Based on the fundamental relationship involving the interaction of microwave radiation with precipitation, microwave-based satellite precipitation estimates hold the most promise for quantitative rain estimation from space. At present, the low-resolution channels onboard the DMSP Special Sensor Microwave Imager (SSM/I) are sampled with a spatial resolution several times larger than the scale at which rainfall is generated in typical convective rainbands. Aircraft-based instruments can provide views of the detailed microwave radiometric characteristics of precipitating clouds.
In this manuscript, the authors present coincident finescale (1–3-km resolution) collocated aircraft radiometric and aircraft precipitation radar measurements collected during the 1993 Tropical Ocean Global Atmosphere Coupled Ocean–Atmosphere Response Experiment in the western Pacific Ocean. By intentionally degrading the resolution of the aircraft datasets from their native resolution to that of current and future spaceborne sensors, the impact of sensor resolution upon a combined radiometer–radar vertical profiling rain-retrieval algorithm (developed and utilized for the Precipitation Intercomparison Program 2) was examined. Retrieved values of the columnar graupel content were more influenced by the addition of the radar profile than was the columnar rain content. The retrieved values of columnar graupel were also significantly smaller than previously published results for land-based rainfall. The results show that the general trend of the rain structure is maintained but finescale details are lost once the observations are reduced to resolutions of 15 km.
One of the more long-standing issues in the remote sensing of precipitation concerns the relation between the sensor resolutions and the horizontal scales of the precipitating cloud, and how this discrepancy affects the satellite estimation of precipitation. Finescale infrared satellite imagery capable of sampling the dimensions of typical precipitating clouds has been available for some time with geostationary satellite platforms, and high-resolution microwave sensors have been proposed (Savage et al. 1994). While geostationary IR-based rain estimates (Arkin et al. 1994; Johnson et al. 1994) provide finer-scale spatial and timely updates of thermal IR temperatures, the connection between the IR measurementand the underlying rain physics is highly variable (Heymsfield and Fulton 1994). Coarser-scale passive microwave satellite cloud imagery has gained increased usage following the deployment of the initial United States Defense Meteorological Satellite Program (DMSP) Special Sensor Microwave Imager (SSM/I) and its companion water vapor sounding instrument, the Special Sensor Microwave Water Vapor Sounder (SSM/T2) (Falcone et al. 1992). The on-earth spatial resolutions of both DMSP instruments are listed in Table 1. Although the SSM/I’s sampling and resolution are capable of capturing the general features and outlines of cyclones and lengthy squall lines, less organized precipitating systems often contain relatively smaller, more individual rain cells. The large 19-GHz SSM/I channel footprint often encompasses only a fraction of the rain cloud, while the finer-scale 85-GHz measurements are related to the mid- and upper-level graupel mass content (Mugnai et al. 1993). Both rain and ice contents are important factors in the estimate of cloud latent heatingand their retrievals are dependent upon how the low-resolution satellite sensor resolves the finer-scale cloud characteristics inside of the sensor footprint. From this standpoint, coincident fine- and coarse-scale microwave radiometric data would be useful for testing of deconvolution techniques and precipitation retrieval algorithms that accomodate the inherent coarseness of current microwave radiometric data.
In this manuscript, we examine coincident passive and active microwave measurements gathered over tropical rainfall by a finescale airborne radar–radiometer combination. The datasets used were gathered during TOGA COARE (Tropical Ocean Global Atmosphere Coupled Ocean–Atmosphere Response Experiment), conducted over the tropical western south Pacific Ocean in late 1992 and early 1993. Science objectives included ground truth and algorithm development support for the Tropical Rainfall Measuring Mission (TRMM) using the National Aeronautics and Space Administration(NASA) Advanced Microwave Precipitation Radiometer (AMPR) and the Millimeter Wave Imaging Radiometer (MIR), both mounted onboard a high-altitude ER-2 aircraft. During flights dedicated to supporting precipitation studies, a lower altitude DC-8 aircraft carrying the Jet Propulsion Laboratory (JPL) 13.8-GHz Airborne Rain Mapping Radar (ARMAR) flew several well-coordinated flight lines below the ER-2. Although the instrument resolutions and scanning geometries are different than the microwave instruments that will be aboard the TRMM satellite, the AMPR–ARMAR combination does simulate the TRMM microwave instrumentation and frequencies from a low-altitude perspective. Figure 1 depicts the scanning geometry of the AMPR and ARMAR. Recently, the ER-2 Doppler Radar (EDOP) was added to the ER-2 microwave instrumentation, making this aircraft the premier platform for overflying precipitating clouds (Caylor et al. 1995).
