• Anagnostou, E. N., 2004: A convective/stratiform precipitation classification algorithm for volume scanning weather radar observations. Meteor. Appl., 11, 291300, doi:10.1017/S1350482704001409.

    • Search Google Scholar
    • Export Citation
  • Atlas, D., , Ulbrich C. W. , , Marks F. D. Jr., , Amitai E. , , and Williams C. R. , 1999: Systematic variation of drop-size and radar-rainfall relations. J. Geophys. Res., 104, 61556169, doi:10.1029/1998JD200098.

    • Search Google Scholar
    • Export Citation
  • Awaka, J., , Iguchi T. , , Kumagai H. , , and Okamoto K. , 1997: Rain type classification algorithm for TRMM precipitation radar. IGARSS’97: 1997 International Geoscience and Remote Sensing Symposium, T. I. Stein, Ed., Vol. 4, IEEE, 16331635. doi:10.1109/IGARSS.1997.608993.

  • Awaka, J., , Iguchi T. , , and Okamoto K. , 2009: TRMM PR standard algorithm 2A23 and its performance on bright band detection. J. Meteor. Soc. Japan, 87A, 3152, doi:10.2151/jmsj.87A.31.

    • Search Google Scholar
    • Export Citation
  • Barnes, H. C. & , and Houze R. A. Jr. 2013: The precipitating cloud population of the Madden-Julian Oscillation over the Indian and west Pacific Oceans. J. Geophys. Res. Atmos., 118, 69967023, doi:10.1002/jgrd.50375.

    • Search Google Scholar
    • Export Citation
  • Biggerstaff, M. I., , and Houze R. A. Jr., 1993: Kinematics and microphysics of the transition zone of the 10–11 June 1985 squall line. J. Atmos. Sci., 50, 30913110, doi:10.1175/1520-0469(1993)050<3091:KAMOTT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Biggerstaff, M. I., , and Listemaa S. A. , 2000: An improved scheme for convective/stratiform echo classification using radar reflectivity. J. Appl. Meteor., 39, 21292150, doi:10.1175/1520-0450(2001)040<2129:AISFCS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Braun, S. A., , and Houze R. A. Jr., 1994: The transition zone and secondary maximum of radar reflectivity behind a midlatitude squall line: Results retrieved from Doppler radar data. J. Atmos. Sci., 51, 27332755, doi:10.1175/1520-0469(1994)051<2733:TTZASM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Bringi, V. N., , Williams C. R. , , Thurai M. , , and May P. T. , 2009: Using dual-polarized radar and dual-frequency profiler for DSD characterization: A case study from Darwin, Australia. J. Atmos. Oceanic Technol., 26, 21072122, doi:10.1175/2009JTECHA1258.1.

    • Search Google Scholar
    • Export Citation
  • Churchill, D. D., , and Houze R. A. Jr., 1984: Development and structure of winter monsoon cloud clusters on 10 December 1978. J. Atmos. Sci., 41, 933960, doi:10.1175/1520-0469(1984)041<0933:DASOWM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Ek, M. B., , Mitchell K. E. , , Lin Y. , , Rogers E. , , Grunmann P. , , Koren V. , , Gayno G. , , and Tarpley J. D. , 2003: Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta model. J. Geophys. Res., 108, 8851, doi:10.1029/2002JD003296.

    • Search Google Scholar
    • Export Citation
  • Hagos, S., , Feng Z. , , Landu K. , , and Long C. N. , 2014: Advection, moistening and shallow-to-deep convection transitions during the initiation and propagation of Madden-Julian Oscillation. J. Adv. Model. Earth Syst., 6, 938949, doi:10.1002/2014MS000335.

    • Search Google Scholar
    • Export Citation
  • Houze, R. A., Jr., 1973: A climatological study of vertical transports by cumulus-scale convection. J. Atmos. Sci., 30, 11121123, doi:10.1175/1520-0469(1973)030<1112:ACSOVT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Hubbert, J. C., , Dixon M. , , Ellis S. M. , , and Meymaris G. , 2009a: Weather radar ground clutter. Part I: Identification, modeling, and simulation. J. Atmos. Oceanic Technol., 26, 11651180, doi:10.1175/2009JTECHA1159.1.

    • Search Google Scholar
    • Export Citation
  • Hubbert, J. C., , Dixon M. , , and Ellis S. M. , 2009b: Weather radar ground clutter. Part II: Real-time identification and filtering. J. Atmos. Oceanic Technol., 26, 11811197, doi:10.1175/2009JTECHA1160.1.

    • Search Google Scholar
    • Export Citation
  • Iacono, M. J., , Delamere J. S. , , Mlawer E. J. , , Shepard M. W. , , Clough S. A. , , and Collins W. D. , 2008: Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, doi:10.1029/2008JD009944.

    • Search Google Scholar
    • Export Citation
  • Janjić, Z. I., 1994: The step-mountain eta coordinate model: Further developments of the convection, viscous layer, and turbulence closure schemes. Mon. Wea. Rev., 122, 927945, doi:10.1175/1520-0493(1994)122<0927:TSMECM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Liebe, H. J., 1985: An updated model for millimeter wave propagation in moist air. Radio Sci., 20, 10691089, doi:10.1029/RS020i005p01069.

    • Search Google Scholar
    • Export Citation
  • Mapes, B. E., , Tulich S. , , Lin J.-L. , , and Zuidema P. , 2006: The mesoscale convection life cycle: Building block or prototype for large-scale tropical waves? Dyn. Atmos. Oceans, 42, 329, doi:10.1016/j.dynatmoce.2006.03.003.

    • Search Google Scholar
    • Export Citation
  • Marshall, J. S., , and Palmer W. M. , 1948: The distribution of raindrops with size. J. Meteor., 5, 165166, doi:10.1175/1520-0469(1948)005<0165:TDORWS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Penide, G., , Protat A. , , Kumar V. V. , , and May P. T. , 2013: Comparison of two convective/stratiform precipitation classification techniques: Radar reflectivity texture versus drop size distribution–based approach. J. Atmos. Oceanic Technol., 30, 27882797, doi:10.1175/JTECH-D-13-00019.1.

    • Search Google Scholar
    • Export Citation
  • Powell, S. W., , and Houze R. A. Jr., 2013: The cloud population and onset of the Madden-Julian Oscillation over the Indian Ocean during DYNAMO-AMIE. J. Geophys. Res. Atmos., 118, 11 97911 995, doi:10.1002/2013JD020421.

    • Search Google Scholar
    • Export Citation
  • Roberto, N., and Coauthors, 2016: Multi-sensor analysis of convective activity in central Italy during the HyMeX SOP 1.1. Atmos. Meas. Tech., 9, 535552, doi:10.5194/amt-9-535-2016.

    • Search Google Scholar
    • Export Citation
  • Rutledge, S. A., , and Houze R. A. Jr., 1987: A diagnostic modeling study of the trailing stratiform region of a midlatitude squall line. J. Atmos. Sci., 44, 26402656, doi:10.1175/1520-0469(1987)044<2640:ADMSOT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Ryzhkov, A. V., , Giangrande S. E. , , and Schuur T. J. , 2005: Rainfall estimates with a polarimetric prototype of WSR-88D. J. Appl. Meteor., 44, 502515, doi:10.1175/JAM2213.1.

    • Search Google Scholar
    • Export Citation
  • Schumacher, C., , and Houze R. A. Jr., 2003: The TRMM precipitation radar’s view of shallow, isolated rain. J. Appl. Meteor., 42, 15191524, doi:10.1175/1520-0450(2003)042<1519:TTPRVO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Skamarock, W. C., and Coauthors, 2008: A description of the Advanced Research WRF version 3. NCAR Tech. Note NCAR/TN-475+STR, 113 pp., doi:10.5065/D68S4MVH.

  • Steiner, M., , and Houze R. A. Jr., 1993: Three-dimensional validation at TRMM ground truth sites: Some early results from Darwin, Australia. Preprints, 26th Conf. on Radar Meteorology, Norman, OK, Amer. Meteor. Soc., 447450.

  • Steiner, M., , and Smith J. A. , 2002: Use of three-dimensional reflectivity structure for automated detection and removal of nonprecipitating echoes in radar data. J. Atmos. Oceanic Technol., 19, 673686, doi:10.1175/1520-0426(2002)019<0673:UOTDRS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Steiner, M., , Houze R. A. Jr., , and Yuter S. E. , 1995: Climatological characterization of three-dimensional storm structure from operational radar and rain gauge data. J. Appl. Meteor., 34, 19782007, doi:10.1175/1520-0450(1995)034<1978:CCOTDS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Thompson, E. J., , Rutledge S. A. , , Dolan B. , , and Thurai M. , 2015: Drop size distributions and radar observations of convective and stratiform rain over the equatorial Indian and West Pacific Oceans. J. Atmos. Sci., 72, 40914125, doi:10.1175/JAS-D-14-0206.1.

