation. The “other” category has a very small sample size, and its performance should not be considered because of the poor reliability. While the DPR–GMI NS has roughly double the number of samples because of the scan geometry, the convective precipitation over sea presents a limited sample size that could affect the corresponding error for both scans.

The RB is close to one (perfect score) for stratiform precipitation and for DPR NS/MS regardless of the surface type, while it increases for convective precipitation where DPR MS outperforms DPR NS. DPR HS has generally higher values, indicating worse performance. The SDB is generally low, around 0.35 mm for stratiform precipitation and DPR NS/MS, while it reaches higher values for DPR HS and convective precipitation. Similar results are obtained from the FSE, highlighting the better reliability of DPR MS for convective precipitation and the better performance of DPR NS for stratiform precipitation.

The DPR–GMI products show RB values that are smaller and closer to one for both stratiform and convective precipitation over land as compared with the DPR products (NS and MS). On the other hand, RB is lower than one over sea, confirming a slight underestimation of Dm by DPR–GMI for these cases. SDB and FSE are similar with respect to the same indicators for the DPR products, showing comparable performance for the two algorithms. The DPR–GMI MS is consistently better than DPR–GMI NS, regardless of the surface or precipitation type.

A more detailed analysis regarding the reliability of the GPM products is addressed by considering the distribution of the error of the single measure through the percentage error (PE), defined as
eq2
where SAT referrers to Dm estimated by DPR or DPR–GMI algorithm and GR refers to Dm estimated by GR. Figure 9 shows the distribution of PE over land for the DPR algorithm (Fig. 9a) and for the combined DPR–GMI algorithm (Fig. 9b) for stratiform and convective precipitation (blue and red lines, respectively, while the different line styles refer to the different scan types). It should be noted that the last class (PE = 100%) contains all the PE ≥ 100%.
Fig. 9.
Fig. 9.

Distribution of PE for the (a) DPR algorithm and (b) DPR–GMI algorithm. The blue and red lines refer to stratiform and convective classification, respectively. The different line styles refer to the DPR scan type.

Citation: Journal of Hydrometeorology 19, 10; 10.1175/JHM-D-18-0002.1

Generally, the distribution of PE is more symmetric with respect to PE = 0% for stratiform precipitation than for convective precipitation for DPR NS/MS (Fig. 9a) and for DPR–GMI (Fig. 9b). The occurrence peak for DPR–GMI is shifted toward negative PE for stratiform precipitation with respect to the DPR products. The overestimation of Dm during convective precipitation is clear for both algorithms, even if it is more marked for DPR products with a large number of the sample exceeding PE = 100%.

6. Sensitivity analysis

Some sensitivity analyses have been carried out in order to further test the reliability of the results and to potentially find reasons for the disagreement between satellite and ground-based estimates of Dm.

a. Distance from GR

A matching pair of points is defined as the intersection of the GR and DPR beams. Potential matches range from the first DPR clutter-free bin (which depends on the radar incidence angle and on the topography and surface characteristics) to the highest level with liquid precipitation (which depends on the characteristics of the rain event).

The farther the distance from the radar, the larger the volume of the radar resolution cell, regardless of elevation angle. As the volume increases, the variability of the rain structure within can become significant, leading to an NUBF issue with the ground radar. Figure 10 shows the variation of mean PE ± one standard deviation with respect to the distance of the matching point from the GR, sampling in classes of 15-km width. Figures 10a and 10b refer to stratiform and convective precipitation, respectively, while the colors refer to the different products (viz., blue is DPR NS, red is DPR MS, black is DPR HS, green is DPR–GMI NS, and magenta is DPR–GMI MS). Generally, the distance of the matching point from the GR does not exceed 150 km, and most points are located at distances less than 100 km. The DPR HS does not show any sample at distance from GR lower than 50 km for convective precipitation (Fig. 10b). Neither the DPR nor the DPR–GMI algorithm show a deterioration in performance with increasing the distance from GR. Parameter ΔDm is generally positive with the exception of DPR–GMI products beyond 60 km. DPR–GMI products have PE values closer to 0% with respect to the DPR products, both for stratiform and convective precipitation.

Fig. 10.
Fig. 10.

Mean ± standard deviation of PE (expressed as difference between GPM and GR Dm estimations) with respect to the distance of the matching point from GR for (a) stratiform and (b) convective precipitation. The line colors are blue for DPR NS, red for DPR MS, black for DPR HS, green for DPR–GMI NS, and magenta for DPR–GMI MS.

