In contrast to the view that deep convection causes heavy rainfall, Tropical Rainfall Measuring Mission (TRMM) measurements demonstrate that heavy rainfall (ranging from moderate to extreme rain rate) over the Korean peninsula is associated more with low-level clouds (referred to as warm-type clouds in this study) than with conventional deep convective clouds (cold-type clouds). Moreover, it is noted that the low-level warm-type clouds producing heavy rainfall over Korea appear to be closely linked to the atmospheric river, which can form a channel that transports water vapor across the Korean peninsula along the northwestern periphery of the North Pacific high. Much water vapor is transported through the channel and converges on the Korean peninsula when warm-type heavy rain occurs there. It may be possible to produce abundant liquid water owing to the excess of water vapor; this could increase the rate and extent of raindrop growth, primarily below the melting layer, causing heavy rain when these drops fall to the surface. The occurrence of heavy rainfall (also exhibited as medium-depth convection in radar observations over Okinawa, Japan) due to such liquid-water-rich lower warm clouds should induce difficulties in retrieving rainfall from space owing to the lack of scattering-inducing ice crystals over land and the warmer cloud tops. An understanding of the microphysical processes involved in the production of warm-type rain appears to be a prerequisite for better rain retrieval from space and rain forecasting in this wet region.
Despite significant interannual variations, summer rainfall over the Korean peninsula is associated primarily with synoptic-scale disturbances, monsoon frontal systems, and typhoon activity (Park et al. 1986, 1989; Lee 1991; Lee et al. 1998; Hwang and Lee 1993; Hong 1992; Lee and Kim 2007; Chu et al. 2012). In particular, heavy rain events are largely responsible for the great losses of lives and property that occur throughout Korea in association with such systems. Therefore, a better understanding of these heavy rain systems, incorporating cloud-related structure, satellite rainfall monitoring, and the relationship of these events to large-scale synoptic forcing, is important for better weather forecasting, which could reduce loss of life and the cost of recovering from damage.
In this context, recent advances in satellite technology have had a profound positive influence on forecasting capability, because spaceborne measurements can help considerably to understand the cloud–rain system (from the perspective of cloud microphysics) and a possible link between the cloud–rain system and large-scale synoptic features. Measurement of precipitation from space has received increasing attention from weather–climate-related scientists because of its potential use in understanding rain-forming cloud systems and cloud microphysical structures, which in turn can help improve weather–climate monitoring capabilities. For example, the Tropical Rainfall Measuring Mission (TRMM; Simpson et al. 1988) has provided not only an improved understanding of rain-forming physics within the cloud layer, but also insight into the tropical circulation system (see Kummerow et al. 1998; Takayabu 2002; Liu et al. 2008, among many others).
Capitalizing on recent advances in satellite technology, we expect successful rainfall measurements to be used for better understanding of cloud–rain physics. Despite such expectations, satellite-derived rain estimates over the Korean peninsula region appear to be less satisfactory in its accuracy. Recently, Sohn et al. (2010) compared four satellite-derived rainfall products (so-called high-resolution precipitation products; Sapiano and Arkin 2009) with surface rain gauge measurements over the Korean peninsula and microwave-based TRMM Microwave Imager (TMI) products for the summer period over multiple years, and concluded that all four satellite-based rainfall amounts were underestimated substantially over the Korean peninsula area. It was proposed that microwave-based rain retrievals were underestimated; thus, other infrared (IR)-based or IR-microwave (MW) blended rainfall products may also be underestimated owing to the calibration of IR-based rain estimates against the microwave-based estimates.
