1. Introduction
Precipitation systems in the midlatitude region consist of various systems such as extratropical cyclones, fronts, and typhoons. These precipitation systems play an important role in transporting heat and moisture from mid–high latitudes and polar regions and often develop over the Pacific and Atlantic Oceans. Thus, understanding three-dimensional precipitation structures can help us estimate quantitative interactions among dynamic, thermodynamic, and precipitation processes.
The Tropical Rainfall Measuring Mission (TRMM)—carrying the precipitation radar (PR), the TRMM Microwave Imager (TMI), and so on—mainly intends to observe tropical precipitation (Kummerow et al. 1998). Even so, the precipitation systems in subtropic and midlatitude regions have been investigated using TRMM data. Since the PR observes three-dimensional rain structure and performs quantitative measurements with fine spatial resolution, the vertical distributions of precipitation were analyzed. Fu and Liu (2003) described average rain profiles in midlatitude eastern Asia in warm seasons. They showed that the contribution of stratiform rain in the midlatitudes is larger than that in the tropics. Using frequency distributions of echo-top height from TRMM PR data, particularly over the Pacific Ocean, Kodama and Tamaoki (2002) suggested that shallow precipitation contributes to precipitation amounts over the midlatitude oceanic regions in winter. Yamamoto et al. (2006) investigated the vertical and horizontal distributions of precipitation over the Sea of Japan, the Yellow Sea, and the Pacific Ocean and found that the profiles differ greatly on whether they came from extratropical cyclones and front and cold outbreaks. However, the PR’s narrow swath (~220 km) make capturing a horizontal expanse of precipitation systems difficult.
A spaceborne microwave radiometer such as the Special Sensor Microwave Imager (SSM/I) and TMI can observe widespread precipitation systems with a relatively large observation swath (~700 km). After the appearance of the spaceborne microwave radiometer, many studies have proposed satellite retrieval algorithms to estimate precipitation (e.g., Wilheit et al. 1977; Petty 1994b; Aonashi et al. 1996). The TRMM TMI rain retrieval employs the Goddard profiling algorithm (GPROF), which is based on the Bayesian inversion approach using a database between cloud–rain profiles simulated by cloud resolving models and forward calculated microwave brightness temperatures (Tbs) (Kummerow et al. 2001). This implies that the accuracy of rain retrieval largely depends on the setting of the database. These algorithms have generally been tuned for tropical regions and are therefore less suitable for estimating the weak rain in the midlatitudes (Negri et al. 1995). Indeed, the amount of precipitation estimated by the PR (PR-rain) and TMI (TMI-rain) have systematic discrepancies. While TMI-rain is greater than PR-rain over the tropics (Kummerow et al. 2000) and over midlatitude regions in summer, PR-rain is greater than TMI-rain in the midlatitudes in winter (Ikai and Nakamura 2003). These discrepancies are explained based on precipitation water content and precipitation water path variations derived from the TMI profiling algorithm (TMI 2A12) and PR profiles (PR 2A25) (Masunaga et al. 2002), freezing-level assumptions in TMI 2A12, the relationships of the radar reflectivity factor to rain rate (Z–R) or attenuation to rain rate (k–R) in PR 2A25 (Ikai and Nakamura 2003), and echo-top height (Furuzawa and Nakamura 2005). Focusing on the western Pacific around Japan, Yamamoto et al. (2006) reported that the discrepancies between PR-rain and TMI-rain vary depending on precipitation patterns. Over regions with cold outbreaks, PR-rain is generally greater than TMI-rain partly because TMI likely misses solid precipitation. In contrast, around extratropical cyclones and fronts, TMI-rain is greater (smaller) than PR-rain in the case of weak (heavy) precipitation.
To eliminate these discrepancies and to understand microwave radiation from precipitation, the relationship between observed microwave signatures and precipitation structure has been statistically investigated. Petty (2001) performed an empirical orthogonal function (EOF) analysis of brightness temperatures in tropical stratiform rain areas inferred from the 19-GHz channel of the SSM/I and related the eigenvectors to precipitation system characteristics. Bauer (2001) also carried out an EOF analysis of brightness temperatures measured by the TMI with rain areas detected by the PR. They compared their results with those of some simulation models. From the global rain systems observed by the TRMM PR, Bauer (2001) concluded that the first EOF represented the major response to rain liquid water, and the second EOF could be interpreted as a sensitivity to cloud ice above opaque raincloud layers. Seo et al. (2007) examined TMI-rain and PR-rain rates for convective, mixed, and nonconvective rain categories. They found the difference of rain rate depends on the intensity and the type of rain. Fu and Liu (2001) investigated the relationship between microwave signatures and principal modes of rain-rate profiles in the tropics. They found that the vertical profiles derived from microwave signatures were similar to those derived from EOF analysis and suggested the importance of the vertical precipitation profile in determining upwelling microwave radiation. Most of these studies, however, only concentrated on tropical regions.
In the mid- and high-latitude regions, surface temperature occasionally falls below 0°C in winter. Therefore, the precipitation phase is very likely to be solid or freezing rain and the above-mentioned interpretation would not apply to midlatitude precipitation systems. Understanding the microwave properties of solid precipitation is crucial in estimating precipitation amounts in high latitudes by microwave radiometers. Some case studies have examined the microwave properties of solid precipitation, such as in a North Atlantic cyclone (Schols et al. 1999) and in cold outbreaks over the Sea of Japan (Lobl et al. 2007), using in situ ground and airborne radar observations. For instance, Aonashi et al. (2007) found a high-sensitivity solid precipitation at high-frequency channels (i.e., 85 and 89 GHz) and investigated relationships between microwave properties and observed liquid cloud water content. Therefore, the above-mentioned interpretation would not apply to midlatitude precipitation system.
The purpose of this paper is to have a better understanding of microwave signatures in midlatitude precipitation systems. First, we identify representative patterns of microwave signatures observed by the TMI in the case of precipitation using the EOF technique. These patterns are interpreted using the frequency distributions of precipitation and cloud systems and atmospheric conditions derived from simultaneous observation data by PR, Visible and Infrared Scanner (VIRS), and reanalysis data. Then, we use case studies to investigate the identified patterns.