    • Search Google Scholar
    • Export Citation
  • Thompson, G., , Field P. R. , , Rasmussen R. M. , , and Hall W. D. , 2008: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part II: Implementation of a new snow parameterization. Mon. Wea. Rev., 136, 50955115, doi:10.1175/2008MWR2387.1.

    • Search Google Scholar
    • Export Citation
  • Yang, Y., , Chen X. , , and Qi Y. , 2013: Classification of convective/stratiform echoes in radar reflectivity observations using a fuzzy logic algorithm. J. Geophys. Res. Atmos., 118, 18961905, doi:10.1002/jgrd.50214.

    • Search Google Scholar
    • Export Citation
  • Yoneyama, K., , Zhang C. , , and Long C. N. , 2013: Tracking pulses of the Madden–Julian oscillation. Bull. Amer. Meteor. Soc., 94, 18711891, doi:10.1175/BAMS-D-12-00157.1.

    • Search Google Scholar
    • Export Citation
  • Yuter, S. E., , and Houze R. A. Jr., 1997: Measurements of raindrop size distributions over the Pacific warm pool and implications for Z–R relations. J. Appl. Meteor., 36, 847867, doi:10.1175/1520-0450(1997)036<0847:MORSDO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Zuluaga, M. D., , and Houze R. A. Jr., 2013: Evolution of the population of precipitating convective systems over the equatorial Indian Ocean in active phases of the Madden–Julian oscillation. J. Atmos. Sci., 70, 27132725, doi:10.1175/JAS-D-12-0311.1.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Fraction of total rainfall observed by S-PolKa during DYNAMO that was classified as convective or stratiform using the SHY95 algorithm.

  • View in gallery

    (a) Sample RHI radar cross section of S-band reflectivity (dBZ) at 2053 UTC 21 Oct 2011. (b) As in (a), but at 2253 UTC 21 Oct 2011.

  • View in gallery

    (a) S-band reflectivity (dBZ) and (b) convective/stratiform classifications using SHY95 from the interpolated S-PolKa dataset at an altitude of 2.5 km at 0131 UTC 1 Jan 2012. In (b), red (purple) denotes stratiform (convective). The green line in (b) denotes the cross section, which runs from north (point A) to south (point B), of reflectivity seen in (e). (c) S-band reflectivity (dBZ) along the 0.5° elevation scan and (d) rain-type classifications using the new algorithm. Red, purple, green, blue, light blue, and pink respectively represent stratiform, convective, uncertain, isolated convective core, isolated convective fringe, and weak echo. (e) Cross section of S-band reflectivity (dBZ) through the green line seen in (b).

  • View in gallery

    As in Figs. 3c and 3d, but at 0231 UTC 16 Oct 2011.

  • View in gallery

    Decision tree diagram illustrating the steps of the new rain-type classification algorithm. Text inside rectangles depicts decision-making steps, and ovals represent final classifications.

  • View in gallery

    Mean profiles of (a) vertical motion and (b) latent heating in stratiform (red), convective (purple), uncertain (green), isolated convective core (blue), and isolated convective fringe (light blue) echo as simulated in the WRF Model using the new rain-type classification algorithm. Each profile is normalized by dividing the entire profile by the absolute value of the largest magnitude of vertical motion in the profile. (c),(d) As in (a),(b), but for stratiform and convective elements when using the SHY95 algorithm.

  • View in gallery

    As in Fig. 5a, but for Rconv = 5 km.

  • View in gallery

    Stacked bar chart showing the cumulative daily-averaged precipitation amounts (mm) during DYNAMO classified as convective (purple), stratiform (red), uncertain (green), isolated convective core (blue), and isolated convective fringe (light blue). Rainfall estimates are made using reflectivity along the 0.5° elevation angle in the eastern half of the radar domain as described in the text.

  • View in gallery

    Fraction of total rainfall observed by S-PolKa during DYNAMO that was classified as convective, stratiform, or uncertain using the updated algorithm and reflectivity along the 0.5° elevation angle. Isolated convective categories are included in the convective rainfall fraction.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 113 113 40
PDF Downloads 80 80 30

Rainfall-Type Categorization of Radar Echoes Using Polar Coordinate Reflectivity Data

View More View Less
  • 1 Department of Atmospheric Sciences, University of Washington, Seattle, Washington
© Get Permissions
Full access

Abstract

An algorithm used to classify precipitation echoes by rain type without interpolating radar data to a constant height is detailed. The method uses reflectivity data without clutter along the lowest available scan angle so that the classifications yield a more accurate representation of the rain type observed at the surface. The algorithm is based on that of Steiner et al. but is executed within a polar coordinate system. An additional procedure allows for more small, isolated, and/or weak echo objects to be appropriately identified as convective. Echoes in the immediate vicinity of convective cores are included in a new transition category, which consists mostly of echoes for which a convective or stratiform determination cannot be confidently made. The new algorithm more effectively identifies shallow convection embedded within large stratiform regions, correctly identifies isolated shallow and weak convection as such, and more often appropriately identifies periods during which no stratiform precipitation is present.

Publisher’s Note: This article was revised on 30 November 2016 to include the following online locations where the code described in this article is now publicly available: Versions for Cartesian and polar coordinate grids, respectively, are found at http://www.github.com/swpowell/raintype_python and http://www.github.com/swpowell/raintype_python_polar.

Corresponding author address: Scott W. Powell, Department of Atmospheric Sciences, University of Washington, Box 351640, Seattle, WA 98195. E-mail: spowell@atmos.uw.edu

Abstract

An algorithm used to classify precipitation echoes by rain type without interpolating radar data to a constant height is detailed. The method uses reflectivity data without clutter along the lowest available scan angle so that the classifications yield a more accurate representation of the rain type observed at the surface. The algorithm is based on that of Steiner et al. but is executed within a polar coordinate system. An additional procedure allows for more small, isolated, and/or weak echo objects to be appropriately identified as convective. Echoes in the immediate vicinity of convective cores are included in a new transition category, which consists mostly of echoes for which a convective or stratiform determination cannot be confidently made. The new algorithm more effectively identifies shallow convection embedded within large stratiform regions, correctly identifies isolated shallow and weak convection as such, and more often appropriately identifies periods during which no stratiform precipitation is present.

Publisher’s Note: This article was revised on 30 November 2016 to include the following online locations where the code described in this article is now publicly available: Versions for Cartesian and polar coordinate grids, respectively, are found at http://www.github.com/swpowell/raintype_python and http://www.github.com/swpowell/raintype_python_polar.

Corresponding author address: Scott W. Powell, Department of Atmospheric Sciences, University of Washington, Box 351640, Seattle, WA 98195. E-mail: spowell@atmos.uw.edu

1. Introduction

Atmospheric precipitation falls from a spectrum of cloud types, and radar is commonly used to characterize the nature of the precipitating clouds. Precipitation echoes seen on radar are broadly categorized as convective and stratiform. These categories imply the nature of the vertical air motions producing the precipitation. Convective echoes are produced by localized intense updrafts capable of advecting precipitation particles upward, while stratiform echoes arise from areas where the air motions are generally weaker and precipitating particles drift downward from aloft. In cases where vertical velocity is known, such as in an atmospheric modeling framework, the profiles of vertical velocity are the direct way of separating model grid points into “convective” and “stratiform” elements. Observations of vertical air motions, however, are rare, whereas radar reflectivity data are commonplace, and texture analysis applied to radar observations can serve as an alternative way to separate rain areas into convective and stratiform components. Such analysis makes use of the fact that regions of convective echoes have a heterogeneous character in the horizontal with maxima in the form of vertically oriented cores, while stratiform echoes are more horizontally homogenous and in the vertical often have a bright band in a horizontally extensive melting layer. A technique based on criteria of these horizontal and vertical structural characteristics has evolved over the past four decades (Houze 1973; Churchill and Houze 1984; Steiner and Houze 1993; Steiner et al. 1995, hereafter SHY95; Yuter and Houze 1997; Awaka et al. 1997; Biggerstaff and Listemaa 2000). A version of the technique is used as part of the algorithm set routinely applied to satellite radar data (Awaka et al. 1997). However, as pointed out by Biggerstaff and Listemaa (2000), the method has a shortcoming in that the precise boundary between convective and stratiform echoes is often ambiguous. Another shortcoming that is encountered especially over tropical oceans is the difficulty of distinguishing shallow, isolated, weakly raining convective elements from fragments of stratiform echo. Schumacher and Houze (2003) addressed this problem for the rain-type classification of shallow, isolated rain using data from the Tropical Rainfall Measuring Mission (TRMM) precipitation radar, and the problem was rectified in the version 6 release of the 2A23 product (Awaka et al. 1997). However, the methodology applied to the downward-looking satellite data cannot be effectively applied to the data of a conically scanning earthbound radar.