Citation: Journal of Hydrometeorology 19, 10; 10.1175/JHM-D-18-0002.1

Regardless of the distance from GR and the precipitation type, MS (red points) has comparable or lower variance than NS, both for DPR and DPR–GMI algorithm, while it is higher for DPR HS. Furthermore, the DPR–GMI algorithm shows generally lower variance than DPR for convective precipitation, while it is comparable for the stratiform precipitation. Even if the LBM GR scan strategy does not ensure a straightforward relationship between the distance from the GR and the height of the matching point, a dedicated analysis does not present any different results with respect to what is shown in Fig. 10 (i.e., there is no dependence of PE on the height of the matching point).

b. NUBF analysis

The inhomogeneity of the rain structure in space and time can affect the rain measurement at different scales (Tokay et al. 2016, 2017), resulting in NUBF situations. We already mentioned that the GPM algorithms, to some degree, take into account this effect (Iguchi et al. 2017), just as the different spatial resolutions of the DPR and GR are considered in the comparison process. This resolution difference allows us to analyze in more detail the effects of NUBF on the Dm estimation. To analyze the impact of NUBF, two indicators have been considered: the percentage of GR pixels below the minimum detectable signal (10 dBZ) and the ZC standard deviation of GR pixels above the minimum detectable signal within a DPR footprint.

The NUBF cases ranged between 11% (DPR NS, stratiform) and 21% (DPR MS, convective) of the total number of samples over land. The DPR HS has the same occurrence of NUBF cases (13%) both for stratiform and convective precipitation, while over sea the frequency of NUBF does not change appreciably for the different DPR products with respect to the corresponding DPR products over land. Comparable NUBF occurrence is obtained for the DPR–GMI algorithm, with 13% (DPR–GMI NS, stratiform) and 24% (DPR–GMI MS, convective) over land.

Figure 11 shows the dependence of PE on the NUBF expressed in terms of the standard deviation of ZC and as the percentage of radar pixels with ZC < 10 dBZ within a DPR footprint. The dots and squares refer to stratiform and convective precipitation, respectively.

Fig. 11.
Fig. 11.

Dependence of PE (expressed as difference between GPM and GR Dm estimations) with respect to the number of no-rain GR pixels and their standard deviation within a DPR footprint for DPR (a) NS, (b) MS, and (c) HS algorithm and for DPR–GMI (d) NS and (e) MS.

Citation: Journal of Hydrometeorology 19, 10; 10.1175/JHM-D-18-0002.1

The standard deviation of GR reflectivity attains high values (up to 7–8 dBZ) for both stratiform and convective precipitation, even though the former is, on average, 0.7–1 dBZ lower than the latter. Generally, the standard deviation of ZC increases when the number of GR pixels below the minimum detectable signal decreases. PE is mainly positive, especially for the DPR algorithm (Figs. 11a–c), while the PE for DPR–GMI (Figs. 11d–e) is closer to 0%. However, there is not a clear relationship between the magnitude of NUBF (interpreted both as percentage of no-signal GR pixels and as ZC standard deviation) and the magnitude of ΔDm. At the same time, a low percentage of no-signal GR pixels with high ZC standard deviation can produce the same error as a high percentage of no-signal GR pixels with low ZC standard deviation. This can be observed in both the DPR and DPR–GMI algorithms. The FSE for the NUBF samples increases for convective precipitation by 22% and 16% for DPR NS and MS, respectively, while for stratiform precipitation the increase is about 16% for both scans. For stratiform precipitation, the error for the DPR-HS scan is lower (by about 8%) than that reported in Table 1. This is probably because the DPR-estimated Dm values are around 1 mm and are not affected by the saturation issue (Fig. 7). For the DPR–GMI results, the increase of FSE is around 4%–7% for all products, indicating that the DPR–GMI algorithm is less sensitive to the effects of NUBF. These results reveal the difficulties for satellite algorithms to correctly estimate Dm in either patchy light rain or during heavy variable rain events. Furthermore, Fig. 11 shows how the two contributions are interdependent and cannot be separated.

c. Reflectivity analysis for missing Dm

In a small number of cases, the DPR algorithm reported a missing Dm. In this section the reflectivity measured by both DPR (i.e., the signal not corrected for attenuation, ground clutter, etc.) and GR are compared. Figure 12a shows the ZC and ZKu scatterplot for the samples with missing Dm, while Fig. 12b refers to the comparison between the ZC and ZKa, where the blue and red colors refer to stratiform and convective precipitation, respectively. In Fig. 12a the dots and squares identify the inner and outer swath, respectively, and in Fig. 12b they indicate the measured ZKa by DPR HS and MS, respectively.