2. Data
This study used TRMM Standard Products in version 6 provided by the National Aeronautics and Space Administration (NASA) and the Japan Aerospace Exploration Agency (JAXA). We also used Global Analysis (GANAL) data provided by the Japan Meteorological Agency (JMA). The analyzed region was over the Pacific Ocean at 30°–40°N and 115°E–180°, where discrepancies in precipitation amounts derived from PR and TMI are large in summer and winter (Ikai and Nakamura 2003). The land region was beyond the scope of this study because microwave properties over land differ greatly from those over ocean. The sampling periods were 2001/02–2003/04 December–February (DJF) and 2002–04 June–August (JJA).
The TMI is a conical scanning passive microwave radiometer with a 760-km swath width. The TMI has nine channels with vertical (V) and horizontal (H) polarization observing at 10.7, 19.35, 21.3, 37, and 85.5 GHz, but the 21.3-GHz channel has V only (hereafter abbreviated to 10V, 10H, 19V GHz, etc.). The spatial resolution of the TMI varies with frequency from 7.2 km at 85 GHz to 63.2 km at 10 GHz. We used Tb data from all channels of TMI 1B11 and the surface rain rate (RR-TMI) of TMI 2A12 data. The algorithm used has been described by Kummerow et al. (1996).
The PR is a 13.8-GHz radar that has a 245-km swath width (after a boost in 2001), with vertical and horizontal resolutions of 250 m and 5 km, respectively. From PR 2A25 data, the radar reflectivity factor (Z) with rain attenuation correction and rain rate (RR−PR) at every range bin and near the surface, defined as the lowest clutter-free bin (500 m at nadir), were used. Details of the algorithm for 2A25 data have been provided by Iguchi et al. (2000). From PR 2A23 data, we used the height of the storm top (StormH); height of the bright band (BBH), which was determined by the maximum Z in a specified range bin window (250 m); and the rain type. Rain in 2A23 is classified as “stratiform,” “convective,” and “others,” with each type expressed as three digits based on results from a vertical profile method (V method) and a horizontal pattern method (H method) (Awaka et al. 2009). The rain types in this study are linked with an aggregation of some rain types in the PR2A23 product (Table 1); stratiform with a bright band (BB) when BB is detected in the V method, stratiform without BB when BB is hardly detected in the V method, convective (convective categories except for shallow nonisolated is detected), shallow isolated rain, and “others.” The frequency of occurrence for the rest of rain types in PR2A23 was low. We summed up the PR pixels with each rain type in 0.25° grid.
Rain types in this study corresponding with those in PR2A23.
The VIRS measures radiance in five spectral bands (0.63, 1.61, 3.75, 10.8, and 12.0 μm) in the visible through the thermal infrared spectral regions. The spatial resolution of the VIRS is 2.1 km at nadir. We used the brightness temperature in the 10.8-μm band (Tbb) of 1B01.
We also utilized GANAL data to identify vertical atmospheric conditions along the TRMM data swath. GANAL data are 6-hourly global reanalysis data, provided on a 1.25° × 1.25° horizontal grid, with 18 vertical layers starting at the surface and delineated at standard pressure levels. The variables of GANAL are geopotential height, temperature, relative humidity, and wind vectors. The GANAL data at the nearest time to TRMM passes were linearly averaged in time and space. The freezing-level height calculated from GANAL (FreezH) was newly defined using geopotential height and temperature. The 2A23 product contains the height of the freezing level estimated from the monthly mean surface temperature data, but we did not use these data because the temperature fluctuation is not well expressed by the monthly mean sea surface temperature (SST) in regions where cold outbreaks occur (Yamamoto et al. 2006). In addition, cloud-top height (CloudH) was also estimated from the GANAL geopotential height and temperature and from Tbb.
The raw orbit data observed by PR, TMI, and VIRS were averaged over 0.25° × 0.25° in the selected region and period to have the same resolution except for StormH and BBH, which are conditionally averaged. GANAL data were linearly interpolated in time and space at the center of the grid and at the TRMM overpass. The samples from TRMM datasets are limited to the PR swath of the angle bins at nadir ±5 because the PR does not observe shallow rain systems (<2 km) at the edge of the swath because of surface clutter. All of the pixels with precipitation observed by PR were selected. The final number of samples for DJF (125 946) was approximately double that for JJA (66 022) because many weak and isolated precipitation features caused by cold outbreaks occur in winter. For the nine-channel TMI Tb data of each season (JJA and DJF), the EOF analysis was carried out to represent the maximum possible characteristics of midlatitudes precipitation from the original TMI Tb data. The EOF analysis was applied with a correlation matrix [i.e., the joint variance structure of the standardized variables 
Correspondence between percentiles and the first three component scores (CS1–3) in JJA and DJF.
Note that a microwave signature is affected not only by precipitation but also by surface roughness, surface and air temperature, moisture, wind, and so on. These may contaminate the results of EOF analyses in cases that insufficient rain liquid water exists on the scale of the sensor’s field of view (FOV), such as with an isolated precipitation system. To reduce the contamination, several microwave algorithms use attenuation and scattering indices (e.g., Petty 1994a; Seo et al. 2007). In this study, however, nine-channel data were used instead of those indices. The data filtering before applying EOF analysis could also be useful, but the filtering was not applied. These are because we are interested in the overall EOF characteristics of the microwave signatures from pixel at least partly containing rain over ocean in the midlatitude region. Thus, microwave signatures at pixels where PR detected rain were simply chosen. Application of scattering and attenuation indices or the filtering remains the next step for a future study.
3. Vertical distributions of precipitation for each EOF pattern
The EOF defines i principal vectors for i variables (i = 9 in this study, which means nine TMI channels). The proportions of every EOF for winter (DJF) and summer (JJA) are shown in Table 3. In a statistical view, only the first few EOFs with high proportions should be used for interpretations to the total variation in Tbs. However, we also described EOF3 despite its small proportion, because it suggests interesting characteristics.