The problem in identifying shallow convective elements is especially important in the tropics because shallow convective radar echoes are an important part of the tropical oceanic cloud population. Shallow convective clouds, both nonprecipitating and precipitating, are numerous over the low-latitude oceans, where they are critical in the transition from shallow to deep convection regimes. They act as a source of lower-tropospheric heating and through them moisture can be deposited at higher levels than might be achieved through large-scale advection alone. Shallow cumulus clouds are often the predominant cloud type present prior to development of larger and deeper convective echoes and broad stratiform regions, and such an evolution of the cloud type is typical during transition from a shallow to a deep convective regime (Mapes et al. 2006; Zuluaga and Houze 2013; Barnes and Houze 2013). Shallow convection is present not only during extremely suppressed periods as isolated objects but during highly active periods, both as isolated convective entities and shallow convective cells embedded within much larger stratiform regions. For tropical studies it is therefore vital to be able to identify the shallow, isolated convective echoes.

Types of algorithms other than ones based on texture analysis of the reflectivity field exist for the purpose of separating convective and stratiform components of precipitation. Each possesses its own strengths and weaknesses. Penide et al. (2013) compare a texture-based SHY95-like method to a method introduced by Bringi et al. (2009) that separated radar data into classifications based on the estimated drop size distribution (DSD) derived from polarimetric data (particularly ZDR and KDP) at Darwin, Australia. The benefit of the DSD-based algorithm is that it introduces a new “mixed” precipitation category that includes echoes that contain both convective and stratiform characteristics. Its separation between convective and stratiform is made by a line in the log10(Nw)–D0 space (see Bringi et al. 2009), in which Nw is the normalized number concentration and D0 is a median volume diameter. Roberto et al. (2016) expand upon the Bringi et al. (2009) approach by using a line in the ZhZDR space. However, such lines of separation may vary significantly depending on the environmental regime (Thompson et al. 2015). Furthermore, strong stratiform echoes might have similar DSDs to weak convective echoes; stratiform echoes in excess of 40 dBZ were occasionally observed in the field during the Dynamics of the Madden–Julian Oscillation (DYNAMO). Statistical methods can also be employed to make rain-type classifications. For example, Yang et al. (2013) used a fuzzy logic algorithm to classify convective and stratiform components. The algorithm is more objective than traditional texture-based algorithms and benefits from providing a measure of the confidence in its classifications, but so far it only allows for two classification categories. Anagnostou (2004) trains a neural network to identify convective and stratiform components observed by a scanning precipitation radar. The network is trained using reflectivity observations from TRMM and the 2A23 convective/stratiform classifications. Because TRMM views precipitation in the vertical, it can easily resolve bright bands and classify precipitation with bright bands as stratiform. The neural network learns the typical reflectivity signal in columns below bright bands and uses this information to make an informed classification using low-elevation scanning radar data. However, in the absence of a clear bright band, the TRMM 2A23 algorithm simply reverts to a texture-based algorithm to make its classification (Awaka et al. 1997). A major limitation of the SHY95 algorithm that was noted by several of the above-mentioned papers was the excessive classification of echoes as convective. We address this problem in section 5.

The objective of this paper is to introduce and detail a rain-type classification algorithm based on texture analysis of the reflectivity field that is an improvement in two respects: This new procedure can more accurately identify shallow convective elements, and, in addition, it can more definitively separate the echoes that are ambiguous as to whether they are convective or stratiform. To demonstrate these improvements, we use an extensive oceanic tropical radar dataset collected during the DYNAMO (Yoneyama et al. 2013) field campaign. From 1 October 2011 through 15 January 2012, the NCAR S-/Ka-band dual-polarimetric radar system (S-PolKa) was stationed on an island in the central equatorial Indian Ocean. Specifically the radar was in Addu City, Maldives (0.63°S, 73.10°E), at 10-m elevation. It nearly continuously sampled oceanic hydrometeor and nonmeteorological echoes during that period. Zuluaga and Houze (2013) have described the scanning strategy, which consisted of surveillance scans at several elevation angles and range–height indicator (RHI) scans, which yielded radar data with higher vertical resolution for parts of the radar domain not affected by beam blockage. Rain-type maps consisting of convective and stratiform echo classifications based on the algorithm of SHY95 applied to the most recent quality-controlled dataset have been publicly released. Powell and Houze (2013) discuss rainfall observed during DYNAMO in detail. Their Fig. 2 is a time series of convective, stratiform, and total radar-estimated rainfall during the field campaign. Figure 1 in this paper is derived from that time series and shows the percentage of total rainfall attributed to convective and stratiform elements in parts of the S-PolKa domain not affected by beam blockage and uses the SHY95 algorithm to determine rain type. Of particular interest are the relative rainfall amounts during January 2012. Although little rain fell, most of the rainfall was classified as stratiform. However, conditions were nearly clear during that period, and the predominant cloud type during the first half of January was shallow cumulus (Powell and Houze 2013). This highlights one limitation of the SHY95 algorithm during convectively suppressed periods. We will use the S-PolKa reflectivities during DYNAMO as the dataset on which we initially test and evaluate the new algorithm, which will identify shallow, weak convection accurately and isolate those echoes that are most ambiguous as to their convective or stratiform character.

Fig. 1.
Fig. 1.

Fraction of total rainfall observed by S-PolKa during DYNAMO that was classified as convective or stratiform using the SHY95 algorithm.

Citation: Journal of Atmospheric and Oceanic Technology 33, 3; 10.1175/JTECH-D-15-0135.1

2. Ambiguities and difficulties in classifying convective and stratiform echoes

In many cases, identification of radar echoes as convective or stratiform is obvious. For example, cumulonimbi containing intense updrafts have large hydrometeors that yield high values of reflectivity. Some stratiform regions can often be easily identified by their robust bright bands and weak echoes at lower altitudes. However the identification of some echoes are more ambiguous. For instance, a transition zone between convective cores and stratiform regions of a squall-line system is sometimes characterized by a deep layer of subsidence and a low-level reflectivity minimum (Biggerstaff and Houze 1993). Biggerstaff and Listemaa (2000) suggest that such echoes are more appropriately classified as convective because the rain rate in transition regions is high despite a decrease in reflectivity (Atlas et al. 1999).

The SHY95 algorithm and similar methods often classify the fringes of precipitation echo associated with isolated, and often shallow, cumulonimbi as stratiform because reflectivity there is low. However, hydrometeors in such regions do not necessarily grow within an upper-tropospheric mesoscale updraft via aggregation before falling through the 0°C level, melting, and continuing downward through a mesoscale downdraft. Rather, they are smaller hydrometeors generated within a convective core that fall out a short lateral distance away from an updraft. Thus, the microphysical growth processes associated with such hydrometeors are distinctively and exclusively convective in nature. A third area of ambiguity occurs in regions of heavy stratiform precipitation. Biggerstaff and Listemaa (2000) argue that SHY95 tends to incorrectly classify heavily precipitating stratiform as convective; however, they do not attempt to adjust a threshold reflectivity above which an echo is considered to be a convective core in their dataset. Furthermore, many vertical cross sections of reflectivity within a broad stratiform region show a feature resembling a bright band near the 0°C level, which would indicate stratiform kinematics. However, they also show a secondary maximum of reflectivity in the lowest 1–2 km, which could indicate that shallow embedded convection is present below the base of the stratiform cloud deck.

For this paper, we follow SHY95 by using reflectivity to classify echoes as convective if the primary microphysical growth processes of the precipitation associated with the echoes likely were associated with an active convective updraft capable of advecting precipitation-sized particles upward. By this definition, stratiform echo consists of hydrometeors that exited a convective core and grew via deposition, aggregation, and possibly some riming as they drifted down to the 0°C level (Rutledge and Houze 1987; Braun and Houze 1994). Echo columns occurring close to convective cores are difficult to separate into convective and stratiform entities using reflectivity only because they may contain characteristics of both. Our updated algorithm will categorize separately echoes whose most likely appropriate classification is highly uncertain.