Fig. 12.
Fig. 12.

Reflectivity measured by (a) DPR ZKu and (b) DPR ZKa and GR (ZC) for the cases where the DPR algorithm reported Dm as missed value.

Citation: Journal of Hydrometeorology 19, 10; 10.1175/JHM-D-18-0002.1

Generally, when the GR measured low reflectivity values (between 10 and 20 dBZ), the corresponding ZKa/ZKu values were either below the DPR minimum detectable signal (around 18 and 13 dBZ for Ka-band and Ku-band radar, respectively) or were at high values, indicating the possible contamination by ground clutter. A limited number of samples reported both high ZC and ZKa/ZKu values, indicating the presence of clutter in both satellite and ground measurements. A few samples consist of high DPR reflectivities (around 35 and 52 dBZ for ZKa and ZKu, respectively) paired with reasonable ZC values (around 30 dBZ) and could be caused by surface clutter signals either from the main or from side lobes of the DPR antenna. Furthermore, Figs. 12a and 12b do not show any difference between ZKa measured by DPR HS and MS or between ZKu measured in inner and outer swaths. This analysis also indirectly emphasizes the characteristics of the DPR radars (i.e., their sensitivity) as well as the goodness of the DPR algorithms in recognizing and filtering situations affected by ground clutter.

7. Conclusions

In this work, the GPM level 2A-DPR and level 2B-CMB mass-weighted mean diameter Dm outputs were compared with the reference Dm estimated by GRs for 14 selected case studies over Italy in the summer and fall seasons between the years 2015 and 2017. Only DPR measurements labeled as liquid precipitation were considered in this study. This makes the analysis challenging because of the complexity of the terrain and the developing of different precipitating systems (i.e., stratiform, convective, stratiform with embedded convection). The GRs used in this study, managed by the DPC, are C-band radars while the GPM Core Observatory radars operate at Ka band and Ku band.

The present work shows the reliability of Dm estimates based on DPR (and DPR–GMI) at midlatitude. It can be considered as a guideline in using the spaceborne DSD parameter (i.e., Dm) estimates to characterize the rain characteristics, especially over areas not covered by ground-based instruments (i.e., radars and disdrometers) able to directly measure or estimate the DSD parameters.

To validate the ground reference, two different approaches have been used to estimate Dm from GR data. The operational chain of DPC uses a neural network (NN) approach to derive the rain variables (i.e., mass-weighted mean diameter, rain rate, etc.) from the GR measurements. In addition, Dm was also estimated through a DmZDR relationship derived by simulating the radar variables at C band from disdrometer observations. The two approaches showed very similar Dm estimation (for a given Dm, at least the 85% of the samples are within Dm ± 0.2), with a slight discrepancy at larger Dm.

The disdrometer data have been also used to derive a relationship between reflectivities at C and Ka/Ku band. This allows the approximate conversion of the Ka-/Ku-band reflectivities to C band to facilitate the comparisons of quantities measured by GR and DPR, after the correction by attenuation, ground clutter, etc. The agreement between ZKaC/ZKuC and ZC is quite good both when the DPR measurement takes into account the single- and double-frequency information, with correlation coefficients ranging between 0.64 and 0.70), while the bias (difference between ZKaC/ZKuC and ZC) is generally positive.