The 10- and 85-GHz channels of the AMPR haveattracted particular interest. The addition of a 10-GHz channel to the channels between 19 and 85 GHz has already been demonstrated to yield definite improvements to rain-rate estimation (Smith et al. 1994). McGaughey et al. (1996) averaged the 85-GHz AMPR data from TOGA COARE to the resolutions of the TRMM Microwave Imager (TMI) and the SSM/I, and showed how these degraded TB varied with the minimum TB taken in the average. The radiometric behavior of the TB ice scattering signatures was shown to occur at scales smaller than either the TMI or SSM/I 85-GHz resolution. In order to examine the effects of passive microwave sensor resolution upon actual rain-rate retrievals, this study presents a series of intentional degradations to a set of combined AMPR and ARMAR data, with the intent of demonstrating the effect of sensor resolution upon the retrieval of rain rate and the associated hydrometeor profiles of columnar rain and graupel contents. Olson and Kummerow (1996) used simulated data to investigate the effects of sensor resolution on future TRMM data. Several teams have developed varieties of rain and hydrometeor profiling-type algorithms to date (Smith and Tripoli 1994; Marzano et al. 1994; Meneghini et al. 1994; Kummerow and Giglio 1994; Evans et al. 1995; Olson et al. 1996; Haddad et al. 1996). We will not address the differences but note that Olson et al. (1996) applied some early AMPR–EDOP combined datasets from the 1993 Convection and Moisture Experiment in a profiling algorithm (EDOP operates at 9.7 GHz).
Many profiling algorithms link the microphysical output of a three-dimensional cloud model to the brightness temperatures or other measurements through optimization techniques and fast forward radiative transfer modeling. The input observations used by profiling algorithms can be obtained from radiometric, radar, or any other datasets (provided that the observation can be forward modeled), although most profiling algorithms have been appropriately designed around the passive microwave TB channels of the SSM/I. In this work, we used these TOGA COARE datasets as input to the profiling algorithm and approach of Marzano et al. (1995) and Marzano et al. (1996), which the authors had originally applied to several land-based AMPR datasets from the 1991 Convection and Precipitation Experiment (CaPE) in Florida (Marzano et al. 1994) and also to SSMI/I data as part of the Precipitation Intercomparison Project (PIP-2) program (Smith et al. 1997). For application to the TOGA COARE data, we have expanded the algorithm’s capability to use not only the passive microwave TB channels, but also the attenuated ARMAR reflectivity profile, which is incorporated as an additional observation to the measurement vector (often the term “tall vector” is used to describe this type of profiling retrieval methodology). Just as the algorithm uses forward radiative transfer calculations to model the AMPR TB, it also calculates a simulated ARMAR profile from each candidate hydrometeor profile; the idea beingthat the radar profile provides additional capabilities that better constrain the retrieved hydrometeor structure. By incorporating radar data in this manner, conventional reflectivity–rain rate or reflectivity–attenuation relations are bypassed, as these often vary throughout the cloud lifetime and are usually tuned for specific climatologies. While still in their development stage and difficult to verify, tall-vector algorithms hold promise for the retrieval of not only rain rate, but the profile of liquid and ice contents needed for the derivation of the latent heat that is released during the precipitation process.
Section 2 discusses the aircraft radar and radiometer systems and specifications, as well as the procedures used to align the distinctly different ER-2 and DC-8 aircraft datasets. Section 3 compares the brightness temperatures between the various AMPR channels with the structure of the ARMAR reflectivity field and some near-coincident SSM/I imagery. Throughout this section, the associated rain rate and integrated rain and graupel contents obtained from the retrieval technique (using intentionally degraded resolution AMPR data) is also shown, in effect simulating various resolutions of the SSM/I and TRMM satellite sensor configurations. To keep the combined radar–radiometer datasets from becoming overly complex, we perform intercomparisons using only the nadir aircraft data, where the individual radar and radiometer instruments view along coincident paths. While this is a simplification of the more difficult case where the radar and radiometer beams are not colocated, the analysis of the nadir data is still far from a trivial task and provides the simplest set of forward measurements required for tall-vector profiling-type precipitation retrieval algorithms.
2. Instrumentation and data fusion
Several articles have now been printed since the AMPR was first developed for NASA Marshall Space Flight Center and tested in late 1990 (Smith et al. 1994; Spencer et al. 1994; McGaughey et al. 1996). The TOGA COARE AMPR measurements reported here are complementary to the measurements that the authors analyzed during the 1991 CaPE experiment in central Florida (Vivekanandan et al. 1993; Turk et al. 1994; Marzano et al. 1994). That is, the CaPE storms were nearly all land-based, while the TOGA COARE storms are all ocean based. Furthermore, the supporting radar data from CaPE came from the ground-based National Center for Atmospheric Research CP-2 multiparameter system and required special processing to obtain the radar data along the radiometer beam, while here the TOGA COARE supporting radar data is from the airborne ARMAR, a downward-viewing profiling-type precipitation radar.