Proportion (%) for nine EOFs in DJF and JJA.
a. EOF1
Almost the same positive elements for EOF1 in JJA values appeared at all channels except for 85 GHz, which had a negative value (the first column in Fig. 1). This distribution corresponded to the results for rain systems over the tropical and subtropical regions (Bauer 2001) but differed from those for stratiform rain systems in the tropics (Petty 2001), possibly because of the difference in sampling methods. From the contoured frequency altitude diagrams (CFADs) of Z for EOF1− (the second columns in Fig. 1a), high frequency of Z was less than 15 dBZ at every altitude. Most pixels for StormH and CloudH were distributed below 2 km. For those in EOF1+ (the third columns in Fig. 1a), on the contrary, high frequencies at Z in less than 4 km for EOF1+ exceeds 30 dBZ, and StormH and CloudH were concentrated around 5 and 8 km, respectively. Relative to FreezH, high (low) StormH appears in most of samples for EOF1+ (EOF1−). This difference between FreezH and StormH in EOF1+ implies many solid particles. The negative values for the 85-GHz Tbs and the structure of EOF1 also show scattering by solid particles above FreezH. For rain types for EOF1+ (the fourth column in Fig. 1a), high frequencies of stratiform and convective clouds appeared. In contrast, EOF1− dominates warm rain from the difference between FreezH and StormH. Rain types showed a large frequency of “others,” and shallow isolated rain corresponded to weak precipitation systems. For CloudH in the isolated rain systems, CloudH tends to underestimate from actual cloud height since averaged Tbb increases with sea surface temperature in a grid. Thus, the shallow–isolated precipitation would affect the distributions of higher StormH than CloudH because of the contamination from surface emission. These characteristics suggest that stronger and taller precipitation systems likely appear when CS1 is larger.
(left to right) Eigenvector of TMI nine-channel Tbs with percentiles of explained variance for EOF1 ; CFAD of Z derived from PR 2A25 (gray image, percentage at every altitude with 0.5-dBZ steps) and histograms of the CloudH (dashed line), StormH (solid line), FreezH (thick dash–dotted line), and BBH (solid line) for EOF1− and EOF1+; and frequency of rain-type flags for each categorized CS1: (a) JJA and (b) DJF. The CFADs indicate the percentages of frequency at every altitude, where the occurrence frequency for the sum of every altitude bin is >1% is shaded.
Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2539.1
The component values of the eigenvector for EOF1 in DJF (the first column in Fig. 1b) showed positive and similar magnitudes at all channels. This means that standardized Tb increases equally as CS1 becomes large. The Z profiles for EOF1− and EOF1+ in DJF (the second and third columns in Fig. 1b) were similar to those in JJA, but those for EOF1+ below FreezH were more variable. Most samples of StormH, CloudH, and FreezH for EOF1− existed below 2 km. These results indicate that shallow precipitation systems dominated for EOF1−. In contrast, EOF1+ had the peak frequency of the distribution at 3 km for StormH and FreezH and at 6 km for CloudH. In comparison with the vertical distributions for JJA, relatively small and weak precipitation systems existed. The positive values for 85-GHz Tbs in EOF1 suggest that there were fewer samples with solid hydrometeors causing scattering signals in DJF than in JJA. The frequency distributions of rain types (the fourth column in Fig. 1b) were generally similar to those in JJA, although “others” were more frequent in the positive CS1. The frequency of stratiform without BB did not largely change regardless of CS1, indicating that rain systems near the freezing height in PR2A23 (about 3 km) were dominant for DJF (Yamamoto et al. 2006).
b. EOF2
The first column in Fig. 2a shows the eigenvector for EOF2 in JJA. The positive (negative) values appeared at low- (high) frequency channels, which is similar to results found in tropical and global cases (Petty 2001; Bauer 2001). The CFADs of Z for EOF2− and EOF2+ (the second and third columns in Fig. 2a, respectively) resembled those for EOF1 but shifted to smaller and larger Z at each altitude. StormH values for EOF2− and EOF2+ were centered at 1.75 and 6 km, respectively. StormH was the most variable parameter for CS2. In contrast, EOF2− and EOF2+ had similar FreezH values. The relationships between StormH and FreezH showed that precipitation for EOF2− was likely to be warm (liquid) rain. For EOF2+, large amounts of solid hydrometeors may have existed above FreezH. Corresponding to these results, the frequencies of stratiform without BB and “others” dominated for EOF2− (the fourth column of Fig. 2a). As CS2 increased (i.e., to higher percentile), the frequency of stratiform with BB became higher. Particularly at EOF2+, the frequency of convective rain rapidly increased and shallow isolated precipitation systems also became low, although the frequencies were relatively small.
As in Fig. 1, but for EOF2.
Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2539.1
The eigenvector for EOF2 in DJF (the first column in Fig. 2b) showed a similar pattern [i.e., positive (negative) eigenvector at low- (high) frequency channels] to EOF2 in summer. CFADs of Z and StormH histograms for EOF2 in DJF were generally similar to those in JJA although Z profiles for EOF2+ (the third column in Fig. 2b) were more scattered, and StormHs for both EOF2− (the second column in Fig. 2b) and EOF2+ were lower (2 and 4 km, respectively) than in JJA. While FreezH for EOF2− indicated the peak frequency of the distribution at 3.5 km, that at the lower peak for EOF2+ was 500 m. The differences between StormH and FreezH suggest that solid hydrometeors tended to dominate for EOF2+ and that not only StormH, but also FreezH, largely varied with CS2. The rain-type frequencies (the fourth column in Fig. 2b) showed that stratiform with BB dominated and convective rain frequently occurred for EOF2+, while “other” and stratiform without BB mainly appeared for EOF2−. This finding suggests that for precipitation systems with solid hydrometeors, both stratiform and convective tend to be active for EOF2+.