The comparison between the Dm estimated by both DPR and the DPR–GMI algorithm and by GR gives good results. The AB is generally lower than 0.5 mm for stratiform precipitation and for the DPR NS/MS and DPR–GMI NS/MS (for DPR HS it is 0.56 mm). However, the performance significantly decreases for convective precipitation for the DPR algorithm (AB ranges between 0.64 mm for DPR MS and 0.91 mm for DPR HS), while the deterioration is much smaller for the DPR–GMI algorithm (AB is 0.43 and 0.40 mm for DPR–GMI NS and MS, respectively). Both the DPR MS and HS sometimes yield saturated Dm values, which might be caused by an overestimation in the attenuation correction (S. Seto 2017, personal communication) resulting in the Dm to be set to its maximum value (3 mm for DPR MS and HS). Although, at the moment, there is no way to identify the saturated samples, a subjective sensitivity analysis shows an improvement of 0.15 and 0.22 mm in AB for stratiform and convective precipitation, respectively, over land for DPR HS when these samples are discarded (the improvement is almost negligible for DPR MS). Over the sea, for classes with significant samples sizes, the performance does not differ appreciably with respect to the performance of the corresponding class over land [i.e., AB of DPR (DPR–GMI) NS for stratiform precipitation over land and over sea is 0.29 (0.31) and 0.26 (0.29) mm, respectively]. Moreover, the PE shows an overestimation of Dm by DPR and DPR–GMI (more marked for the DPR algorithm) for convective precipitation, while it is close to normally distributed around 0% for stratiform precipitation even though there is a higher occurrence of positive PE.

A sensitivity study did not show a deterioration of the Dm estimation with respect to the distance (or the height, not shown) of the matching point from the GR. This feature has been observed for both stratiform and convective precipitation for the DPR and DPR–GMI algorithms. The impact of nonuniform beam filling (NUBF) has been addressed both in terms of the number of GR pixels not exceeding the minimum detectable DPR signal and in terms of ZC standard deviation within a DPR footprint. The larger Dm difference between GPM products and GR can be due to a high number of no-signal GR pixels (even with a low ZC standard deviation) as well as higher ZC standard deviations (up to 7–8 dBZ) but with a negligible number of GR pixels below the signal threshold.

Acknowledgments

The 2A-DPR V05 GPM data were downloaded from the Precipitation Processing System (ftp://arthurhou.pps.eosdis.nasa.gov/).

REFERENCES

  • Adirosi, E., L. Baldini, F. Lombardo, F. Russo, F. Napolitano, E. Volpi, and A. Tokay, 2015: Comparison of different fittings of drop spectra for rainfall retrievals. Adv. Water Resour., 83, 5567, https://doi.org/10.1016/j.advwatres.2015.05.009.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Aires, F., A. Chédin, N. Scott, and W. B. Rossow, 2002: A regularized neural network approach for retrieval of atmospheric and surface temperatures with the IASI instrument. J. Appl. Meteor., 41, 144159, https://doi.org/10.1175/1520-0450(2002)041<0144:ARNNAF>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Andsager, K., K. V. Beard, and N. F. Laird, 1999: Laboratory measurements of axis ratios for large rain drops. J. Atmos. Sci., 56, 26732683, https://doi.org/10.1175/1520-0469(1999)056<2673:LMOARF>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Battaglia, A., S. Tanelli, G. M. Heymsfield, and L. Tian, 2014: The dual wavelength ratio knee: A signature of multiple scattering in airborne Ku–Ka observations. J. Appl. Meteor. Climatol., 53, 17901808, https://doi.org/10.1175/JAMC-D-13-0341.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Beard, K. V., and C. Chuang, 1987: A new model for the equilibrium shape of raindrops. J. Atmos. Sci., 44, 15091524, https://doi.org/10.1175/1520-0469(1987)044<1509:ANMFTE>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bech, J., B. Codina, J. Lorente, and D. Bebbington, 2003: The sensitivity of single polarization weather radar beam blockage correction to variability in the vertical refractivity gradient. J. Atmos. Oceanic Technol., 20, 845855, https://doi.org/10.1175/1520-0426(2003)020<0845:TSOSPW>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Brandes, E., G. Zhang, and J. Vivekanandan, 2002: Experiments in rainfall estimation with a polarimetric radar in a subtropical environment. J. Appl. Meteor., 41, 674685, https://doi.org/10.1175/1520-0450(2002)041<0674:EIREWA>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bringi, V. N., V. Chandrasekar, N. Balakrishnan, and D. S. Zrnić, 1990: An examination of propagation effects in rainfall on radar measurements at microwave frequencies. J. Atmos. Oceanic Technol., 7, 829840, https://doi.org/10.1175/1520-0426(1990)007<0829:AEOPEI>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cao, Q., G. Zhang, E. Brandes, and T. Schuur, 2010: Polarimetric radar rain estimation through retrieval of drop size distribution using a Bayesian approach. J. Appl. Meteor. Climatol., 49, 973990, https://doi.org/10.1175/2009JAMC2227.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Cao, Q., G. Zhang, and M. Xue, 2013: A variational approach for retrieving raindrop size distribution from polarimetric radar measurements in the presence of attenuation. J. Appl. Meteor. Climatol., 52, 169185, https://doi.org/10.1175/JAMC-D-12-0101.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • D’Adderio, L. P., F. Porcu, and A. Tokay, 2015: Raindrop size distribution in the presence of break-up. J. Atmos. Sci., 72, 34043416, https://doi.org/10.1175/JAS-D-14-0304.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Drobinski, P., and et al. , 2014: HyMeX: A 10-year multidisciplinary program on the Mediterranean water cycle. Bull. Amer. Meteor. Soc., 95, 10631082, https://doi.org/10.1175/BAMS-D-12-00242.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Grecu, M., W. S. Olson, S. J. Munchak, S. Ringerud, L. Liao, Z. Haddad, B. L. Kelley, and S. F. McLaughlin, 2016: The GPM combined algorithm. J. Atmos. Oceanic Technol., 33, 22252245, https://doi.org/10.1175/JTECH-D-16-0019.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hou, A. Y., and et al. , 2014: The Global Precipitation Measurement Mission. Bull. Amer. Meteor. Soc., 95, 701722, https://doi.org/10.1175/BAMS-D-13-00164.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Houze, R., Jr., and et al. , 2017: The Olympic Mountains Experiment (OLYMPEX). Bull. Amer. Meteor. Soc., 98, 21672188, https://doi.org/10.1175/BAMS-D-16-0182.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Iguchi, T., and et al. , 2016: Precipitation rates estimated with GPM’s dual-frequency radar. 2016 IEEE Int. Geoscience and Remote Sensing Symp., Beijing, China, IEEE, 3918, https://doi.org/10.1109/IGARSS.2016.7730017.