The AMPR window channel frequencies are also present onboard the future TRMM TMI (10.7, 19.35, 37.1, 85.5 GHz). We will truncate the decimal and refer to these as 10, 19, 37, and 85 GHz below. The 10-GHzchannel has the advantage of being nearly unaffected by columnar water vapor and ice, but its larger sampling area is less capable of resolving smaller-scale precipitation features noticed by the 85-GHz channel, and it is also more affected by variations in the underlying surface emissivity. The AMPR scans ±45° across track and images 50 pixels per scan. Each scan requires 3 s, during which time the ER-2 moves forward 700 m. This maintains contiguous 85-GHz imaging. The resolutions of the 10- and 19-GHz AMPR channels are matched at 2.8 km; 37 and 85 GHz have resolutions of 1.5 and 0.6 km, respectively. The AMPR feedhorn is configured such that its receive polarization state skews from H to V across the scan (right to left as viewed by the pilot). For clear skies over ocean, the skew is noticeable as a gradual TB warming across each scan, due to the higher vertically polarized emissivity of the ocean surface. However, over the range ±15°, the polarization state of the AMPR is relatively constant at 45° linear (Vivekanandan et al. 1993). Turk and Vivekanandan (1995) quantified the net TB polarization difference due to the presence of nonspherically shaped precipitation particles. After incorporating oblate spheroidally shaped raindrops, conically shaped graupel, and plate-shaped ice crystals into a multistream radiative transfer model with a polarized ocean surface, the final net upwelling radiation was largely unpolarized near nadir. However, a net 85-GHz TVB − THB polarization difference of 15 K remained for view angles greater than 45° from nadir. This supports recent SSM/I observations of Heymsfield and Fulton (1994). Hence, we believe that it is possible to use the AMPR measurements quantitatively over ±15° of nadir, where the assumption that the scene THB ≈ TVB is valid.
The pulsed-Doppler, polarization-agile ARMAR was designed and constructed by the Jet Propulsion Laboratory (JPL). It employs pulse compression to achieve low-range sidelobe levels, which can otherwise mask near-ocean rain signals due to the presence of the strong ocean return (Durden et al. 1994). The DC-8 flight altitude of ≈10 km is half that of the ER-2, and in its normal mode of operation ARMAR scans ±20° across track. Instrument characteristics are listed in Table 2. Depending upon the vertical resolution requirements, ittypically acquires 20 beams per scan, and its on-earth nadir resolution is about 600 m. During these TOGA COARE flights, the range gate spacing was set at 60 m. At 14 GHz, attenuation rates of 2 dB km−1 are typical for a rain rate of 30 mm h−1, so much of the TOGA COARE data exhibits artifacts of propagation attenuation below the melting level of 4.5-km altitude. Yeh et al. (1995) performed calculations of the simulated 13.8-GHz radar reflectivity profile from 3D cloud model output over land. They noted up to 20–30 dB difference between the peak and near-surface reflectivities in convective cores due to attenuation. Caylor et al. (1995) retrieved similar values of attenuation with the ER-2 EDOP radar, which operates near 10 GHz. There are a variety of techniques specifically designed to capitalize upon the attenuation, most notably surface reflection techniques pioneered at NASA Goddard Space Flight Center (Meneghini et al. 1989), to which we refer the interested reader.
Most of the ARMAR data gathered for these data had the radar configured for the collection of reflectivity, Doppler velocity, and spectrum width, but some measurements of the linear depolarization ratio LDR (the ratio of the cross-polarized to the copolarized radar return) were available (Durden et al. 1994). The LDR is dependent upon the magnitude of the cross-polarized return, which is dependent upon the hydrometeor shape, canting, and effective dielectric constant. It is a useful parameter for detection of mixed phase and melting conditions (Herzegh and Jameson 1992). ARMAR also records a radiometric TB for each beam position, which proved very useful when aligning these data with the AMPR.
a. Combining AMPR and ARMAR nadir data
Overall, we have extracted 12 ER-2 straight-line storm overpasses that were well coordinated with the DC-8 flight lines on 4, 8, 20, and 22 February 1993. An examination of the AMPR and MIR data from a series of intense convection flights on 22 February was undertaken by Heymsfield et al. (1995). In this study, we examine the aircraft overflights from 4 and 8 February, since they gathered data during the developing and mature stages of Tropical Cyclone Oliver, respectively, in addition to having well-coordinated, aligned flights between the ER-2 and DC-8. In order to examine these data quantitatively, special care was taken into account to properly align the data from the AMPR, MIR, and ARMAR instruments. The procedure is briefly outlined below.
After locating a flight segment where the ER-2 flew a straight-line path with the DC-8, the AMPR and MIR data were aligned by examining the nadir trace of their 85- and 89-GHz channels, respectively. Since the polarized water surface is substantially depolarized by the intervening atmosphere at these frequencies, the imagery structure from these two channels is very similar.The scanning operation of the MIR and AMPR instruments was aligned by plotting the nadir trace of the 85- and 89-GHz channels. Typically the MIR gathered a few additional scans for a given flight time interval, and this was accounted for by interpolating the AMPR records at equally spaced intervals throughout the time interval and saving these universal time and latitude–longitude records for each nadir pixel in the AMPR–MIR data. Next, individual scans of the AMPR–MIR dataset and the ARMAR data were coordinated by the ER-2 and DC-8 latitude and longitude records. Data records were rejected when the ER-2 and DC-8 drifted more than 3 km apart (usually due to the DC-8 maneuvering to avoid turbulence). Fine-tuning was accomplished by examining the nadir trace of the AMPR 10-and 19-GHz TB with the ARMAR 14-GHz TB, such that the 14-GHz TB peaks and valleys fell between those at 10 and 19 GHz. The best alignment occured on 4 and 8 February, where the on-earth alignment of the ER-2 and DC-8 was within 500 m, although the aircraft were often a few minutes apart. The 14-GHz reflectivities and 14-GHz TB were averaged over two adjacent pixels, which somewhat accounts for the finer spatial resolution of the ARMAR compared to the 10-GHz AMPR TB.