3. The rain-type classification of SHY95

Convective–stratiform classification of radar echoes has evolved, and the version of the scheme frequently referred to as SHY95 is just a step in that evolution. SHY95 was an improvement of Steiner and Houze (1993), which was an update of the algorithm introduced by Churchill and Houze (1984), which in turn was based on a method that Houze (1973) applied to high-resolution rain gauge data. Churchill and Houze (1984) were the first to apply the method to fields of radar reflectivity. They converted the reflectivity to rain rate based on a simple Marshall and Palmer (1948) relationship in order to work in units of rain rate. They then classified as convective the radar echo areas that exceeded some rain-rate threshold or those in which local peaks of rain rate were significantly greater than the nearby background rate. Steiner and Houze (1993) also applied the method to the reflectivity field but used thresholds of reflectivity instead of rain rate in order to avoid the large uncertainty involved in radar-estimated rainfall rates. Yuter and Houze (1997) made some further refinements to SHY95 to be able to apply the scheme to airborne radar data. Awaka et al. (1997) adapted features of the scheme to satellite radar data. The present paper is the latest step in the evolution of this methodology.

SHY95 implemented a threshold reflectivity value above which any echo is automatically considered convective and identified echoes as convective based on their reflectivity relative to the background intensity of echoes. They further tested the reflectivity-based method for consistency with dual-Doppler-derived vertical velocity data. For any echoes classified as a convective core using the threshold or the background intensity as criteria, neighboring grid points would also be classified as convective. The number of grid points classified as such was related to the intensity of the echo in the convective core—more intense convective cores would be assigned wider radii. Yuter and Houze (1997, see their appendix B) made an additional adjustment to how a single echo is compared to the background reflectivity. While their change to the scheme did not result in many new classifications compared to SHY95, they allowed the scheme more flexibility for use with a variety of datasets by introducing two additional parameters that the user can “tune” to best classify convection for a dataset of specific resolution or from a particular radar system or setup. The algorithm we refer to as SHY95 in this paper contains the adjustment by Yuter and Houze (1997). Its use is complicated by the inclusion of a number of parameters that must be tuned by the user. These parameters are adjusted for particular environmental regimes, radar platforms, dataset resolutions, and scanning strategies. Even in the same environmental regime, for two radar systems having beams of different frequencies or widths, for example, one must alter the input parameters for proper rain-type classification.

As traditionally written, the SHY95 algorithm is applied on a field of reflectivity that has been interpolated from the native polar coordinates of a scanning radar onto a Cartesian grid at a single altitude. The interpolation degrades data close to the radar and creates data to fill in gaps in the grid near the edges of the domain of the radar’s observations. For most radar platforms, the method is applied to the reflectivity field at a height between 2 and 3 km. Thus, the method does not classify echoes close to the surface, and it sometimes fails to identify shallow precipitating convection as convective because such elements generally have low reflectivities, do not stand out significantly from the background reflectivity, are too short, and/or are too narrow.

For example, Fig. 2a illustrates an RHI cross section from S-PolKa of nongridded reflectivity data at 2053 UTC 21 October 2011. The cross section is taken through a developing stratiform region that, at the time, consisted primarily of convection of various depths. After the data are interpolated, SHY95 identifies much of the convection seen in this figure; however, many of the short and narrow echoes that were obviously convective, such as the 40+ dBZ echoes near 50 km, are not classified as such. Another motivating example is seen from a nearby RHI cross section ~2 h later, as seen in Fig. 2b. A small convective element was present on the edge of a larger stratiform region, and its 40 dBZ contour only extended upward to about 2 km. SHY95 treats the echo as part of the stratiform region if interpolated data at an altitude above 2 km are used for the classification.

Fig. 2.
Fig. 2.

(a) Sample RHI radar cross section of S-band reflectivity (dBZ) at 2053 UTC 21 Oct 2011. (b) As in (a), but at 2253 UTC 21 Oct 2011.

Citation: Journal of Atmospheric and Oceanic Technology 33, 3; 10.1175/JTECH-D-15-0135.1

When many small, and frequently shallow, echo objects are present, SHY95 often classifies the edges of such echo objects as stratiform. Figures 3a and 3b show the gridded reflectivity field at 2.5 km and the corresponding convective/stratiform classification at 0131 UTC 1 January 2012. SHY95 detects many convective cores but also depicts about half of the total precipitating echo as stratiform. In reality, during the suppressed conditions prevailing at that time, no mechanism was operative that could have produced deep stratiform precipitation of the type discussed in section 2. To demonstrate, a cross section through one of the convective elements is shown in Fig. 3e. A green line indicates the location of the cross section in Fig. 3b. Convection extends vertically to 4–6 km, and no bright bands were present. SHY95 makes the stratiform classification because the reflectivities on the edges of the echo objects were far lower than the reflectivities within the convective echo cores; hence, their reflectivities were neither significantly larger than the background reflectivity nor large enough to exceed the threshold for convective classification.

Fig. 3.
Fig. 3.

(a) S-band reflectivity (dBZ) and (b) convective/stratiform classifications using SHY95 from the interpolated S-PolKa dataset at an altitude of 2.5 km at 0131 UTC 1 Jan 2012. In (b), red (purple) denotes stratiform (convective). The green line in (b) denotes the cross section, which runs from north (point A) to south (point B), of reflectivity seen in (e). (c) S-band reflectivity (dBZ) along the 0.5° elevation scan and (d) rain-type classifications using the new algorithm. Red, purple, green, blue, light blue, and pink respectively represent stratiform, convective, uncertain, isolated convective core, isolated convective fringe, and weak echo. (e) Cross section of S-band reflectivity (dBZ) through the green line seen in (b).

Citation: Journal of Atmospheric and Oceanic Technology 33, 3; 10.1175/JTECH-D-15-0135.1

4. Classification in a radar’s native spherical coordinate grid

A scanning precipitation radar has a fixed beamwidth and usually scans azimuthally at a constant elevation angle. Data are averaged and recorded at selected azimuthal intervals. The distance between adjacent range gates along any radial is also an interval of adjustable size. For example, the S-PolKa radar during DYNAMO used 1° azimuthal spacing and a beamwidth of 0.91°. Range gates were 150 m apart. Along a radial, hydrometeors or other objects within the beam volume between range gates influence reflectivity, and a single reflectivity value is recorded for each range bin and azimuth interval in the range–azimuth coordinate system centered on the radar antenna. The rain-type classification of SHY95 is applied after the data recorded at points in range and azimuth are interpolated onto a Cartesian grid. For research purposes, a horizontal grid spacing of 2 km and a vertical spacing of 500 m are often used for the Cartesian grid. However, small and shallow precipitating convective elements are often narrower than 2 km wide, and interpolating the original radar data onto the Cartesian grid omits details about such echoes close to the radar, where the native spatial resolution is high enough to detect such features. Additionally, SHY95 generally uses the reflectivity field interpolated along a constant height 2–3 km above the surface. Thus, shallow echoes with tops below this level are missed altogether.

Alternatively, the SHY95 algorithm can be run on a curvilinear grid. It can be applied to radar data in the native, polar coordinate system on which it is collected. The radar-centered polar coordinates may be represented by azimuth angle φ, elevation angle θ, and radius r from the center. By working in these polar coordinates, we take advantage of the high spatial resolution of the radar data along a radial, and we avoid eliminating small features that are spatially resolved close to the radar, where the distance between adjacent radials is small. Instead of using data along a constant height, we use data along one sweep at a constant elevation angle and map the data onto regularly spaced (r, φ) coordinates. In particular, we use the lowest available elevation angle. This approach preserves sharp changes in reflectivity that may occur from range bin to range bin or between neighboring azimuthal bins. It is ideal for classification of the precipitation echoes close to the surface and gives us the best estimate of surface rain type that can be retrieved from a scanning radar. In addition, classification of the echoes closest to the surface is more compatible with rain-rate estimates derived from relationships between reflectivity, differential reflectivity (ZDR), and/or specific differential phase (KDP). Such relationships are often derived from disdrometer data, if they exist, at the surface or radar data along the lowest available tilt angle (e.g., Ryzhkov et al. 2005). As such, radar estimates of rainfall using such relationships are best computed using the lowest elevation angle not under the strong influence of clutter; therefore, a rain-type classification for the same echoes is preferable. Clutter in the dataset we used for testing the algorithm was present mostly near the radar site, and a clutter removal algorithm (Steiner and Smith 2002; Hubbert et al. 2009a,b) was executed on the reflectivity field before rain-type classifications were made. One limitation of our approach is that more shallow echoes will be omitted as one uses data farther from the radar site. Even at low θ = 0.5° elevation angles, the center of a radar beam 150 km from the radar is about 2 km above the surface. However, we gain the advantage of not losing valuable information close to the radar.