    • Crossref
    • Export Citation
  • Iguchi, T., S. Seto, R. Meneghini, N. Yoshida, J. Awaka, M. Le, V. Chandrasekar, and T. Kubota, 2017: GPM/DPR level-2. Algorithm Theoretical Basis Doc., 81 pp., http://www.eorc.jaxa.jp/GPM/do AU6 c/algorithm/ATBD_DPR_201708_whole_1.pdf.

  • Kruger, A., and W. F. Krajewski, 2002: Two-dimensional video disdrometer: A description. J. Atmos. Oceanic Technol., 19, 602617, https://doi.org/10.1175/1520-0426(2002)019<0602:TDVDAD>2.0.CO;2.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kubota, T., T. Iguchi, M. Kojima, L. Liao, T. Masaki, H. Hanado, R. M. Meneghini, and R. Oki, 2016: A statistical method for reducing sidelobe clutter for the Ku-band precipitation radar onboard the GPM Core Observatory. J. Atmos. Oceanic Technol., 33, 14131428, https://doi.org/10.1175/JTECH-D-15-0202.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Lee, G. W., and I. Zawadzki, 2005: Variability of drop size distributions: Time-scale dependence of the variability and its effects on rain estimation. J. Appl. Meteor., 44, 634652, https://doi.org/10.1175/JAM2222.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Meneghini, R., H. Kim, L. Liao, J. A. Jones, and J. M. Kwiatkowski, 2015: An initial assessment of the surface reference technique applied to data from the Dual-Frequency Precipitation Radar (DPR) on the GPM satellite. J. Atmos. Oceanic Technol., 32, 22812296, https://doi.org/10.1175/JTECH-D-15-0044.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mishchenko, M. I., 2000: Calculation of the amplitude matrix for a nonspherical particle in a fixed orientation. Appl. Opt., 39, 10261031, https://doi.org/10.1364/AO.39.001026.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Mroz, K., A. Battaglia, T. J. Lang, D. J. Cecil, S. Tanelli, and F. Tridon, 2017: Hail-detection algorithm for the GPM Core Observatory satellite sensors. J. Appl. Meteor. Climatol., 56, 19391957, https://doi.org/10.1175/JAMC-D-16-0368.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ni, X., C. Liu, D. J. Cecil, and Q. Zhang, 2017: On the detection of hail using satellite passive microwave radiometers and precipitation radar. J. Appl. Meteor. Climatol., 56, 26932709, https://doi.org/10.1175/JAMC-D-17-0065.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Nurmi, P., 2003: Recommendations on the verification of local weather forecasts. ECMWF Tech. Memo. 430, 19 pp., https://www.ecmwf.int/en/elibrary/11401-recommendations-verification-local-weather-forecasts.