b. Comparisons with near-coincident SSM/I imagery
With regard to the nearest-time DMSP satellite imagery, the SSM/I conical-scan orientation and 1400-km swath assures the presence of coverage gaps in the equatorial regions, so adequate coverage of a specific equatorial area may occur several days apart. The high-resolution 85-GHz SSM/I pixels sampled every 12.5 km along each scan and each scan line is 12.5 km apart, while the lower five channels are sampled every other line and sample (25-km sampling). For a plan view of the storm structure and evolution, the area covered by the limits of the AMPR scan (40 km on-earth distance across track) is plotted on top of recent F-11 SSM/I imagery as a boxed region corresponding to the on-earth coverage. Care has been taken to display these DMSP data at their native resolution, without any low-pass filtering or deconvolution techniques. The 1905 UTC F-11 descending pass from 4 February 1993 is shown in the top two panels of Fig. 2, where the 19- and 37-GHz TB are remapped onto a rectangular projection (grid lines 2° apart) with the two AMPR scan-coverage areas from this date drawn as a boxed region and annotated with the pass date and time. As a result, there is near-coincident coverage of the storm structure as viewed from polar orbit. Unfortunately, these SSM/I data are corrupted by several missing scan lines during the second ER-2 overflight on this date (during 1901–1912 UTC), when Cyclone Oliver was in its forming stage.
The next decent scan coverage of the full cyclone area occured for the 0710 UTC F-11 ascending (localevening) pass on 7 February, where Oliver had reached its mature phase. Here, the location of the AMPR scan coverage is shown for the two passes on 8 February as well as an ER-2-only overflight (no DC-8) from 7 February. The 8 February ER-2 overflights occured over a day later than this F-11 pass. A well-developed eye in the 85-GHz (lower right) panel is evident as well as 85-GHz TB near 180 K in the outer rainbands, indicative of ice scattering associated with convective rainfall.
3. Effects of aircraft data resolution upon rain-rate retrieval
Referring to the AMPR coverage regions drawn upon Fig. 2, we now compare the above SSM/I imagery with the finer-scale aircraft data gathered at the same frequencies during ER-2 overflights from 4 and 8 February 1993. The idea is to examine how the rain and ice retrievals from tropical rainfall systems are affected by the sensor resolution and the along-track sampling distance. This is accomplished through a series of intentional degradations to the finescale AMPR brightness temperatures and applications to a precipitation-retrieval technique.
The effect of antenna aperture size and sensor resolution upon retrievals of rainfall and integrated ice content has been simulated in a recent article by Skofronik-Jackson and Gasiewski (1995). Their rain-rate retrievals suffered more than the integrated ice content as the sensor resolution was degraded from 1.5 to 40 km. The results presented here use actual measured data from tropical rain systems and also incorporates additional radar data, so any differences are expected to be revealing. McGaughey et al. (1996) examined the effects arising from degradations to the AMPR data at scales of 24 km. This study will focus on the actual AMPR microphysical retrievals based on the degraded AMPR TB. For one overpass case, the degraded retrieval results are presented alongside the corresponding SSM/I-based rain and ice retrievals.
a. Degradations to the AMPR resolution
The spatial resolution and sampling of satellite microwave radiometers has a strong impact on the estimation accuracy of the cloud and precipitation structure. In order to demonstrate these effects quantitatively, we have applied in the along-track direction both low-pass filtering (to degrade the finescale resolution) and a resampling algorithm (to simulate how a satellite sensor samples in the along-track direction) to the AMPR data. The low-pass filter consists of a moving-average window with resultant along-track sizes equal to the AMPR 10-GHz field of view (FOV) (3 km), future TRMM 85-GHz FOV (6 km), SSM/I 85-GHz FOV (15 km), and SSM/I 37-GHz FOV (about 30 km). Considering the AMPR sampling distance of 0.7 km along track, then the number of pixels N to be averaged is given by N = NoddInt[(F − R)/0.7], where F is the output FOV, R is the AMPR channel resolution, and NoddInt is the operator “nearest odd integer” (e.g., to obtain F = 30 km, N = 39 at 10 GHz, and N = 43 at 85 GHz).