The Tb histograms for the CS2 category for all channels were investigated to see detailed characteristics of Tb for EOF2. The histograms for EOF2+ had double peaks in JJA and DJF, particularly at higher frequencies (i.e., 37 and 85 GHz). Accordingly, EOF2 could be interpreted as consisting of two different effects with similar eigenvectors. Then, EOF2+ was divided into lower Tb (EOF2+low) and higher Tb (EOF2+high) by a threshold of Tb37V at 225 K for DJF and at 230 K for JJA. The Tb37V thresholds were determined to give the clearest Tb separation between EOF2+low and EOF2+high. Figure 3 shows the frequency distribution of Tbs. The Tb values at the peak frequency of the distribution for EOF2+high in all the channels were the highest for all categories both in JJA and DJF, except at 85 GHz. The polarization (H and V difference) became small at the high-frequency channels. These results indicate that large amounts of liquid (solid) hydrometeors for EOF2+high are due to emissions (scattering) at the low- (high) frequency channels. While small polarization at 85 GHz appeared for EOF2−, EOF2+low got large polarization. This difference suggests that relatively large emissions occurred for EOF2−. For EOF2+low, in contrast, polarization was large at all frequency channels, suggesting the penetration of emissions from sea surface. This implies the existence of few solid and liquid hydrometeors and/or of isolated precipitation systems.
Histograms of brightness temperature for EOF2− (dotted line), EOF2+ <225 K for the V channel of 37 GHz (EOF2+low, light solid line), and that ≥225 K for the V channel of 37 GHz (EOF2+high, dark solid line) in DJF: (top to bottom) 10–85 GHz.
Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2539.1
Figure 4 shows the frequency distributions of the Z profile, StormH, FreezH, CloudH, and BBH for EOF2+low and EOF2+high. For EOF2+low both in JJA and DJF (Figs. 4a,c), weak Z (~15 dBZ) appeared regardless of height. Most pixels for CloudH were distributed at less than 1 km. This low CloudH may have been caused by contamination from SST since precipitation systems may be small relative to the horizontal resolution of the TMI. This is supported by the large polarization difference between 85V and 85H for EOF2+low (Fig. 3). The high frequency of StormH was centered around 1 and 3 km in JJA and DJF, respectively. These frequency distributions for EOF2+low are apparently similar to those for EOF2− (the second column of Fig. 2). However, while some samples in which StormH was higher than FreezH can be seen for EOF2+low, there are few such samples for EOF2−. This tendency is more significant for the DJF cases. Moreover, for EOF2+low in DJF (Fig. 4c), most pixels in FreezH were less than 2 km and were lower than StormH, and StormH was generally low compared with FreezH for EOF2−. The relationship between FreezH and StormH suggests that EOF2+low (EOF2−) consisted of solid (warm) precipitation particles. For EOF2+high (Figs. 4b,d), Z below FreezH and BBH was centered around 35 dBZ. Peak frequencies of distribution for StormH in JJA and DJF were present around 6 and 4.5 km, respectively. From these distributions, EOF2+high could represent tall systems with heavy precipitation.
The CFAD of Z derived from PR2A25 (gray image, percentage at every altitude with 0.5-dBZ steps) and histograms of the CloudH (dotted line), StormH (solid line), FreezH (thick dash–dotted line), and BBH (solid line) for EOF2+ (a) <230 K of Tb37V (EOF2+low) and (b) those ≥230 K of Tb37V (EOF2+high) in JJA, and those for (c) EOF2+low and (d) EOF2+high in DJF.
Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2539.1
c. Relationship between CS and precipitation amount
Figure 5 shows the relationship between CS and near surface RR–PR for EOF1 and EOF2. For EOF1, significant positive correlation appeared above the thresholds of 0.2 mm h−1 in JJA (Fig. 5a) and 1 mm h−1 in DJF (Fig. 5b). Similar distributions between CS1 and vertically integrated RR–PR regardless of the phase (total, liquid, and solid) were also found (not shown). These results and the positive eigenvector of EOF1 (Fig. 1) suggest that as surface RR-PR increases, the emission signatures at all channels strengthen above the thresholds. In cases in JJA, however, solid hydrometeors also caused scattering at 85 GHz. These findings are consistent with the interpretations of Bauer (2001) and Petty (2001) for surface rain. In contrast, near surface RR-PR less than the thresholds did not contribute to significant emissions (Tb increase). This shows that it is difficult to estimate surface RR less than 0.2 (1) mm h−1 in JJA (DJF) from CS1 in 0.25° average, possibly because the RR-PR less than the thresholds is too small to fill a horizontal TMI pixel with precipitation. Particularly in DJF, more shallow and isolated precipitation systems may have existed, contributing to the threshold difference between JJA and DJF.
Scatter diagram of the component score and near-surface rain rates retrieved from PR in (a) EOF1 JJA, (b) EOF1 DJF, (c) EOF2 JJA, and (d) EOF2 DJF. The frequency is taken at every rain rate.
Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2539.1
For EOF2, positive correlations (although less significant than for EOF1) also appeared at more than 1 mm h−1 both in JJA (Fig. 5c) and DJF (Fig. 5d). As mentioned in regard to the CFADs of Z and histograms of StormH and FreezH in EOF2 (the second and third columns of Fig. 2), CS2 was likely sensitive to solid hydrometeors. In general, solid particles increase with the development of a precipitation system because the system becomes tall. These results suggest that solid particles at more than 1 mm h−1 in surface RR-PR caused scattering at higher-frequency channels, while rain systems of less than 1 mm h−1 would be warm rain or have few solid hydrometeors with scattering.
To compare the amount of solid and liquid hydrometeors among EOF2+high, EOF2+low, and EOF2−, RR-PR above and under the FreezH (ΣRRsol and ΣRRliq, respectively) were vertically accumulated (Fig. 6). Since PR has low sensitivity to small ice particles because of 13.8 GHz of frequency (Masunaga et al. 2002), PR echoes above FreezH would be supercooled, melting, or particularly large frozen hydrometeors. Then, the large value of ΣRRsol should be expected to substantial solid hydrometeors. For EOF2+high, ΣRRliq in JJA and DJF (Figs. 6a,b) was distributed higher than 10 mm h−1 (5 mm h−1) with nearly the same amount of ΣRRsol. This corresponds to the results from the CFADs and histograms for EOF2+high (Figs. 4a,b). In comparison with the distributions between EOF2+low (Figs. 6c,d) and EOF2− (Figs. 6e,f), relatively large contributions of ΣRRsol (ΣRRliq) to total ΣRR occurred for EOF2+low (EOF2−). As mentioned in the previous subsection, the CFADs and StormH for EOF2+low (Figs. 4c,d) were similar to those for EOF2− (Fig. 2); furthermore, the maximum frequency of StormH was generally higher than that of FreezH for EOF2+low, and vice versa for EOF2− although the frequency of ΣRRliq frequency showed a seasonal difference. These differences in the precipitation systems for EOF2+low and EOF2− could be characterized by the strength of the liquid precipitation rate in JJA and by the phase of precipitation in DJF.