  • Panegrossi, G., and et al. , 2016: Use of the GPM constellation for monitoring heavy precipitation events over the Mediterranean region. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 9, 27332753, https://doi.org/10.1109/JSTARS.2016.2520660.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Petracca, M., L. P. D’Adderio, F. Porcù, G. Vulpiani, S. Sebastianelli, and S. Puca, 2018: Validation of GPM Dual-Frequency Precipitation Radar (DPR) rainfall products over Italy. J. Hydrometeor., 19, 907925, https://doi.org/10.1175/JHM-D-17-0144.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Seto, S., T. Iguchi, and T. Oki, 2013: The basic performance of a precipitation retrieval algorithm for the Global Precipitation Measurement mission’s single/dual frequency radar measurements. IEEE Trans. Geosci. Remote Sens., 51, 52395251, https://doi.org/10.1109/TGRS.2012.2231686.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Seto, S., T. Shimozumal, T. Iguchi, and T. Oki, 2016: Spatial and temporal variations of mass-weighted mean diameter estimated by GPM/DPR. IEEE Int. Geoscience and Remote Sensing Symp., Beijing, China, IEEE, 3938–3940, https://doi.org/10.1109/IGARSS.2016.7730023.

    • Crossref
    • Export Citation
  • Speirs, P., M. Gabella, and A. Berne, 2017: A comparison between the GPM Dual-Frequency Precipitation Radar and ground-based radar precipitation rate estimates in the Swiss Alps and Plateau. J. Hydrometeor., 18, 12471269, https://doi.org/10.1175/JHM-D-16-0085.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tabary, P., 2007: The new French operational radar rainfall product. Part I: Methodology. Wea. Forecasting, 22, 393408, https://doi.org/10.1175/WAF1004.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Thurai, M., and V. N. Bringi, 2005: Drop axis ratios from 2D video disdrometer. J. Atmos. Oceanic Technol., 22, 966978, https://doi.org/10.1175/JTECH1767.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Thurai, M., G. Huang, V. Bringi, W. Randeu, and M. Schönhuber, 2007: Drop shapes, model comparisons, and calculations of polarimetric radar parameters in rain. J. Atmos. Oceanic Technol., 24, 10191032, https://doi.org/10.1175/JTECH2051.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tokay, A., L. P. D’Adderio, D. B. Wolff, and W. A. Petersen, 2016: A field study of pixel scale variability of raindrop size distribution in the Mid-Atlantic region. J. Hydrometeor., 17, 18551868, https://doi.org/10.1175/JHM-D-15-0159.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Tokay, A., L. P. D’Adderio, F. Porcù, D. B. Wolff, and W. A. Petersen, 2017: A Field study of footprint-scale variability of raindrop size distribution. J. Hydrometeor., 18, 31653179, https://doi.org/10.1175/JHM-D-17-0003.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vulpiani, G., F. S. Marzano, V. Chandrasekar, A. Berne, and R. Uijlenhoet, 2006: Polarimetric weather radar retrieval of raindrop size distribution by means of a regularized artificial neural network. IEEE Trans. Geosci. Remote Sens., 44, 32623275, https://doi.org/10.1109/TGRS.2006.878438.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vulpiani, G., S. Giangrande, and F. S. Marzano, 2009: Rainfall estimation from polarimetric S-band radar measurements: Validation of a neural network approach. J. Appl. Meteor. Climatol., 48, 20222036, https://doi.org/10.1175/2009JAMC2172.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vulpiani, G., M. Montopoli, L. Delli Passeri, A. Gioia, P. Giordano, and F. S. Marzano, 2012: On the use of dual-polarized C-band radar for operational rainfall retrieval in mountainous areas. J. Appl. Meteor. Climatol., 51, 405425, https://doi.org/10.1175/JAMC-D-10-05024.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Vulpiani, G., L. Baldini, and N. Roberto, 2015: Characterization of Mediterranean hail-bearing storms using an operational polarimetric X-band radar. Atmos. Meas. Tech., 8, 46814698, https://doi.org/10.5194/amt-8-4681-2015.

    • Crossref
    • Search Google Scholar
    • Export Citation
Save