By placing each moving-average window on the AMPR nadir pixel, we obtain TB traces with the original along-track AMPR sampling of 0.7 km. With the aim of simulating actual satellite measurements, we have also resampled each moving-average window at the corresponding sampling rate. The latter choice implies to resample the simulated TRMM 85-GHz FOV every 3 km, SSM/I 85-GHz FOV every 12.5 km, and SSM/I 37-GHz FOV every 25 km. To each window centered on the sampled pixel a constant TB value has been assigned equal to the center pixel TB value. Of course, the resampling is affected by the initial pixel and may be shifted arbitrarily. In the figures, we have initiated the resampling by choosing to place an entire FOV at the beginning of the track.
b. Microphysical retrieval methodology
The precipitation retrieval algorithm used is based on the statistical–physical method developed jointly by the Istituto di Fisica dell’Atmosfera and the University of Rome “La Sapienza” (Marzano et al. 1995; Marzano et al. 1996). It is an extension of the maximum a posteriori probability estimation technique (Pierdicca et al. 1996), where the coincident radar-reflectivity measurements are included into the measurement vector. The combination of radar–radiometric techniques is appealing and is attracting attention from several algorithm development teams with regard to the TRMM era (Olson et al. 1996; Haddad et al. 1996).
In particular, the manner in which the radar observations is treated is fundamentally different from typical range-gated relations that relate the (attenuation corrected) reflectivity to a microphysical quantity such as rain rate or mass content. Usually, an estimate of the total path attenuation is retrieved and then apportioned along the beam to correct for cumulative attenuation. In this and other tall-vector precipitation retrieval algorithms, rather than attempting to correct the radar data for path attenuation, the entire radar reflectivity profile is included as an additional dimension in the input measurement vector along with the TB observations. Both the multiple TB values and attenuated radar profile are then simulated by solving the forward problem with hydrometeor profiles extracted from a microphysically detailed cloud-model database. In general, since the forward problem is better understood than the inverse problem (i.e., the inversion of the measurements into hydrometeor profiles), the resultant hydrometeor profiles can be consistent with both the observations and mesoscale model microphysics. The resultant profiles are extracted from the database via optimization techniques and computationally fast radar and radiative transfer forward models (Smith and Tripoli 1994).
With regards to the TOGA COARE data, we have applied this technique to the combined datasets of 4 and 8 February 1993. The measurement tall-vector consists of the AMPR TB together with the ARMAR reflectivity profile, where the radar data has been vertically averaged to layers of about 1.5 km. After assuming the error statistics of the TB and Z, the estimated profile of hydrometeor contents is computed by minimizing a discriminant function where the probability of the hydrometeor profile is also included in terms of a mean and a correlation matrix. The latter statistics are derived from a microphysical cloud-model simulation of a hurricane over a grid of 330 × 330 km2 (Panegrossi et al. 1998). Marzano et al. (1996) demonstrated that the incorporation of the reflectivity profile in this fashion mainly affects the estimation of columnar rain, surface rain rate, and, partly, the precipitation-sized graupel.
Below, we compare also the estimation results obtained by excluding the reflectivity profile within the measurement tall vector. This is a realistic situationwhen the swath of a spaceborne radar is narrower than that of the scanning radiometer system. Moreover, since on the DMSP platforms there is no spaceborne radar, we do not include the reflectivity profile when doing calculations at 15 km and 30 FOV resolution. The simulated SSM/I retrievals are made by averaging the TB at 19, 37, and 85 GHz (the AMPR does not contain the 22.235-GHz SSM/I channel).
c. Overflights over rainbands from 4 February
The AMPR TB imagery from 1901 to 1912 UTC on 4 February is presented in Fig. 3, with the TB scale depicted on the right side. This can be thought of as a magnified view of the radiometric TB inside of the boxed region drawn onto the SSM/I image of Fig. 2, since the SSM/I and ER-2 times are near coincident. The swath width of these images is 40 km (about 0.35° on earth), and the instrument-received polarization skew is evident along the leftmost side of the 10-GHz image. The banded structure of the 10-GHz AMPR image in Fig. 3 reveals a smaller-scale nature of precipitation structure formation about the forming eye, with 37- and 85-GHz scattering features noticeable between scan numbers 110 and 140. The 19-GHz TB is well-saturated past 275 K in the heavily raining regions and above 225 K for the lightest rainfall. Coincident 19-GHz SSM/I pixels, while few in such a small region, do also generally fall above 250 K with limited dynamic range as well. The characteristic of the 19 GHz channel to saturate over heavy rain has generally placed an upper limit on the maximum rain rate that can be trusted from many SSM/I rain algorithms.
