https://ntrs.nasa.gov/search.jsp?R=20180000191 2018-11-21T05:32:01+00:00Z Chapter 13 Remote Sensing of Precipitation from Airborne and Spaceborne Radar S. Joseph Munchak NASA Goddard Space Flight Center Greenbelt, MD, USA Abstract Weather radar measurements from airborne or satellite platforms can be an effective remote sensing tool for examining the three-dimensional structures of clouds and precipitation. This chapter describes some fundamental properties of radar measurements and their dependence on the particle size distribution (PSD) and radar frequency. The inverse problem of solving for the vertical profile of PSD from a profile of measured reflectivity is stated as an optimal estimation problem for single- and multi-frequency measurements. Phenomena that can change the measured reflectivity Z from its intrinsic m value Z , namely attenuation, non-uniform beam filling, and multiple scattering, are e described and mitigation of these effects in the context of the optimal estimation framework is discussed. Finally, some techniques involving the use of passive microwave measurements to further constrain the retrieval of the PSD are presented. [[H1]] Introduction The ability of weather radar to measure the location and intensity of precipitation was rapidly realized in the late 1940s following World War II. However, ground-based radars are limited in their ability to directly detect precipitation close to the ground far from the radar site due to ground clutter, refraction of the radar beam, and the curvature of the earth. Beam blockage by terrain also poses problems for radar coverage in mountainous areas. Coverage over oceans and other remote areas, where maintaining a ground radar would be difficult and costly, is also impractical, yet the precipitation that falls in these regions has important impacts on the global atmospheric circulations via latent heating (e.g., Hoskins and Karoly, 1981, Hartmann et al., 1984, Matthews et al., 2004) and can have a profound influence on weather patterns thousands of kilometers away. Likewise, knowledge of precipitation over land, particularly in the form of snow, is a crucial component of the mass balance equation for glaciers and ice sheets, which must be properly characterized for realistic climate simulations (Shepherd et al., 2012). Airborne radar systems can provide high sensitivity and finely resolved vertical profiles to characterize precipitation microphysics for the benefit of model parameterizations and process understanding (e.g., Reinhardt et al, 2010, Heymsfield et al., 2013, Rauber et al., 2016). However, in order to achieve true global coverage, it had been proposed from nearly the beginning of the space age to put a weather radar in space (Kreigler and Kawitz, 1960), and efforts to do so began in earnest in the late 1970s and 1980s with the planning of the Tropical Rainfall Measuring Mission satellite (TRMM; Simpson et al., 1987, Okamoto et al., 1988) with its Ku-band Precipitation Radar (PR; Table 1), which was launched in 1997. TRMM was followed in 2014 by the Global Precipitation Measurement (GPM; Hou et al., 2014) misson with its Ku- and Ka-band Dual-frequency Precipitation Radar (DPR; Table 1), which provides increased accuracy, sensitivity, and extension to higher latitudes. Both the TRMM PR and GPM DPR were intended not only to estimate precipitation directly from the radar data but also to construct a database of precipitation profiles to unify precipitation retrievals from passive microwave radiometers (Hou et al., 2014, Kummerow et al., 2015), enabling more frequent coverage than is possible from a narrow-swath on a single satellite. Meanwhile, in 2006, the CloudSat mission (Stephens et al., 2002) with a W-band Cloud Profiling Radar (CPR; Table 1), was launched into polar orbit, complementing TRMM and GPM by providing estimates of light precipitation at high latitudes (Behrangi et al., 2014). <<Table 00-01>> Key parameters of spaceborne weather radars launched prior to 2016. The GPM DPR consists of two radars with matched beams: the KuPR and KaPR. [begin box] In radar engineering and meteorology, it is common to refer to specific frequency ranges (bands) by letter designation. This table lists bands commonly used for meteorological radars according to the IEEE Standard 521-2002. Some of these bands contain gas absorption lines, where atmospheric extinction can be orders of magnitude higher than the surrounding “window” regions. However, it has been proposed to use radars