Scatter diagrams of the vertically accumulated liquid rain rate ΣRRliq and solid rain rate ΣRRsol derived from PR and GANAL for (a),(b) EOF2+high, (c),(d) EOF2+low, and (e),(f) EOF2− in (left) JJA and (right) DJF.
Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2539.1
In summary, EOF2 represented the amount of solid water content, although liquid water content also largely affected CS2 in JJA. In conjunction with the frequency distributions of rain types for these categories, EOF2−, EOF2+low, and EOF2+high were suggested to be warm rain, isolated solid precipitation particularly in the DJF case, and highly developed precipitation, respectively.
d. EOF3
Despite its small proportion, we also describe EOF3 because it suggested interesting characteristics. The eigenvector for EOF3 in JJA (the first column in Fig. 7a) showed positive (negative) values at 10 and 85 GHz (37 GHz). For EOF3− (the second column in Fig. 7a), near surface Z was distributed around 25 dBZ, and almost the same Z appeared below FreezH (4 km). StormH was also centered at 4.25 km, corresponding to FreezH. Below 500 m from FreezH, a significant BB appeared in the CFAD, corresponding to the histograms of BBH. For EOF3+ (the third column in Fig. 7a), there were many cases of heavy precipitation indicating more than 30 dBZ, although large variations existed. Most of the pixels of StormH were higher than those of FreezH. These results indicate that tall precipitation systems with many solid hydrometeors dominated for EOF3+. From the rain-type distributions (the fourth column in Fig. 7a), a high frequency of stratiform appeared for both EOF3− and EOF3+. In contrast, “others” and shallow isolated precipitation systems peaked from 70th to 85th percentile (from +0.5 to +1.0) in CS3. The frequency of convective precipitation increased for EOF3+ to approximately 10%.
As in Fig. 1, but for EOF3.
Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2539.1
The first column in Fig. 7b shows the eigenvector for EOF3 in DJF. While the lowest- and highest-frequency channels (10 and 85 GHz) showed positive values, those at the other channels (19, 23, and 37 GHz) were negative. In comparison with the CFAD for EOF3− in JJA (the second column in Fig. 7a), mean Z increased toward the surface with a less significant BB in DJF (the second column in Fig. 7b). The CFAD of Z for EOF3+ (the third column in Fig. 7b) distributed mostly less than 20 dBZ in every altitude; however, Z near the surface for a few percent of samples exceeded 30 dBZ. StormH at the peak frequency of distribution for EOF3+ (2 km) was lower than that for EOF3− (3.5 km). The other parameters such as FreezH and BBH showed no large differences between EOF3− and EOF3+. The rain-type distributions (the fourth column in Fig. 7b) indicated that stratiform with BB (“others”) dominated for EOF3− (EOF3+). Almost the same frequencies in stratiform without BB and convective systems appeared regardless of CS3. Shallow rain systems only appeared in positive or small negative CS3.
For EOF3, histograms of Tb for EOF3+ had double peaks as well as EOF2 (Fig. 8). Those in DJF did not show double peak but large variations with higher Tb at all frequency channels. We also divided EOF3+ into EOF3+low and EOF3+high with thresholds of 250 (240) K in JJA (DJF) at 37V GHz. Regardless of the seasons, Tb of all channels at the peak frequency of distribution for EOF3+high was the highest of all categories except for 85 GHz. Particularly at 10 GHz, Tb for EOF3+high was higher than that for EOF3− by about 15–30 K. There was little polarization difference at 19 GHz for EOF3+high, while clear difference was shown between EOF3− and EOF3+low. Those large emissions at 10 and 19 GHz for EOF3+high suggest large amounts of hydrometeors. At the higher-frequency channels, the frequency distributions for EOF3+high and EOF3− did not show apparent difference.
As in Fig. 3, but for EOF3 and a threshold of 250 K.
Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2539.1
Figure 9 shows the frequency distributions of Z, StormH, CloudH, FreezH, and BBH for EOF3+low and EOF3+high in JJA and DJF. For EOF3+low in JJA and DJF (Figs. 9a,b), Z was centered around 15 dBZ with StormH of 4.5 and 2 km. There were no large differences of EOF3+low from EOF1− and EOF2−. For EOF3+high (Figs. 9b,d), the high frequency of Z exceeded 30 dBZ near the surface with StormH values of 6 and 3.75 km; this result indicates highly developed precipitation systems. Focusing on the distributions for EOF3+high and EOF3− (the second column in Fig. 7), some differences could be found. The Z at the near surface for EOF3+high was higher than that for EOF3− by 10 dBZ both in JJA and DJF. In the JJA case, most StormH was above FreezH, and there was no clear BB for EOF3+high. For EOF3−, the peak frequency of distribution for StormH corresponded to that of FreezH, and significant BBs were present in the Z profiles. The characteristics of the vertical structures of EOF3+high and EOF3− correspond to typical convective and stratiform precipitation types, respectively (Houze 1993). Vertical structures of Z were very similar to the mean vertical structures of convective and stratiform rain rate over the midlatitude oceanic region (Figs. 5 and 3 in Fu and Liu 2003). Hong et al. (1999) developed a scheme that classifies convective and stratiform areas using microwave Tb. The scheme is based on strong emission (scattering) at the center pixel surrounded by pixels at 19 and 37 GHz (85 GHz). Our Tb histogram result, showing higher Tb for EOF3+high than for EOF3− at lower channels, agrees with their results. In comparison with the JJA case, determining a significant BB from vertical structures of Z in the DJF cases was difficult because of weak rain; precipitation systems in winter are not as heavy as those in summer. However, the characteristics in DJF may be common to those in JJA. We will confirm this suggestion with case studies in a later section.
As in Fig. 4, but for EOF3 and a threshold of 250 K in JJA and 240 K in DJF.
Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2539.1
To summarize this section, Table 4 presents the interpretations of the representative patterns for the first three EOFs. EOF1 represents the rain liquid water. There are good correlations between CS1 and the near surface rain rate. The eigenvector for the 85.5-GHz channel suggests that summer precipitation appears to be more solid particles in developing precipitation systems than winter precipitation. EOF2 represents the number of solid particles. EOF2+ consists of two kinds of precipitation system. One is highly developed precipitation system (EOF2+high), which contains both liquid and solid particles from high emissions at the 37-GHz channel. The other is isolated and/or solid precipitation (EOF2+low) from weak emissions at all channels. The CFADs in EOF2+low suggest that the summer precipitation likely appears to be an isolated system, while the winter precipitation dominates as shallow solid precipitation systems. In contrast, EOF2− shows shallow and weak precipitation systems with warm rain. EOF3 is related to rain type. The CFADs in EOF3+high and EOF3− suggest convective and stratiform precipitation system, respectively. An EOF3+low likely appears around the coastal region (see the next section). Although the microwave signatures of midlatitude precipitation systems in summer seasons are similar to cases in tropical or subtropical warm season (Petty 2001; Bauer 2001), dominant precipitation patterns in the midlatitude in winter are less distinct because weak, shallow, and isolated precipitation systems frequently appear particularly in winter season.
Interpretation of representative EOF patterns.
4. Classification examples
As an application of the patterns suggested above, horizontal distributions of CSs were compared with meteorological parameters in the following cases: 1) the baiu front and an extratropical cyclone on 21 June 2002 (Figs. 10 and 11) and 2) an extratropical cyclone and cold outbreak on 10 February 2002 (Figs. 12 and 13). CSs were calculated for all the TMI pixels including those over the no-rain pixels and outside the PR swath because one of our goals was to see the performance of the automatic detection of midlatitude precipitation systems characteristics using TMI data. Rain areas were determined from pixels observed by both PR and TMI. Land regions were not included because our EOFs were obtained only over the ocean.
(a) Sea level pressure (hPa; solid lines every 4 hPa) and temperature at 1000 hPa (°C; dashed lines every 3°C) derived from GANAL data and the surface rain rate derived from TRMM TMI (mm h−1; colors), (b) TRMM VIRS IR image (K; gray shades), (c) surface rain rate derived from TRMM PR (mm h−1; colors), and (d) rain type derived from PR 2A23 at orbit number 26 230 (21 Jun 2002).
Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2539.1
Horizontal distributions of (a) CS1, (b) CS2, and (c) CS3 (percentiles) derived from TRMM TMI for the same orbit of Fig. 10.
Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2539.1
As in Fig. 10, but for orbit numbers 24 184 and 24 185 (10 Feb 2002).
Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2539.1
As in Fig. 11, but for the same orbit of Fig. 12.
Citation: Journal of Applied Meteorology and Climatology 50, 6; 10.1175/2010JAMC2539.1
a. Baiu front and extratropical cyclone
Figure 10a shows the distributions of the near surface pressure field and RR-TMI on 21 June 2002. An extratropical cyclone existed around 38°N and 150°E and an associated front around 35°N connected with the baiu front around 30°N and 140°E. Precipitation appeared over the baiu front and rainbands were present over 30°N from 130° to 140°E and over 32.5°N from 150° to 160°E. According to the Tbb distribution from VIRS (Fig. 10b), high clouds extended over these regions. RR-PR and rain-type distributions (Figs. 10c and 10d, respectively) also showed significant convective rainbands with heavy precipitation and stratiform rain extending around the bands.
The distributions for CS1 (Fig. 11a) interpreted as rain liquid water correlated with those for RR-PR (Fig. 10c). Heavier RR-PR led to larger CS1. Thus, positive CS1 generally indicates rain areas. Large positive CS2 (EOF2+, red colors in Fig. 11b), which implies many solid particles, also corresponded to rainbands over the baiu front with heavy RR-PR and high clouds. In contrast, large negative CS2 (EOF2−, blue colors) was present in regions of relatively weak RR-PR and low cloud. For example, in the area with heavy rainfall over 130°–140°E, corresponding to relatively heavy precipitation more than 4 mm h−1 with high cloud, EOF2+ appeared widely. These regions extended around 140°E. To the east of 140°E, RR-PR rapidly weakened and CloudH became lower. These characteristics agree with EOF2−. EOF2+low (green color), interpreted as shallow precipitation with solid particles, was widely distributed over the region without rain. This may indicate that PR detected weak noiselike precipitation from the sea surface. For CS3 (Fig. 11c), we again focused on the area with heavy rainfall over 130°–140°E. A heavy convective precipitation band parallel to 30°N around 130°E observed by the PR well correlated with EOF3+high (red color). CS3 rapidly decreased below −1.0 (EOF3−, blue color) over the south of the convective band. At 142.5°E, the areas with weak stratiform rain were close to those of EOF3−. The RR-TMI and Tbb implied a relatively moderate rain system with high clouds, which may have consisted of stratiform rain, in conjunction with the interpretation of EOF3−. The distributions of EOF3+high and EOF3− were similar to that of the ratio of convective RR-TMI to total RR-TMI in 2A12, corresponding to the convective-area fraction defined by Hong et al. (1999) (not shown). EOF3+low (green color) was distributed along the coastal areas; this may have been due to contamination from land surface emissions.
b. Extratropical cyclone and cold outbreak
According to the surface pressure and temperature field (Fig. 12a), an extratropical cyclone was present at 33°N and 150°E with a warm front extending to the east and a cold front to the southwest. Cold advection was brought to the rear of the extratropical cyclone and the cold front, causing precipitation near the west coast of Kyushu, Japan. Heavy precipitation with high clouds (Fig. 12b) was located near the center of the extratropical cyclone and the cold front. Near that area, some convective lines were clearly visible, where RR-PR showed rain of more than 16 mm h−1 (Fig. 12c). Stratiform rain systems widely spread around the convective lines (Fig. 12d).