On the left-hand side of Fig. 3 is a contour of the reconstructed Constant Altitude Plan-Position Indicator (CAPPI) ARMAR reflectivity field at 500-m altitude. The reflectivity scale is depicted on the bottom of the figure. This is high enough off the surface to avoid contamination by the strong ocean return contained in the range sidelobes of the ARMAR pulse compression processing (Durden et al. 1994). With the ARMAR ±20° across-track scanning, the on-earth swath width is ≈8 km (vs 40 km for the AMPR), so in actuality this CAPPI is comparable to a narrow strip centered along nadir in the 10-GHz AMPR image. These dimensions are noted in the bottom of Fig. 3 (the CAPPI is intentionally stretched in the across-track direction for viewing purposes). The CAPPI provides an even finer view of the surface precipitation structure than revealed by the radiometric 10-GHz image by about a factor of 5. The range-attenuated radar Z nearest the surface is useful to tall-vector rainfall-retrieval algorithms as a “consistency check” on rain rate retrieved nearest the surface. For example, the 85-GHz scattering feature near scan 130 in Fig. 3 is associated with Z ≈ 45 dBZ, but this reflectivity is depreciated by the two-way path attenuation betweeen the 500-m range gate and the radar. After accounting for radar beam attenuation, the attenuated-corrected Z value is likely in excess of 50 dBZ (at 30 mm h−1, the attenuation rate is ≈2 dB km−1 and the intrinsic reflectivity ≈48 dBZ). Furthermore, 10-GHz model results demonstrate that the TB is largely dependent upon the path extinction alone, making the 10-GHz TB a robust estimator of the total radar path attenuation over the range of wind speeds and surface temperatures likely to be encountered in the tropical oceans (Turk et al. 1995; Smith et al. 1997). Therefore, the surface rain rate can be obtained by an appropriate Z–R relation, and used as a consistency check (as opposed to an actual ground-truth verification) against the rain rate produced by the hydrometeor profiling algorithm (Marzano et al. 1996).
Extracting the vertical profile of the radar data along the aircraft nadir track, we plot a segment of the height versus time cross section in Fig. 4, with the nadir AMPR TB traces in the bottom panel. The region of 85-GHz scattering between scans 120 and 140 contains Z between 20 and 30 dB up to 11 km (where the DC-8 flewthrough the frozen hydrometeors), with evidence of vertical air motion elevating large melting hydrometeors above the 0°C temperature level of ≈4.5-km altitude. Up until scan 100 or so, the DC-8 likely flew through regions of smaller ice water content and smaller vertical air motion. About the 0°C level, a thin radar bright band between scans 70 and 100 is present, indicating of the onset of falling ice crystals that very quickly melt and fall as rain. The LDR profile (not shown) does indicate this onset of melting quite nicely.
Since the SSM/I data during this ER-2 pass is corrupted by several missing scan lines, we have chosen to apply the retrieval results to the earlier ER-2 pass on this date. This is the nadir AMPR data collected on 4 February 1993 from 1715 to 1727 UTC (see the top of Fig. 2 for the AMPR scanning region from this time period overlaid on the nearby SSM/I imagery), to which we refer to for the remainder of the 4 February analysis. In the top two panels of Fig. 5, the resultant 10- and 85-GHz AMPR channels averaged to effective FOV resolutions of 3, 6, 15, and 30 km are plotted. In addition, the along-track sampling distance has been adjusted at each of these FOV degradations to 0.7, 3, 12.5, and 25 km, respectively. The latter two sampling distances are representative of the SSM/I A and B scan along-track sampling distances. For the 19-, 37-, and 85-GHz plots, the nearest colocated SSM/I pixels have been extracted, and the 19- and 85-GHz pixels are plotted in the bottom two panels of Fig. 5. Note that the timing of this SSM/I pass is about 2 h later than the time of the 1715–1727 UTC ER-2 overpass. With this two hour offset, the rain structure has likely evolved, so any comparisons can only be made by shifting the data and keeping this in mind when interpreting any conclusions. Here, the SSM/I data seems to be shifted by about 45 AMPR pixels (about 30 km), if the 19-GHz AMPR maximum around pixel 50 is aligned with the SSM/I maximum near pixel 95. Even so, most of the fine detail in the very coarsest AMPR data still contains more detail in the TB structure than does the SSM/I. Similarly, McGaughey et al. (1996) showed high variability in the 85-GHz scattering-based TB depressions when averaging the 85-GHz TB.
The results of the retrieval algorithm when appliedto the various stages of degraded AMPR data are plotted in the top two panels of Figs. 6 (for rain rate and nonprecipitating ice) and Fig. 7 (for columnar rain and columnar graupel contents). In both Figs. 6 and 7, the bottom two panels depict the corresponding SSM/I retrieved values. Given the finescale nature of the input AMPR measurement vector, the retrievals of the four microphysical quantities (rain rate, columnar rain, columnar ice, and columnar graupel) are increasingly different for each stage of the resolution degradation. These smoothings of the AMPR FOV produce rain-rate estimates in fairly good agreement with the highest resolution until the 15-km resolution. For the 30-km resolution and 25-km sampling, the structure of the retrieved rain rate is somewhat recognizable, but the small rainbands are not captured. From pixels 70–100, the rain rate is higher for the degraded case due to the higher-averaged TB values at 10 GHz. The corresponding SSM/I-retrieved rain rates are about half those retrieved from the AMPR and shifted due to the two hour time difference, and the rain details are lost.