operating at two or more closely-spaced frequencies near some of these bands to estimate vertical profiles of water vapor (e.g., Meneghini et al., 2005, Lebsock et al., 2015) by taking advantage of the differential attenuation. <<Table 00-02>> Band designations, frequency ranges, and significant absorption lines in each band. [end box] In order to minimize size and weight, which strongly correspond to the cost of a satellite mission, it is necessary to use higher frequencies (Ku-, Ka-, or W-band) than are typical for ground radar systems. With increasing frequency, power and antenna size requirements for a desired sensitivity and horizontal resolution are reduced, but attenuation and multiple scattering, which can lead to ambiguity in converting reflectivity to precipitation rate, increase. Even at higher frequencies, the distance from low earth orbit results in ground footprints that are large relative to the scale of variability in most precipitation systems and this non-uniformity must also be considered in precipitation retrieval algorithms. This chapter is intended to provide an overview of the theoretical basis and some practical implementations of precipitation retrieval algorithms for nadir (or near-nadir) looking airborne and spaceborne weather radars at attenuating frequencies, without consideration of polarimetric quantities or Doppler velocity. While dual- polarimetric radars are widely used from ground-based platforms to identify preferentially- oriented, non-spherical hydrometeors, at near-nadir incidence angles these measurements are of limited utility although the linear depolarization ratio measurements can be useful for identifying melting layers and non-spherical ice particles (Pazmany et al., 1994; Galloway et al., 1997). Doppler velocities are useful for inferring hydrometeor fall speeds and at multiple frequencies can be highly effective in discerning cloud liquid from rain (e.g., Kollias et al., 2007) as well as identification of ice particle habits (Kneifel et al., 2016), but obtaining them from rapidly- moving satellite platforms is a difficult engineering challenge that will first be attempted in the EarthCARE mission (Illingworth et al., 2015). Table 1 – Characteristics of TRMM, GPM, and CloudSat Spaceborne Weather Radars [[H1]] Radar Precipitation Measurement Fundamentals The earliest attempts to measure rainfall with radar (Marshall et al., 1947) found that, in general, a power law relationship between radar reflectivity factor Z and rainfall rate R existed: 𝑍 = 𝑎𝑅𝑏. (1) The coefficient a and exponent b of this power law were later provided by Marshall and Palmer (1948), whose values are still in wide use today. Despite this common usage, it was quickly recognized (e.g., Atlas and Chmela 1957) that these parameters varied widely and seemed to be associated with synoptic conditions. It is now recognized (e.g., Brandes et al., 2006) that the power law of Marshall and Palmer (1948) is more representative of frontal stratiform rainfall, which is the predominant rainfall type in Ontario, Canada where the radar and rainfall observations upon which this power law was based were taken. Convective and tropical rainfall, for example, is observed to have a smaller coefficient a (Tokay and Short, 1996). A more comprehensive review of varying power law relations is given by Battan (1973). For the purposes of this chapter, is sufficient to recognize that the non-uniqueness of the Z-R relationship is a fundamental result of the general equations for radar reflectivity and rainfall rate: 𝑍 = ∫𝐷𝑚𝑎𝑥𝑁(𝐷)𝐷6𝑑𝐷, (2) 𝐷 𝑚𝑖𝑛 𝑅 = 𝜋∫𝐷𝑚𝑎𝑥𝐷3𝑁(𝐷)𝑉(𝐷)𝑑𝐷, (3) 6 𝐷𝑚𝑖𝑛 where v(D) is the drop fall speed. Owing to the fact that vertical air motions are small near the ground and that raindrops achieve terminal fall velocity within about 100m (Section 10.3.6, Pruppacher and Klett, 1997) formulae relating terminal fall speed to drop size are often used. A simple power law such as v(D)=17.67D0.67 (where V is in m s-1 and D is in cm; Atlas and Ulbrich, 1977) is convenient for calculation of Z-R power law coefficients by combining (2) and (3), especially when an analytic form of the drop size distribution N(D) is assumed. Slightly more accurate piecewise power laws such as the one given by Beard (1976) account for different hydrodynamic regimes as drops grow in size, and this is the relationship used for all rain rate calculations in this chapter. In (2), there is no dependence of Z on the radar wavelength. This is only valid when the particle size is much