Figure 13 shows the distributions for CS(1–3). The variations for CS1 (Fig. 13a) corresponded well to RR-PR and RR-TMI, as in the previous case. CS1 showed an increase over the center of the extratropical cyclone and the front. For solid precipitation, CS1 was negative. Even for the western coast in Fukuoka, where relatively heavy precipitation occurs, CS1 was approximately more than 70th percentile. Relatively small CS1 values (30th < CS1 < 70th percentile) appeared widely south of Japan, despite the absence of RR-PR. These regions were outside of the EOF analysis area (our analyses covered an area above 30°N), but relatively large rain liquid water was inferred from the interpretation of CS1. For CS2 (Fig. 13b), EOF2+high was found in the northern part of the extratropical cyclone and the warm and cold fronts. These regions also showed relatively heavy RR-PR (>4 mm h−1) and high clouds. This may indicate a highly developed precipitation system with many solid hydrometeors. Over the area of precipitation derived from cold outbreaks such as southern Japan around 30°N, small negative CS2s appeared. This precipitation, including liquid hydrometeors, may have been caused by relatively high air temperatures (6°–9°C). EOF2+high was widely distributed over the cloud-free areas with relatively low SSTs. EOF2− was widely spread over the southern part of the warm and cold fronts. In comparison with the RR-PR and rain types, EOF2+high and EOF2− could indicate developed convective systems and relatively small stratiform systems, respectively. For CS3 (Fig. 13c), a significant band of EOF3+high extended along the cold and warm fronts. In comparing Fig. 13c with Fig. 12d, it is seen that these bands correlated well with convective precipitation. In contrast, EOF3− delineated the southern part of the front and spread to the northern part of the extratropical cyclone. The regions where the high ratio of convective RR-TMI to total RR-TMI in 2A12 appeared correspond to those in EOF3+high (not shown), although the ratio was relatively small, maybe because of weaker systems when compared with precipitation systems in JJA. EOF3+low was found only in the coastal area in both the summer and winter cases.
5. Summary
Microwave properties in midlatitude precipitation systems in summer and winter were analyzed using the EOF technique. Major EOFs were detected and interpreted using the frequency distributions of Z profiles and meteorological parameters. The results are summarized as follows:
The eigenvector for EOF1 indicated positive values at almost all channels. This shows that Tbs rise with CS1 because of emission from liquid water. As CS1 increases, Z below FreezH is larger and StormH, CloudH, and the frequency of stratiform and convective rain become high. While almost the same amplitude of eigenvectors appeared for all frequencies in winter, only the 85-GHz channels in summer had small negative values. The negative eigenvalues at 85 GHz in summer may have been caused by more solid particles formed by the development of precipitation systems. CS1 correlated with near-surface RR-PR more than 0.2 and 1 mm h−1 in the summer and winter cases, respectively. These thresholds imply a minimum rain rate when the whole TMI FOV is covered by rain. The case studies showed that the CS1 distributions corresponded well to RR-PR both summer and winter.
The eigenvectors for EOF2 were positive (negative) at the lower- (higher) frequency channels. The Tb at 85 GHz fell when high CS2 appeared, in relation to increased scattering due to solid particles. As well as EOF1, near surface Z, StormH, and CloudH become high as CS2 increased. At the same time, the difference between FreezH and StormH and the frequency of convective rain increased. In addition, CS2 also correlated with RR-PR >1 mm h−1 both in summer and winter. This would have corresponded to an increase of solid particles by the development of precipitation systems. The CS2 variations were largely affected by StormH in summer and by StormH and FreezH in winter. In the case of EOF2+, Tbs showed double peaks. By separating the vertical profiles and the parameters using thresholds of the 37V channel, EOF2+ was found to consist of two patterns: one is characterized by shallow systems with weak precipitation having Tb lower than the threshold (EOF2+low), and the other is by tall systems with heavy precipitation having Tb higher than the threshold (EOF2+high). Particularly in winter, EOF2+low and EOF2+high corresponded to shallow precipitation systems caused by a cold outbreak and heavy precipitation systems, respectively. In contrast, EOF2− showed the characteristics of shallow warm rain systems.
EOF3 explained only a small proportion, but it suggested interesting characteristics. Significant positive values were found at the lowest and highest (10 and 85 GHz) channels, while the other channels indicated small positive or negative values. For JJA case, EOF3− generally showed relatively moderate (20–30 dBZ) precipitation with StormH of 4 km. A significant brightband structure was seen in the vertical Z distributions in this pattern. Similar to EOF2+, EOF3+ is consisted of two patterns according to Tb distributions: one was shallow and weak precipitation (EOF3+low), and the other was heavy convective precipitation (EOF3+high). The vertical distributions of EOF3+high did not show significant brightband structures. These results resembled characteristics in winter, although the differences of structure were relatively insignificant because of generally weaker precipitation in winter compared to summer. In the case studies, EOF3+high mainly appeared over rainbands with convective precipitation systems and fronts, and EOF3− was widely distributed surrounding areas of EOF3+high. These results suggest to us that EOF3 represents rain types, with EOF3+high (EOF3−) implying convective (stratiform) precipitation systems.
This study has related EOF patterns to characteristics of midlatitude precipitation–cloud systems in the summer and winter seasons. For the first three EOFs, the midlatitude summer precipitation patterns were similar to those of tropic and global cases. On the other hand, the EOFs and vertical distributions of precipitation had less significant difference between EOF+ and EOF− because of the generally weaker precipitation in winter cases. The results are useful for detecting characteristic patterns in precipitation systems using only TMI data. By combining TMI data with data from other spaceborne microwave imagers, such as Advanced Microwave Scanning Radiometer for Earth Observing System (EOS) (AMSR-E) on board the Aqua satellite, we could obtain more frequent images and extend our study to the development process of precipitation systems. Moreover, the proposed Global Precipitation Measurement (GPM) core satellite and constellation polar-orbit satellites equipped with microwave radiometers will be very helpful instruments in the post-TRMM era. We plan to conduct similar and expanded studies at higher latitudes to better understand the formation and development mechanisms of precipitation systems.
Acknowledgments
TRMM products were provided by NASA and JAXA. The authors express their thanks to members of the Satellite Meteorology Laboratory of the Hydrospheric Atmospheric Research Center (HyARC) and the 21st Century Center for Excellence (COE) Program “Dynamics of the Sun-Earth-Life Interactive System (SELIS),” Nagoya University, for useful discussions. Dr. D. Short kindly checked the manuscript. This study was partly supported by JAXA, SELIS, and the “Formation of a virtual laboratory for diagnosing the earth’s climate system” program of the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan.