Our method of classification treats a dataset collected on a conical surface by a full radar sweep (on a single plane in spherical space) as if it were obtained on a single plane at constant altitude. Local to a single data point, such treatment is approximately valid. The relative altitude at which an echo is observed is only important when considering the background reflectivity as in SHY95. The radius of influence for any grid point, or the distance at which neighboring reflectivities are considered in determining the background reflectivity, is typically about 5–10 km. A typical base elevation angle for scanning precipitation radar is 0.5°. Within 150 km of a radar for such an elevation angle, and depending on the distance of a beam from the radar, a typical increase in altitude along a radial over a horizontal distance of 10 km is between 100 and 250 m after accounting for upward refraction of the beam by water vapor (Liebe 1985). Thus, the variation in the altitudes along such a radial of data points used to determine the background reflectivity is less than the typical vertical resolution (500 m) of gridded, interpolated datasets.

In view of the above-mentioned information, a polar coordinate–based algorithm can run much like the classification scheme of SHY95 but with better spatial resolution. Table 1 lists and describes all of the parameters in the polar coordinate algorithm that can be tuned by a user, including several that will be described in section 5. As in SHY95, a threshold reflectivity Zth above which all echoes are considered convective, is applied. In convective elements, reflectivity often decreases with height, which might necessitate that Zth decrease with increasing range. We have implemented a variety of Zth values close to the one currently used that are a function of range (not shown), and none significantly changed the classifications. Therefore, a constant Zth is sufficient for the dataset used here. Echoes that are sufficiently higher than background reflectivity are also classified as convective following Yuter and Houze (1997). The background reflectivity is simply the mean equivalent reflectivity (converted to dBZ) of all echoes within a radius of influence Rbg. Distances between radar data points can be found by remapping the data onto a rectilinear grid. Then we use the Pythagorean distance formula to find all points within Rbg of the data point under evaluation. Differences in altitude across the grid are ignored because they are negligible in the distance calculation over small lengths on the order of 10 km. Because the original grid is curvilinear, only one such remapping is required for each ring of radar data. For a grid point to be considered convective, its reflectivity (dBZ), if not equal to or greater than Zth, must exceed the background reflectivity Zbg by Zcc, such that
e1
in which a and b are user-defined parameters that require adjustment based on the spatial resolution of the radar data (Yuter and Houze 1997). SHY95 also classified echoes immediately surrounding those identified as convective cores as convective based on the intensity of the convective core echo relative to its environment. The maximum distance from convective cores within which such echoes could be classified as convective was Rconv. Such echoes could only be classified as convective if they were within a radius Radj of the convective core; the value of Radj was determined based on the reflectivity of the convective core and could not exceed Rconv. However, as we will show in section 6, such echoes often contain some stratiform heating characteristics. We will instead classify echoes surrounding convective cores as “uncertain” regions. Some such echoes may include convective or stratiform echoes as well as transition regions as described in section 1. The radius Radj is small for echoes in regions of weaker background reflectivity to minimize erroneous classification of surrounding weak or stratiform echo, such that
e2
and Zconv is the minimum reflectivity a convective core echo may have so that all surrounding echoes within a radius Rconv are classified as uncertain. In Table 1, we also include the values for each parameter used in this article. Again, the values used in this article are not intended for universal use across different datasets. Input must be tuned appropriately before using the algorithm.
Table 1.

List of input parameters in algorithm that can be adjusted by the user. Descriptions of each parameter are provided. Values for each parameter used in this study are listed. They are not intended for universal use across different datasets. Input should be tuned appropriately for a user’s specific environment and radar platform before using the algorithm.

Table 1.

5. A simple method for identifying shallow, isolated, and weak convection

The resolution of S-PolKa data along a radial is fine enough to resolve most small precipitating elements, and the algorithm described in section 4 effectively detects fairly intense shallow convection—particularly that embedded within stratiform regions—because it uses the lowest sweep of data available instead of data from a fixed height. However, weakly reflective shallow and isolated convective echo objects, which may consist of relatively small hydrometeors, do not produce a reflectivity that exceeds Zth nor do large parts of many such echo objects sufficiently exceed the background reflectivity. As discussed above, SHY95, which runs on an interpolated grid, is not well suited to simultaneously identify shallow and deep convection. Generally, wider convective echo objects are associated with stronger or more numerous updrafts, either of which cause higher reflectivity by means of suspending larger or more numerous hydrometeors. Small, isolated echo objects tend to be associated with young convection, convection in unfavorable environments, or convection with weak updrafts; their maximum reflectivities and echo/cloud-top heights are, on average, observed to be less than those of wider and sometimes isolated but more robust convective entities, such as cumulus congestus clouds (i.e., Fig. 8 in Hagos et al. 2014). As suggested by Schumacher and Houze (2003), version 7 of the TRMM 2A23 algorithm (Awaka et al. 2009) now classifies small echo objects as convective; echo objects that consist of only one or two contiguous pixels are classified as convective based on the premise that isolated entities are usually shallow convection.

For echo elements with small areal coverage, we adjust the minimum reflectivity threshold required for convective classification to some value less than or equal to Zth. The new minimum threshold Znewth is a function of the areal coverage of an echo object. Here, we define a two-dimensional echo object as any contiguous area of reflectivity on the (r, φ) plane in exceedance of a minimum echo threshold, Zweak, which represents the minimum reflectivity value an echo must have to be classified by our algorithm. Term Znewth then is defined as not exceeding Zth and has the property
e3
where A is the area of an echo object, Alow is a user-specified minimum area that echo objects must achieve to contain a convective core, Amed is the user-specified area below which sufficiently large echo objects are assigned a convective threshold of Zshallow, and Ahigh is a user-specified area above which all echoes are assigned a convective threshold equal to Zth. Any echoes that exceed a reflectivity Znewth (or Zth) are first assigned to the “isolated convective core” (or convective) category; the grouping contains mostly echoes that are at or near the center of small echo objects. In other words, an echo must have a minimum reflectivity of Zshallow to be classified as an isolated convective core. The value of Znewth varies between Zshallow and Zth based on the areal coverage of the parent echo object: Echo objects with areal coverages between Amed and Ahigh are assigned a Znewth that is linearly interpolated between Zshallow and Zth based on the difference between that object’s area and Amed. Echoes that are part of an echo object area with an area between Alow and Ahigh and with reflectivity less than Znewth are classified as “isolated convective fringe.” Per the definition in (3), isolated convective fringe contains mostly echoes surrounding convective cores in shallow, isolated convection, but some decaying convection no longer associated with an active convective core is also included in the category. Any echo object with areal coverage less than Alow is considered weak echo, a category that also contains, for example, biological echoes or Bragg scatter echoes that are not meaningful in studies of precipitating clouds. Table 2 contains a summary of the six different rain-type classification categories discussed above. The portion of the algorithm described in this paragraph can be executed on data in any grid as long as the user knows the area covered by each grid cell.
Table 2.

List of echo classifications made by the updated algorithm.

Table 2.

Figures 3c and 3d show the 0.5° reflectivity and rain-type classification for the same echoes seen in Figs. 3a and 3b using the new algorithm. Only the eastern half of the domain is displayed because the low-elevation radar beam was blocked by vegetation and man-made structures west of the radar. Note that we switch from a Cartesian grid in Figs. 3a and 3b to a polar grid in Figs. 3c and 3d. All of the echoes are now classified, probably more accurately, as either weak echo (pink) or isolated convective core (blue)/fringe (light blue). Several instances of such improvement occur in particular during convectively suppressed periods or a few days prior to the development of large stratiform regions associated with a large-scale convective event of the Madden–Julian oscillation. Such are times when weak, isolated convective elements are most likely to be the most common echo object present (Powell and Houze 2013).

Another example of the new classifications is seen in Fig. 4. The cumulonimbi were observed around 0231 UTC 16 October 2011 (Fig. 4a). At that time, some isolated convection and deep and wide convection were present, and stratiform regions of precipitation were beginning to develop within the radar domain. As viewed in Fig. 4b, weak echoes around the isolated convective cores are classified as isolated convective fringe, and large areas around echoes classified as convective (purple) are considered to have uncertain classifications (green). In this case, convective cores are generally located near other convective cores such that uncertain areas are regions within a larger stratiform region (red) in which convective cores are somewhere located. Using a gridded dataset, dividing the uncertain regions more finely into convective and stratiform areas might be possible. Biggerstaff and Listemaa (2000) attempted to do so in a Cartesian framework by using the vertical gradient in reflectivity to detect bright bands in columns of data. However, this approach is not practical when using a polar coordinate–based dataset whose data are not stacked vertically, and it requires a volume scan consisting of several closely spaced elevation angles to be effective even on an interpolated dataset.

Fig. 4.
Fig. 4.

As in Figs. 3c and 3d, but at 0231 UTC 16 Oct 2011.

Citation: Journal of Atmospheric and Oceanic Technology 33, 3; 10.1175/JTECH-D-15-0135.1

Figure 5 provides a visual representation of the algorithm described in the previous two sections. Rectangles represent steps in the algorithm, and ovular shapes depict categorizations of echoes.