smaller than the wavelength. For larger sizes, the equivalent reflectivity factor Z is used instead: e 𝑍 = 𝜆4 ∫𝐷𝑚𝑎𝑥𝑁(𝐷)𝜎 (𝐷,𝜆)𝑑𝐷, (4) 𝑒 𝜋5|𝐾|2 𝐷𝑚𝑖𝑛 𝑏 where K is a function of the complex index of refraction, λ is the radar wavelength in mm, σ is b the backscattering cross section (in mm2), and N(D) is the number concentration of raindrops (in m-3) per size interval, resulting in units of mm6m-3 for Z . Note that (3) and (4) are also valid for e frozen and melting particles, however, the definition of particle size (and fall speeds) becomes ambiguous for non-spherical particles, and even large raindrops exhibit some departures from sphericity. For the remainder of this chapter, the convention will be that D represents the diameter of an equal-mass homogeneous (solid or liquid) sphere, so that N(D) is equivalent to a mass distribution, and N(D) will be referred to as the particle size distribution (PSD). The backscattering cross section describes the amount of electromagnetic radiation that is scattered towards the source of incident radiation. For particles much smaller than the wavelength, the individual dipoles that comprise the particle can be treated as coherent scatterers, and the Rayleigh approximation holds: 𝜋5|𝐾|2𝐷6 𝜎 = (5) 𝑏 𝜆4 and (4) becomes (2). As the size parameter πD/λ approaches one, the Rayleigh approximation breaks down and Mie theory, which provides exact results for spheres, should be used.. For particles that are rotationally symmetric about one axis (such as oblate or prolate spheroids, cylinders, or cones), σ can be calculated from the T-Matrix (Mishchenko and Travis, 1998). b Finally, σ for arbitrarily-shaped particles can be calculated with the discrete-dipole b approximation (Draine and Flatau, 1994), a method often used for realistically-shaped snowflakes (e.g., Liu (2004), Kim (2006), Petty and Huang (2010), Kwo et al. (2016)) and melting particles (Johnson et al., 2016). Plots of σ from all of these approximations can be b found in Figure 1 for spherical and oblate raindrops and spherical, cylindrical, and synthetically- grown ice particles and aggregates. <<Figure 1: Backscattering efficiencies calculated at Ku-, Ka-, and W-band frequencies for rain and ice particles. Spheroidal raindrops were modeled according to aspect ratio and canting angle distributions given in Beard et al., 2010. The cylindrical snow particles, which are an effective representation of hexagonal plates (Adams et al., 2012), were modeled with an aspect ratio (D/h) = 6 and effective density of 0.6 g/cm3. The DDSCAT particles (color indicates relative density) are from the database of Kwo et al. (2016). >> Although the departures of reflectivity from Rayleigh theory are important, particularly for multi-frequency radars, it is still the case that while Z is approximately proportional to the 6th moment of the DSD, rain rate is proportional to a much lower 3.67th moment. This is illustrated in Figure 2, which shows the relative contribution of different drop sizes to reflectivity and rain rate for a typical exponential DSD. Thus the fundamental problem is radar meteorology is that multiple values of R can be associated with a single value of Z. <<Figure 2: Relative contribution (in 0.1 mm bins) to reflectivity (Z) and rainfall rate (R) of an exponential drop size distribution with a median volume-weighted diameter of 1.5 mm. >> Aside from backscatter, another characteristic of precipitation particles that is critical for understanding radar measurements is the extinction cross-section σ . This quantity describes the e amount of electromagnetic radiation absorbed and scattered by the particle, and like σ depends b on the dielectric constant, particle size, and shape (for size parameters close to or great than one). Figure 3 shows the extinction efficiency for the same particle types in Figure 1. <<Figure 3: Extinction efficiencies calculated at Ku-, Ka-, and W-band frequencies for rain and ice particles. Particle type description can be found in the Figure 1 caption.