REFERENCES
Aonashi, K., A. Shibata, and G. Liu, 1996: An over-ocean precipitation retrieval using SSM/I multichannel brightness temperatures. J. Meteor. Soc. Japan, 74, 617–637.
Aonashi, K., T. Koike, K. Muramoto, K. Imaoka, and N. Takahashi, 2007: Physical validation of microwave properties of winter precipitation over Sea of Japan. IEEE Trans. Geosci. Remote Sens., 45, 2247–2258.
Awaka, J., T. Iguchi, and K. Okamoto, 2009: TRMM PR Standard Algorithm 2A23 and its performance on bright band detection. J. Meteor. Soc. Japan, 87A, 31–52.
Bauer, P., 2001: Over-ocean rainfall retrieval from multisensor data of the Tropical Rainfall Measuring Mission. Part I: Design and evaluation of inversion databases. J. Atmos. Oceanic Technol., 18, 1315–1330.
Fu, Y., and G. Liu, 2001: The variability of tropical precipitation profiles and its impact on microwave brightness temperature as inferred from TRMM data. J. Appl. Meteor., 40, 2130–2143.
Fu, Y., and G. Liu, 2003: Precipitation characteristics in mid-latitude east Asia as observed by TRMM PR and TMI. J. Meteor. Soc. Japan, 81, 1353–1369.
Furuzawa, F. A., and K. Nakamura, 2005: Differences of rainfall estimates over land by Tropical Rainfall Measuring Mission (TRMM) precipitation radar (PR) and Microwave Imager (TMI)—Dependence on storm height. J. Appl. Meteor., 44, 367–382.
Hong, Y., C. D. Kummerow, and W. S. Olson, 1999: Separation of convective and stratiform precipitation using microwave brightness temperature. J. Appl. Meteor., 38, 1195–1213.
Houze, R. A., 1993: Cloud Dynamics. Academic Press, 573 pp.
Iguchi, T., T. Kozu, R. Meneghini, J. Awaka, and K. Okamoto, 2000: Rain-profiling algorithm for the TRMM precipitation radar. J. Appl. Meteor., 39, 2038–2052.
Ikai, J., and K. Nakamura, 2003: Comparison of rain rates over the ocean derived from TRMM Microwave Imager and precipitation radar. J. Atmos. Oceanic Technol., 20, 1709–1726.
Kodama, Y.-M., and A. Tamaoki, 2002: A re-examination of precipitation activity in the subtropics and the mid-latitudes based on satellite-derived data. J. Meteor. Soc. Japan, 80, 1261–1278.
Kummerow, C., W. S. Olson, and L. Giglio, 1996: A simplified scheme for obtaining precipitation and vertical hydrometeor profiles from passive microwave sensors. IEEE Trans. Geosci. Remote Sens., 34, 1213–1232.
Kummerow, C., W. Barnes, T. Kozu, J. Shiue, and J. Simpson, 1998: The Tropical Rainfall Measuring Mission (TRMM) sensor package. J. Atmos. Oceanic Technol., 15, 809–817.
Kummerow, C., and Coauthors, 2000: The status of the Tropical Rainfall Measuring Mission (TRMM) after two years in orbit. J. Appl. Meteor., 39, 1965–1982.
Kummerow, C., and Coauthors, 2001: The evolution of the Goddard profiling algorithm (GPROF) for rainfall estimation from passive microwave sensors. J. Appl. Meteor., 40, 1801–1820.
Lobl, E. S., K. Aonashi, B. Griffith, C. Kummerow, G. Liu, M. Murakami, and T. Wilheit, 2007: Wakasa Bay: An AMSR precipitation validation campaign. Bull. Amer. Meteor. Soc., 88, 551–558.
Masunaga, H., T. Iguchi, R. Oki, and M. Kachi, 2002: Comparison of rainfall products derived from TRMM Microwave Imager and precipitation radar. J. Appl. Meteor., 41, 849–862.
Negri, A. J., E. J. Nelkin, R. F. Adler, G. J. Huffman, and C. Kummerow, 1995: Evaluation of passive microwave precipitation algorithms in wintertime midlatitude situations. J. Atmos. Oceanic Technol., 12, 20–32.
Petty, G. W., 1994a: Physical retrievals of over-ocean rain rate from multichannel microwave imagery. Part I: Theoretical characteristics of normalized polarization and scattering indices. Meteor. Atmos. Phys., 54, 79–99.
Petty, G. W., 1994b: Physical retrievals of over-ocean rain rate from multichannel microwave imagery. Part II: Algorithm implementation. Meteor. Atmos. Phys., 54, 101–121.
Petty, G. W., 2001: Physical and microwave radiative properties of precipitation clouds. Part I: Component analysis of observed multichannel microwave radiances in tropical stratiform rainfall. J. Appl. Meteor., 40, 2105–2114.
Schols, J. L., J. A. Weinman, G. D. Alexander, R. E. Stewart, L. J. Angus, and A. C. L. Lee, 1999: Microwave properties of frozen precipitation around a North Atlantic cyclone. J. Appl. Meteor., 38, 29–43.
Seo, E.-K., B.-J. Sohn, and G. Liu, 2007: How TRMM precipitation radar and Microwave Imager retrieved rain rates differ. Geophys. Res. Lett., 34, L24803, doi:10.1029/2007GL032331.
Wilheit, T. T., A. T. C. Chang, M. S. V. Rao, E. B. Rodgers, and J. S. Theon, 1977: A satellite technique for quantitatively mapping rainfall rates over the oceans. J. Appl. Meteor., 16, 551–560.
Wilks, D. S., 2006: Principal component (EOF) analysis. Statistical Methods in the Atmospheric Sciences, 2nd ed. Academic Press, 463–508.
Yamamoto, M. K., A. Higuchi, and K. Nakamura, 2006: Vertical and horizontal structure of winter precipitation systems over the western Pacific around Japan using TRMM data. J. Geophys. Res., 111, D13108, doi:10.1029/2005JD006412.