Fig. 5.
Fig. 5.

Decision tree diagram illustrating the steps of the new rain-type classification algorithm. Text inside rectangles depicts decision-making steps, and ovals represent final classifications.

Citation: Journal of Atmospheric and Oceanic Technology 33, 3; 10.1175/JTECH-D-15-0135.1

6. Evaluation of the algorithm using WRF

In consideration of the definitions of convective, stratiform, and transition precipitation/precipitation of uncertain type, and their related subcategories discussed above, we evaluate the accuracy of our classifications using a regional cloud-resolving model to simulate a cloud population on which we can test the algorithm. We use version 3.5.1 of the Weather Research and Forecasting (WRF; Skamarock et al. 2008) Model with a domain located over the Indian Ocean. The domain was centered at 0°, 73.15°E and was 3280 km long zonally by 2240 km wide. The Mellor–Yamada–Janjić (MYJ) planetary boundary layer scheme (Janjić 1994), Rapid Radiative Transfer Model for GCMs (RRTMG) radiation physics (Iacono et al. 2008), unified Noah land surface physics (Ek et al. 2003), and Thompson microphysics (Thompson et al. 2008) were used, and cumuli were explicitly resolved. The model resolution was 2 km, and 38 vertical pressure levels were used with a model top at 50 hPa. The simulation period was 1–20 October 2011.

The run simulated a convectively suppressed period over the central Indian Ocean and a buildup of convection into a large-scale convective event of the Madden–Julian oscillation (Powell 2016, manuscript submitted to J. Atmos. Sci.). Reflectivity output from the model is computed during model integration based on hydrometeor concentrations and assumed size distributions output by the microphysics scheme. We have run the new algorithm on the simulated reflectivity field. Because S-band radiation is not heavily attenuated by water vapor or liquid water, the simulated reflectivities (also at S band) should be similar to reflectivities obtained from greater distances along a radial that starts at a point near the surface.

The advantage of using a model framework to test the algorithm is that the model provides profiles of vertical motion and latent heating, whereas these variables are difficult to obtain using contemporary observational techniques. We can thus verify whether the columns classified as convective or stratiform based on the simulated reflectivity field are correctly classified. This test is analogous to the test performed with high-resolution dual-Doppler radar observations in SHY95. However, dual-Doppler observations were not available in DYNAMO. The model output grid is Cartesian, like an interpolated dataset of radar reflectivity. This is not problematic because the portion of the algorithm described in section 4 is essentially that of SHY95, which was originally run on a Cartesian grid, and the algorithm in section 5 can be implemented on any type of grid.

Figures 6a and 6b show the mean profiles of vertical motion and latent heating in columns classified as stratiform (red), convective (purple), uncertain (green), isolated convective core (blue), and isolated convective fringe (light blue). Each profile is divided by the absolute value of the maximum magnitude of vertical motion or latent heating in the profile, so that the profiles shown have magnitudes between −1 and 1 at all pressure levels. In general, vertical motions have the same sign and relative magnitudes as latent heating. Stratiform regions have a maximum in upward motion near 300 hPa and a maximum in downward motion near 600 hPa. The profile indicates near-zero or slightly positive vertical motion and a small peak in stratiform heating below 800 hPa. This possibly occurred because shallow convection was frequently produced by the model below 2.5 km, and weak convection embedded within large stratiform regions will often be classified by our algorithm as stratiform because the echoes above the shallow convection are weak, horizontally uniform, and consistent with stratiform precipitation echoes. The stratiform category may also erroneously include a small number of echoes that are structurally and microphysically part of isolated convective entities but are still connected to larger systems via contiguous precipitation echoes.

Fig. 6.
Fig. 6.

Mean profiles of (a) vertical motion and (b) latent heating in stratiform (red), convective (purple), uncertain (green), isolated convective core (blue), and isolated convective fringe (light blue) echo as simulated in the WRF Model using the new rain-type classification algorithm. Each profile is normalized by dividing the entire profile by the absolute value of the largest magnitude of vertical motion in the profile. (c),(d) As in (a),(b), but for stratiform and convective elements when using the SHY95 algorithm.

Citation: Journal of Atmospheric and Oceanic Technology 33, 3; 10.1175/JTECH-D-15-0135.1

Convective echoes consist of a deep layer of upward motion, with a maximum between 500 and 600 hPa. Downward motion is seen below 800 hPa, and it may be related to downdrafts occurring in the most heavily precipitating convective cores. Low-level upward motion is more likely to occur outside of the convective precipitation cores in transition regions. Uncertain regions, which sometimes represent transition regions of the type described by Biggerstaff and Houze (1993), demonstrate upward motion throughout the column on average but with two maxima. One maximum is located between 700 and 800 hPa, and is likely related to updrafts into nearby deep convective cores. A minimum occurs around 550 hPa, and the other maximum occurs near 300 hPa, where the maximum in stratiform upward motion was located. Thus, the uncertain category contains some echoes that possess heating and vertical motion profiles consistent with convective elements and others with stratiform regions. Isolated convective cores associated with horizontally smaller echo objects have profiles that are distinctly different from those of stronger convective cores. Upward motion and heating extend up to 400–500 hPa with maxima near 700 hPa, indicating that, on average, the echoes classified as isolated convection are, as we expect, usually shallower than more intense convective cores. Profiles for isolated convective fringe are similar to those for isolated convective cores, but with maxima near 800 hPa and downward motion/cooling in clear air above 700 hPa. Most of these echoes are therefore probably echoes surrounding those classified as isolated convective cores, and the two categories can probably be combined in cases for which large amounts of decaying convection are not obviously present.

Figures 6c and 6d depict the profiles of vertical motion and heating in model columns classified as stratiform (purple) or convective (red) using the SHY95 algorithm and the same input parameters as in Table 1 (except Rconv = 5 km). Convective vertical motion and heating are similar to that determined using the updated algorithm, with a maximum in upward motion and heating between 500 and 600 hPa. Most strikingly, the stratiform category contains heating and upward motion that peaks near 300 hPa, but it also contains a secondary peak—with nearly the same magnitude—around 800 hPa, not far from the level of peak vertical motion and heating in isolated convection as classified by the updated algorithm. The low-level stratiform heating as depicted by the model when using the SHY95 algorithm is present because SHY95 erroneous classifies most of isolated echoes as stratiform (e.g., Fig. 3). Clearly, the inclusion of additional rain-type categories for isolated convection vastly improves the classification.

Above when using the updated algorithm, we used the same value for Rconv as seen in Table 1. In other words, any echo within 10 km of a convective core that was not itself a convective core was placed into the uncertain category. Ideally, our goal should be to minimize the areal coverage of echo classified as uncertain. We repeated the above-mentioned procedure, instead setting Rconv = 5 km, the value used originally by SHY95. The resulting profiles of vertical velocity are shown in Fig. 7. Two small, but important, changes are noted. First, a maximum in normalized upward motion of almost 0.2 is seen in stratiform regions near 850 hPa. In other words, the mean vertical motion at 850 hPa is 20% of the mean vertical motion at the level (300 hPa) at which stratiform upward motion is maximum. Second, the normalized vertical velocity at 300 hPa in uncertain regions drops from over 0.9 to slightly above 0.7. These results are not surprising. When Rconv is reduced, the frequency of stratiform echo being placed into the uncertain category decreases, and the uncertain category assumes a slightly more convective profile. However, more convective echo is also allowed into the stratiform category, and so we see appreciable low-level vertical motion. In other words, setting Rconv to 5 km probably overstates our confidence in how close to a convective core we can differentiate between the convective and stratiform entities. A value of Rconv = 10 km minimizes the low-level stratiform heating and upward motion, while larger values (not shown) have little additional effect other than making the profile of vertical motion in uncertain regions more like what we expect in stratiform regions.

Fig. 7.
Fig. 7.

As in Fig. 5a, but for Rconv = 5 km.

Citation: Journal of Atmospheric and Oceanic Technology 33, 3; 10.1175/JTECH-D-15-0135.1

Although our classification improves upon the SHY95 rain type, some problems remain. Because distributions of vertical motion or latent heating between all categories overlap, it is therefore probably not a good idea to use any rain-type classification algorithm, including this one, to confidently classify any particular single echo. Instead, one can apply the algorithm to large datasets and determine differences between the various categories of convection on average. We have shown here that, on average, significant differences in vertical motion and latent heating do exist for the categories used in our new classification algorithm. Therefore, we advocate inclusion of the new features for identifying weak echo, isolated convection, and uncertain echoes in addition to the basic convective and stratiform categories.