The integrated columnar rain and graupel retrievals of Fig. 7 show peak values of approximately 6 and 1 kg m−2 for columnar rain and graupel, respectively. Therain value is similar to that retrieved from the ER-2 AMPR overflights from the 1991 CaPE Florida flights, which were all over heavy land-based convection containing 85-GHz TB depression near 100 K. However, the columnar graupel from the Florida storms often exceeded 12–15 kg m−1 (Marzano et al. 1994), much larger than the values retrieved here. The small ice content has been noted by others who have analyzed the TOGA COARE datasets (McGaughey et al. 1996) as indicative of the nature of much tropical rainfall (heavy rain with a weak ice phase). If so, then the scattering-based rain-rate algorithms that rely upon the 85-GHz channel may require special adaptations to work with rainfall in the Tropics versus rainfall at midlatitudes.
d. Overflights near the eye on 8 February
For the AMPR imagery between 2044 and 2104 UTC on 8 February, the cyclone eye is strong but decaying, and the ER-2 and DC-8 performed an aligned overflight directly over the eye. We present a similar set of AMPR and ARMAR figures as above in Figs. 8 and 9. We do not use the SSM/I data quantitatively in this case due to the 1-day time difference. The nonraining background from the 85-GHz AMPR channel is near 275 K, which is near the value measured by the SSM/I 85V channel (the AMPR polarization mixing is not noticeable at 85 GHz). In such a tightly organized mature state, there appears to be a wide variability in rain intensity about the eye. While the SSM/I 85-GHz resolutions are sharp enough to detect the eyewall pattern, the AMPR 10-GHz detects the presence of 3–4 individual rain features within the overall eyewall banded about the low pressure center. The only noticeable 85-GHz scattering signature occurs near scan 175, and then it is only near 200 K, similar to the SSM/I 85-GHz values in this region. While the 10-GHz channel appears to best identify likely rain regions, the 85-GHz image provides more of a description of the rainfall convective intensity. Overall, the relatively weak scattering TB depressions are in contrast to the steep scattering-based 19-, 37-, and 85-GHz TB depressions noted in AMPR measurements over land-based precipitation cores in central Florida (Vivekanandan et al. 1993). The ARMAR profile in Fig. 9 has its cloud tops lower and sheared off at an altitude of ≈10 km, lower than the 4 February pass shown in Fig. 4. The 30–35 dBZ contour also does not extend as high as the 4 February pass.
By comparison, the MIR imagery corresponding to the Fig. 8 AMPR imagery is depicted in Fig. 10. Note the nearly identical 85-GHz AMPR imagery and 89-GHz MIR imagery, matched to within 1–2 K near nadir. The MIR 183.31 ± 1-GHz TB imagery is fairly constant near 240 K except around the eyewall region, where it falls by several tens of degrees due to scattering from ice within the storm (this serves to further reduce the TB beyond that due to water vapor alone). The 183.31 ± 7-GHz channel is further removed from the center of the water vapor absorption line, and hence it shows more storm structure in its TB image. It is more affected by slight changes in the storm convective activity than the 183.31 ± 1-GHz channel. The 220-GHz channel, while in a window region of the atmospheric transmissivity spectrum, is sufficiently far down in the watervapor absorption continuum that it appears generally similar to the 183.31 ± 7-GHz image. Using these same TOGA COARE datasets, Marzano et al. (1995) found that the MIR 150- and 220-GHz channels could help with the retrieval of the ice content estimate at cloud top. However, since these channels are strongly dependent upon the ice scattering properties, a proper microphysical parameterization needs to be carried out. Bauer and Grody (1995) also noted that the SSM/I–SSM/T2 combination may assist in discriminating regions of snow cover from precipitation.
For this overpass, we present the degraded AMPR 10- and 85-GHz nadir data in Fig. 11, in a similar fashion to Fig. 5, except that the effect of along-track sampling is taken into account in the bottom panels (the along-track sampling is maintained at 0.7 km in the top panels). During this overpass the aircraft encountered several rainbands on its way towards the hurricane eye. As for the 4 February pass, the 85-GHz TB structure becomes smoothed in the top panels. The top panels of Fig. 12 depict the rain-rate retrievals for stages of degradation between 6 and 30 km and a constant along-track sampling distance of 0.7 km, whereas the bottom panels degrade both the sensor resolution and the along-track sampling. The finescale rainbands are lost with the larger sampling distance, although for degradations in both the footprint and along-track sampling distance, several of the rainfall maximums are maintained. The nonprecipitating ice (nonprecipitating cloud water is not added in) is an additional output of this type of algorithm, due to its reliance upon explicit categories of cloud-particle types and associated microphysics.
The retrievals of the precipitation-sized rain and graupel columnar contents are presented in Fig. 13. This series of panels depicts the differences that result when the ARMAR data is excluded from the tall vector of input observations. Although the rain structure follows the same general trend for both situations, the result when omitting the ARMAR profile is to underestimate the regions of small columnar graupel, while maintaining the ability to capture some of the peak values. Evidently, the addition of the ARMAR profile to the AMPR TB provides information on the weaker graupel structure that otherwise may not be captured even with the 85-GHz AMPR channel. This is consistent with the relatively high 85-GHz TB values (most values above 220 K) noted over the TOGA COARE storms by McGaughey et al. (1996).