>> For particles with small size parameters, σ is given by: e 𝜋2𝐷3 𝑚2−1 𝜎 = ℑ( ), (6) 𝑒 𝜆 𝑚2+2 where m is the complex index of refraction and ℑ(𝑥) denotes the imaginary part of x. The bulk extinction coefficient, k , can be calculated by integrating σ over the particle size distribution: ext e 𝑘 = ∫𝐷𝑚𝑎𝑥𝜎 𝑁(𝐷)𝑑𝐷. (7) 𝑒𝑥𝑡 𝐷 𝑒 𝑚𝑖𝑛 Note that in addition to precipitation particles, atmospheric gases such as oxygen and water vapor can have non-negligible contributions to k at some wavelengths common to airborne and ext spaceborne radars (Li et al., 2001; Tanelli et al., 2006, Ellis and Vivekanandan, 2010). Cloud water, which has a negligible contribution to radar reflectivity at Ka-band and lower frequencies, nevertheless can contribute significantly to k at these frequencies as well (Grecu and Olson, ext 2008). Note the independence of k with respect to D in the limit of small D in Figure 5. This ext m m follows from (6) and shows that in the limit of small particles, k is directly proportional to the ext water/ice content. The effect of bulk extinction on the measured reflectivity can be calculated by integrating k along the two-way radar propagation path: ext 𝑟 𝑍𝑚(𝑟) = 𝑒−0.2ln⁡(10)∫0𝑘𝑒𝑥𝑡(𝑠)𝑑𝑠𝑍𝑒(𝑟) (8) Note that while Z is an intrinsic property of the PSD at a given location, Z also depends on the e m integrated bulk extinction between the source of the radar signal and that location, creating another source of uncertainty when converting from Z to R. Therefore, understanding the vertical profile of the PSD, on which Z, k , and R depend, is fundamental to the radar-precipitation ext profiling algorithms that are described in this chapter. [[H1]] The Particle Size Distribution The previous section demonstrated that knowledge of the particle size distribution (PSD) is needed to convert Z to the physical integrated quantities such as precipitation rate R and water content W. It is often convenient to assume an analytical form of the PSD that describes the shape with a few (relative to a discrete bin representation) free parameters. Although PSD models with as many as eight free parameters have been proposed (Kuo et al., 2004), most often the modified gamma distribution with three free parameters (Ulbrich, 1983) is used: 𝑁(𝐷) =⁡𝑁 𝐷𝜇exp⁡(−Λ𝐷). (9) 0 The free parameters are often referred to as the intercept (N ), slope (Λ), and shape (μ). These 0 names describe the mathematical form of the distribution moreso than physical quantities, and it is difficult to impart any physical meaning to a value for anyone one of the parameters given in isolation. However, it is possible to re-cast these parameters in terms of physical quantities through the following relationships (Testud et al., 2001; Williams et al., 2014): 4+𝜇 𝐷 = , (10) 𝑚 Λ 𝜇 Γ(4+𝜇) 256 𝑁 = 𝑁 𝐷 , (11) 𝑤 0 𝑚 Γ(4) (4+𝜇)4+𝜇 𝜎 = 𝐷𝑚 , (12) 𝑚 √4+𝜇 where D is the mass-weighted mean diameter (also defined as the ratio of the 4th to 3rd moment m of the PSD), N is the normalized intercept parameter defined such that it is equal to N for an w 0 exponential (μ=0) PSD of the same water content and D (Bringi and Chandrasekhar 2001), and m σ is the mass spectrum standard deviation. The median volume diameter, D , describes the m 0 particle size such that ∫𝐷0 𝐷3𝑑𝐷 =∫𝐷𝑚𝑎𝑥𝐷3𝑑𝐷, and for a gamma distribution 𝐷 = 𝐷𝑚𝑖𝑛 𝐷0 0 3.67+𝜇 𝐷 .⁡Note that these expressions are only valid for a PSD where particle density is constant 𝑚 4+𝜇 with size. For realistic frozen particles, this is often not the case, and formulae to convert PSD parameters from observed to solid-sphere-equivalent particle sizes can be found in Petty and Huang (2011). Analysis of disdrometer observations has shown that neither the three free parameters in (9) nor the physical quantities in (10)-(12) are statistically independent in rain (e.g., Haddad et al., 1996; Zhang et al., 2003; Munchak and Tokay, 2008; Williams et al., 2014.) and relationships between the parameters can be formulated to reduce the degrees of freedom in (9). This is particularly useful for radar precipitation retrieval algorithms because it allows for a common basis from which to compute Z-R and Z-k relationships, and, when joint probability distribution functions ext (pdfs) of PSD parameters are known, these a priori statistics can be used to constrain the retrieval of PSD parameters from radar profiles of reflectivity and other information. To demonstrate the relationship between PSD parameters and radar reflectivity Z, Figures 4 and 5 show reflectivity and extinction coefficients integrated over PSDs with D ranging from 0.1 to m 3 mm and μ ranging from -1 to 3 for snow and following the σ -D relationships given by m m Williams et al. (2014) for rain. The integrated water content of all PSDs was normalized to W=1 g m-3 to emphasize the importance of the shape of the PSD in terms of its mean and dispersion on the reflectivity and extinction. For these PSDs, the equivalent intercept N can be calculated w from D and W (Testud et al., 2001): m 256 𝑊 𝑁 = . (13) 𝑤 𝜋𝜌𝑤𝐷𝑚4