7. Rainfall estimates using the updated algorithm

Rainfall estimates are obtained using a hybrid rain-rate algorithm (http://www.eol.ucar.edu/projects/dynamo/spol/parameters/rain_rate/rain_rates.html) with data from the 0.5° elevation scans. The estimated daily-averaged rain rates for each category in the eastern half of the S-PolKa domain are shown in Fig. 8. The stacked bar chart shows the relevant contributions of each category to the total precipitation estimated for each day. The color scheme is consistent with that displayed in Figs. 3, 4, and 68. Most of the precipitation that occurred during convectively suppressed periods, such as in early October and early November, fell within isolated convective entities. On such days, the total rainfall amount was usually 10 mm or less. During more convectively active days, the contribution of convective and stratiform precipitation was frequently at least 10–20 mm. The contribution of rainfall with uncertain classification during such periods often exceeds the contribution of precipitation more confidently classified as convective or stratiform. While much of the uncertain precipitation could be classified into convective or stratiform categories if the associated profile of vertical motion or diabatic heating were known, our results highlight the fact that the use of the reflectivity field alone requires that a large fraction of the total precipitation on convectively active days cannot be classified into either rain-type category.

Fig. 8.
Fig. 8.

Stacked bar chart showing the cumulative daily-averaged precipitation amounts (mm) during DYNAMO classified as convective (purple), stratiform (red), uncertain (green), isolated convective core (blue), and isolated convective fringe (light blue). Rainfall estimates are made using reflectivity along the 0.5° elevation angle in the eastern half of the radar domain as described in the text.

Citation: Journal of Atmospheric and Oceanic Technology 33, 3; 10.1175/JTECH-D-15-0135.1

Figure 9, like Fig. 1, shows a time series of the relative percentages of precipitation classified as convective or stratiform, but in this instance the results are based on the algorithm detailed in this paper. In this figure, convective precipitation includes contributions from the isolated convective core and fringe categories. In Fig. 1, at least 10% of precipitation was attributed to stratiform elements at all times—even when stratiform was obviously not present. The relative percentage of stratiform depicted in Fig. 9 is usually near 0% for very suppressed days during DYNAMO (first halves of October and November, and late December through mid-January) and the large relative amount of stratiform precipitation depicted in Fig. 1 in January is mostly eliminated. Most of the echoes occurring in January that had been classified as stratiform are now classified as isolated convective core or fringe. During convectively active periods when large stratiform regions are present (particularly the second halves of October and November and mid-December), the percentage of precipitation classified as convective, stratiform, or uncertain is typically 30%–60%, 10%–25%, and 30%–50%, respectively.

Fig. 9.
Fig. 9.

Fraction of total rainfall observed by S-PolKa during DYNAMO that was classified as convective, stratiform, or uncertain using the updated algorithm and reflectivity along the 0.5° elevation angle. Isolated convective categories are included in the convective rainfall fraction.

Citation: Journal of Atmospheric and Oceanic Technology 33, 3; 10.1175/JTECH-D-15-0135.1

8. Conclusions

We have adapted a radar-based classification of convective and stratiform rain type from Steiner et al. (1995) and Yuter and Houze (1997) for use in the polar coordinate system in which radar data are naturally obtained. By doing so, we do not require interpolation of radar data, and we can take advantage of the accuracy and high resolution of the reflectivity as it is measured along a radial beam. By using data along beams at a low-elevation angle, we obtain a more accurate and detailed classification of the rain type likely observed at the surface beneath an echo object. Importantly, we are able to more effectively detect particularly shallow convective elements embedded within large stratiform regions. However, because our algorithm uses data along the lowest available scan angle, it will detect more of the shallowest convection closest to the radar site and less near the edges of a radar domain. The severity of such bias depends on the value of the lowest elevation angle available. In any case, one should use caution when considering statistics of the isolated convective echoes over an entire radar domain.

In addition, we have created a simple, yet effective, way of classifying echo objects based on their apparent lateral size, such that smaller echo objects are more likely to have convective origins even if they have fairly low reflectivities. A shallow, isolated convection algorithm prevents the erroneous classification of convective echo as stratiform, a problem that occurs when using the SHY95 algorithm in conditions where shallow cumuliform clouds are the dominant cloud type present. Large majorities of precipitation observed during convectively suppressed periods are attributed to such weak, and often isolated, convective echoes. Additional categories further divide convection based on the implicit uncertainty of convective/stratiform classification in the area surrounding convective cores. We find that a large portion of precipitation near convective cores classified by SHY95 as either convective or stratiform should not actually be confidently included in either category. Within as much as 10 km of a convective core, echoes may take on vertical motion and latent heating characteristics of either convective or stratiform regions and sometimes both.

Using the lowest scan angle for the rain-type classification can also allow for identification of echoes much closer to the radar than SHY95 allows on a Cartesian grid at fixed height. In the absence of clutter, or after running an algorithm designed to eliminate clutter, the rain-type classification algorithm herein can be run using data at each tilt angle to provide high temporal resolution of small echo objects passing within a few kilometers of a radar. This approach may be particularly useful for identifying small convective elements that pass over a nearby vertically pointing instrument. For example, during DYNAMO, a vertically pointing Ka-band zenith radar (KAZR) was located a few kilometers southeast of S-PolKa (see Fig. 1 in Powell and Houze 2013).

Future methods to identify rain type will likely take into account polarimetric variables not used in this paper in order to make classifications of echoes based on the microphysical process(es) most likely occurring within each. A rain-type classification using only reflectivity, such as described herein, will remain important as a first guess that can be refined by multipolarimetric observations. Also, as of writing this paper, some currently used research radars and many operational radars around the world do not yet have polarization capabilities. Furthermore, the method can be used retroactively on data from nonpolarized radar, such as WSR-88D radars in the United States, prior to deployment of dual-polarization capabilities on those platforms.

Finally, we note that the methodology detailed herein is optimized for use in tropical environments where the 0°C level is well above the beam on the lowest elevation angle out to 150 km, and where stratiform precipitation is that associated with mature or developing mesoscale convective systems. One should use caution when identifying convective and stratiform elements using such a method in other environments. In wintertime frontal situations, the stratiform precipitation is produced by frontogenetic mechanisms, and the snow layer may be close to or in contact with the ground or ocean. In tropical cyclones, stratiform precipitation may be due to the secondary circulation of the vortex. Algorithms for identifying these other types of stratiform precipitation may need to emphasize different aspects of the radar observations. They are beyond the scope of this article.

Acknowledgments

This research was supported by the Department of Energy Grant DE-SC0008452 and the National Science Foundation Grant AGS-1355567. Beth Tully edited the graphics. The authors thank M. Dixon (NCAR) for helping us with viewing S-PolKa data in their native coordinates. Computing resources for running WRF were provided by the National Energy Research Scientific Computing Center (NERSC). DYNAMO radar data used in this article can be obtained online (data.eol.ucar.edu/codiac/dss/id-347.017). Contact the first author to obtain the most recent version of the code for rain-type classification.

REFERENCES

  • Anagnostou, E. N., 2004: A convective/stratiform precipitation classification algorithm for volume scanning weather radar observations. Meteor. Appl., 11, 291300, doi:10.1017/S1350482704001409.

    • Search Google Scholar
    • Export Citation
  • Atlas, D., , Ulbrich C. W. , , Marks F. D. Jr., , Amitai E. , , and Williams C. R. , 1999: Systematic variation of drop-size and radar-rainfall relations. J. Geophys. Res., 104, 61556169, doi:10.1029/1998JD200098.

    • Search Google Scholar
    • Export Citation
  • Awaka, J., , Iguchi T. , , Kumagai H. , , and Okamoto K. , 1997: Rain type classification algorithm for TRMM precipitation radar. IGARSS’97: 1997 International Geoscience and Remote Sensing Symposium, T. I. Stein, Ed., Vol. 4, IEEE, 16331635. doi:10.1109/IGARSS.1997.608993.

  • Awaka, J., , Iguchi T. , , and Okamoto K. , 2009: TRMM PR standard algorithm 2A23 and its performance on bright band detection. J. Meteor. Soc. Japan, 87A, 3152, doi:10.2151/jmsj.87A.31.

    • Search Google Scholar
    • Export Citation
  • Barnes, H. C. & , and Houze R. A. Jr. 2013: The precipitating cloud population of the Madden-Julian Oscillation over the Indian and west Pacific Oceans. J. Geophys. Res. Atmos., 118, 69967023, doi:10.1002/jgrd.50375.