The finescale microwave radiometric features of precipitating clouds were examined in this study, and these measured data provided an opportunity to examine the effect of sensor resolution upon the vertically profiling rainfall-retrieval algorithm developed by Marzano et al. (1995) and Marzano et al. (1996). While there were obvious discrepancies when trying to relate finescale microwave data from across-track scanning radiometers such as the AMPR to a conical scan geometry of the coarser-scale SSM/I, the fact that these two instruments operate at identical frequencies provided an opportunity to examine this complex issue with actual combined radar–radiometric data. Through direct assimilation of the radar reflectivity profile at several levels below the freezing level, the algorithm produces an output rain rate and radar path attenuation that is consistent with the observations. As the TRMM satellite will contain radar and radiometric instruments comparable in frequency to these aircraft instruments, several experiments were performed to coarsen the data to resolutions and along-track sampling distances typical of the TMI and the SSM/I instruments. Using these degraded-resolution datasets in the retrieval algorithm, the rain rateand columnar rain and graupel estimates at different resolutions were compared.
In general, when the data were coarsened beyond ≈ 15 km, the retrieved rain rate began to substantially diverge from the initial state (the initial state being the undegraded AMPR data). The mean trend was maintained, but much of the finescale variability was lost. The columnar graupel retrievals were affected more than the columnar rain contents by the degraded sensor resolution. This is suprisingly different than previous results on simulated data (Skofronik-Jackson et al. 1995). Columnar nonprecipitating ice, which is a quantity that is usually not distinguishable from precipitation-sized graupel, was affected less by the degraded sensor resolution than was the columnar graupel.
Within these tropical storms, 85-GHz ice scattering effects were generally weaker than those noticed over land-based storms (Vivekanandan et al. 1993; McGaughey et al. 1996), and the retrieved columnar graupel contents were only near 1–2 kg m−2. The 85-GHz background TB appeared radiometrically warm due to the high columnar water vapor and also any nonprecipitating cloud liquid water. Therefore rain profiling algorithms that rely upon cloud model-derived microphysics may need to properly take into account not only regions of nonprecipitating cloud liquid water above the freezing level, but also the profile of water vapor in and above the ice hydrometeors (Panegrossi et al. 1998). Scattering-based rainfall algorithms may need to be adapted or tuned to produce consistent rainfall amounts in the tropical regime.
We noted that the AMPR–ARMAR combination showed promising potential in two particular areas. The addition of the ARMAR profile was shown to improve rain vertical profiles near the ocean surface where the variable sea surface emissivity strongly influences the 10-GHz TB. Second, the addition of the ARMAR profile to the measurement vector also enhanced the columnar graupel content when compared to the AMPR-only retrievals. This is thought to be due to the relatively weak 85-GHz scattering-based TB depressions noted over these storms, and in these situations the retrieval algorithm capitalizes upon the availability of the radar profile to constrain the output graupel profile in precipitation structures with relatively little ice.
Looking ahead to the future, pending a successful deployment of the TRMM satellite, coordinated AMPR–EDOP datasets (Caylor et al. 1995) should be available in coordination with the TRMM radar and TMI data during the validation phase of the TRMM experiment. These combined datasets will greatly expand the capabilities available for refining tall-vector rain profiling algorithms during the TRMM era and beyond. Development of improved cloud model microphysical parameterizations and improvements to computationally efficient, accurate radiative transfer models that take cloud geometries into account may ultimately yield positive benefits to tall-vector rain profiling algorithms. There remains the issue of how to properly combine the radar profile with the radiometer channels in the inner workings of these algorithms, especially away from the nadir view. The fusion of the radar profile in its basic measured state (where the attenuation is handled internally with a forward model) maintains a consistent framework for both the passive and active measurements, but the overall algorithm must be able to function with or without the availability of radar data. The TRMM radar will have an on-earth resolution of ≈4 km, so that it will be providing much finer-scale data than the 10-GHz TMI resolution. When treating the beamfilling issue for the 10-GHz TRMM channel, one suggestion has been to use the surrounding radar beam geometry to obtain the larger-scale rain-rate variability (Kozu and Iguchi 1995). These issues are very critical to the ability of tall-vector algorithms to operate effectively within the TRMM era and beyond.
The support of the sponsor, the Office of Naval Research, Program Element 0602435N, is gratefully acknowledged. J. Turk also acknowledges visiting scientist support at the Istituto di Fisica dell’Atmosfera, Frascati, Italy, in 1994 and supportwhile at Colorado State University under NASA Grant NAG8-890. The authors thank Ms. Robbie Hood and Frank LaFontaine of NASA/Marshall Space Flight Center for providing the AMPR data, Dr. Jim Wang of NASA/Goddard Space Flight Center for providing the MIR data, and Dr. Steve Durden of the Jet Propulsion Laboratory for providing the ARMAR data and its access software. SSM/I data were obtained from the Marshall Space Flight Center Distributed Active Archive Center and processed using the TeraScan software developed by SeaSpace Corp., San Diego. Mr. Steve Bishop provided timely assistance in figure preparation. We greatly thank two anonymous reviewers for their suggestions.
Corresponding author address: Dr. F. Joseph Turk, Marine Meteorology Division, Naval Research Laboratory, 7 Grace Hopper Avenue, Monterey, CA 93943-5502.