    • Search Google Scholar
    • Export Citation
  • Biggerstaff, M. I., , and Houze R. A. Jr., 1993: Kinematics and microphysics of the transition zone of the 10–11 June 1985 squall line. J. Atmos. Sci., 50, 30913110, doi:10.1175/1520-0469(1993)050<3091:KAMOTT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Biggerstaff, M. I., , and Listemaa S. A. , 2000: An improved scheme for convective/stratiform echo classification using radar reflectivity. J. Appl. Meteor., 39, 21292150, doi:10.1175/1520-0450(2001)040<2129:AISFCS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Braun, S. A., , and Houze R. A. Jr., 1994: The transition zone and secondary maximum of radar reflectivity behind a midlatitude squall line: Results retrieved from Doppler radar data. J. Atmos. Sci., 51, 27332755, doi:10.1175/1520-0469(1994)051<2733:TTZASM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Bringi, V. N., , Williams C. R. , , Thurai M. , , and May P. T. , 2009: Using dual-polarized radar and dual-frequency profiler for DSD characterization: A case study from Darwin, Australia. J. Atmos. Oceanic Technol., 26, 21072122, doi:10.1175/2009JTECHA1258.1.

    • Search Google Scholar
    • Export Citation
  • Churchill, D. D., , and Houze R. A. Jr., 1984: Development and structure of winter monsoon cloud clusters on 10 December 1978. J. Atmos. Sci., 41, 933960, doi:10.1175/1520-0469(1984)041<0933:DASOWM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Ek, M. B., , Mitchell K. E. , , Lin Y. , , Rogers E. , , Grunmann P. , , Koren V. , , Gayno G. , , and Tarpley J. D. , 2003: Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta model. J. Geophys. Res., 108, 8851, doi:10.1029/2002JD003296.

    • Search Google Scholar
    • Export Citation
  • Hagos, S., , Feng Z. , , Landu K. , , and Long C. N. , 2014: Advection, moistening and shallow-to-deep convection transitions during the initiation and propagation of Madden-Julian Oscillation. J. Adv. Model. Earth Syst., 6, 938949, doi:10.1002/2014MS000335.

    • Search Google Scholar
    • Export Citation
  • Houze, R. A., Jr., 1973: A climatological study of vertical transports by cumulus-scale convection. J. Atmos. Sci., 30, 11121123, doi:10.1175/1520-0469(1973)030<1112:ACSOVT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Hubbert, J. C., , Dixon M. , , Ellis S. M. , , and Meymaris G. , 2009a: Weather radar ground clutter. Part I: Identification, modeling, and simulation. J. Atmos. Oceanic Technol., 26, 11651180, doi:10.1175/2009JTECHA1159.1.

    • Search Google Scholar
    • Export Citation
  • Hubbert, J. C., , Dixon M. , , and Ellis S. M. , 2009b: Weather radar ground clutter. Part II: Real-time identification and filtering. J. Atmos. Oceanic Technol., 26, 11811197, doi:10.1175/2009JTECHA1160.1.

    • Search Google Scholar
    • Export Citation
  • Iacono, M. J., , Delamere J. S. , , Mlawer E. J. , , Shepard M. W. , , Clough S. A. , , and Collins W. D. , 2008: Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, doi:10.1029/2008JD009944.

    • Search Google Scholar
    • Export Citation
  • Janjić, Z. I., 1994: The step-mountain eta coordinate model: Further developments of the convection, viscous layer, and turbulence closure schemes. Mon. Wea. Rev., 122, 927945, doi:10.1175/1520-0493(1994)122<0927:TSMECM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Liebe, H. J., 1985: An updated model for millimeter wave propagation in moist air. Radio Sci., 20, 10691089, doi:10.1029/RS020i005p01069.

    • Search Google Scholar
    • Export Citation
  • Mapes, B. E., , Tulich S. , , Lin J.-L. , , and Zuidema P. , 2006: The mesoscale convection life cycle: Building block or prototype for large-scale tropical waves? Dyn. Atmos. Oceans, 42, 329, doi:10.1016/j.dynatmoce.2006.03.003.

    • Search Google Scholar
    • Export Citation
  • Marshall, J. S., , and Palmer W. M. , 1948: The distribution of raindrops with size. J. Meteor., 5, 165166, doi:10.1175/1520-0469(1948)005<0165:TDORWS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Penide, G., , Protat A. , , Kumar V. V. , , and May P. T. , 2013: Comparison of two convective/stratiform precipitation classification techniques: Radar reflectivity texture versus drop size distribution–based approach. J. Atmos. Oceanic Technol., 30, 27882797, doi:10.1175/JTECH-D-13-00019.1.

    • Search Google Scholar
    • Export Citation
  • Powell, S. W., , and Houze R. A. Jr., 2013: The cloud population and onset of the Madden-Julian Oscillation over the Indian Ocean during DYNAMO-AMIE. J. Geophys. Res. Atmos., 118, 11 97911 995, doi:10.1002/2013JD020421.

    • Search Google Scholar
    • Export Citation
  • Roberto, N., and Coauthors, 2016: Multi-sensor analysis of convective activity in central Italy during the HyMeX SOP 1.1. Atmos. Meas. Tech., 9, 535552, doi:10.5194/amt-9-535-2016.

    • Search Google Scholar
    • Export Citation
  • Rutledge, S. A., , and Houze R. A. Jr., 1987: A diagnostic modeling study of the trailing stratiform region of a midlatitude squall line. J. Atmos. Sci., 44, 26402656, doi:10.1175/1520-0469(1987)044<2640:ADMSOT>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Ryzhkov, A. V., , Giangrande S. E. , , and Schuur T. J. , 2005: Rainfall estimates with a polarimetric prototype of WSR-88D. J. Appl. Meteor., 44, 502515, doi:10.1175/JAM2213.1.

    • Search Google Scholar
    • Export Citation
  • Schumacher, C., , and Houze R. A. Jr., 2003: The TRMM precipitation radar’s view of shallow, isolated rain. J. Appl. Meteor., 42, 15191524, doi:10.1175/1520-0450(2003)042<1519:TTPRVO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Skamarock, W. C., and Coauthors, 2008: A description of the Advanced Research WRF version 3. NCAR Tech. Note NCAR/TN-475+STR, 113 pp., doi:10.5065/D68S4MVH.

  • Steiner, M., , and Houze R. A. Jr., 1993: Three-dimensional validation at TRMM ground truth sites: Some early results from Darwin, Australia. Preprints, 26th Conf. on Radar Meteorology, Norman, OK, Amer. Meteor. Soc., 447450.

  • Steiner, M., , and Smith J. A. , 2002: Use of three-dimensional reflectivity structure for automated detection and removal of nonprecipitating echoes in radar data. J. Atmos. Oceanic Technol., 19, 673686, doi:10.1175/1520-0426(2002)019<0673:UOTDRS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Steiner, M., , Houze R. A. Jr., , and Yuter S. E. , 1995: Climatological characterization of three-dimensional storm structure from operational radar and rain gauge data. J. Appl. Meteor., 34, 19782007, doi:10.1175/1520-0450(1995)034<1978:CCOTDS>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Thompson, E. J., , Rutledge S. A. , , Dolan B. , , and Thurai M. , 2015: Drop size distributions and radar observations of convective and stratiform rain over the equatorial Indian and West Pacific Oceans. J. Atmos. Sci., 72, 40914125, doi:10.1175/JAS-D-14-0206.1.

    • Search Google Scholar
    • Export Citation
  • Thompson, G., , Field P. R. , , Rasmussen R. M. , , and Hall W. D. , 2008: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part II: Implementation of a new snow parameterization. Mon. Wea. Rev., 136, 50955115, doi:10.1175/2008MWR2387.1.

    • Search Google Scholar
    • Export Citation
  • Yang, Y., , Chen X. , , and Qi Y. , 2013: Classification of convective/stratiform echoes in radar reflectivity observations using a fuzzy logic algorithm. J. Geophys. Res. Atmos., 118, 18961905, doi:10.1002/jgrd.50214.

    • Search Google Scholar
    • Export Citation
  • Yoneyama, K., , Zhang C. , , and Long C. N. , 2013: Tracking pulses of the Madden–Julian oscillation. Bull. Amer. Meteor. Soc., 94, 18711891, doi:10.1175/BAMS-D-12-00157.1.

    • Search Google Scholar
    • Export Citation
  • Yuter, S. E., , and Houze R. A. Jr., 1997: Measurements of raindrop size distributions over the Pacific warm pool and implications for Z–R relations. J. Appl. Meteor., 36, 847867, doi:10.1175/1520-0450(1997)036<0847:MORSDO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Zuluaga, M. D., , and Houze R. A. Jr., 2013: Evolution of the population of precipitating convective systems over the equatorial Indian Ocean in active phases of the Madden–Julian oscillation. J. Atmos. Sci., 70, 27132725, doi:10.1175/JAS-D-12-0311.1.

    • Search Google Scholar
    • Export Citation
Save