The underestimates of microwave-based rain rate over the Korean peninsula have been examined using surface rain gauge measurements collocated with TRMM TMI 85-GHz brightness temperatures (Ryu et al. 2012). Surprisingly, it was found that heavy rains could occur for clouds with 85-GHz brightness temperatures (Tbs) much warmer than those in the National Aeronautics and Space Administration (NASA) Goddard Profiling Algorithm (GRPOF) version 6 algorithm (Olson et al. 2006) for similar rain rates (Ryu et al. 2012). Because scattered signatures of microwave radiation at 85 GHz are used for rain estimation over land and scattering is caused largely by ice particles in the upper parts of convective clouds (Ferraro et al. 1998), a simple explanation may be “warm-type rain,” which can occur without substantial upper-level ice particles. However, this warm-type rain appears to be different to the orographic precipitation described by Houze (1993) and the warm rain due to the so-called seeder–feeder enhancement mechanism over the U.K. region (Browning et al. 1974). This is because heavy rain rates (i.e., greater than 20 mm h−1) are frequently observed from clouds with much warmer 85-GHz Tbs than those producing the same rain rates in the GPROF algorithm.
If clouds with low amounts of ice crystals cause heavy rainfall, the cloud–rain system over Korea may be considerably different to the highly convective clouds that cause heavy rainfall such as that over the Great Plains of the United States. Cloud–precipitation schemes have been developed in state-of-the-art weather prediction models to simulate the severe weather in the United States; however, such schemes may not work well over Korea owing to the different cloud–rain system (e.g., Hong 2004). Moreover, as discussed earlier, the IR-based rain-rate estimation method may hold less physical background because of the warmer cloud-top temperature, despite heavy rainfall. Even if IR-based estimates are well adjusted to microwave-based estimates, many problems may arise because a microwave-based algorithm may not resolve heavy rainfall over the Korean peninsula area.
To address the potential problems inherent in flood forecasting or rainfall estimation, the present study intends to document characteristic features of warm-type rain over the Korean peninsula area. We also examine how those warm-type rain features are different to those found over the central U.S. region (the Oklahoma area), where the TRMM rain estimation algorithm for land is adjusted to surface observations (Olson et al. 2006). Then, combining TRMM products with reanalysis data, we further examine the large-scale atmospheric environment and the possible conditions necessary to induce warm-type rain events over the Korean peninsula. The results obtained may lead to a better understanding of the rain system causing flood, which could in turn improve the forecasting of heavy rainfall.
2. Analysis domain and data used
In this study, we hypothesize that the TMI algorithm for land underestimates the rain rate over the Korean peninsula substantially because the cloud–rain system over Korea is considerably different to that found over the Oklahoma region of the United States (U.S.–Oklahoma region), where the TMI land algorithm has been tuned (Olson et al. 2006). To examine the proposed hypothesis, we compare cloud structures and their associated rain characteristics between the two regions using TRMM and CloudSat data.
a. Analysis domain
In our comparison of the rainfall characteristics of Korea and the U.S.–Oklahoma (hereafter, referred to as US-OK) region, the southern part of Korea is contrasted with the US-OK area south of 36.5°N (boxes with solid lines in Fig. 1), the northern limit of TRMM Precipitation Radar (PR) coverage. Over the domains outlined, TRMM TMI 85-GHz Tbs and PR reflectivity and their derived products (such as rain rate and storm height) are compared for the summer months (June–August) over a 10-yr period (2002–11). The characteristics of ice water profiles are also examined over these two regions using CloudSat data, but over a different analysis period (i.e., only the four summers during 2007–10), owing to the limited availability of CloudSat. Last, through examination of large-scale synoptic features related to the development of certain types of rain, warm-type (or cloud-type) rains observed over the Korean peninsula are related to large-scale meteorological fields using European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis Interim (ERA-Interim) data. While detailed description of our analysis methods is deferred to sections 3 and 4, more detailed discussion of the data used is provided below.
b. TRMM/TMI and PR data
The TMI is a scanning passive microwave radiometer with vertical and horizontal polarization channels at 10.65, 19.35, 37.0, and 85.5 GHz, and a vertical polarization channel at 21.3 GHz. The spatial sampling resolution along the satellite track varies from 63.2 km (10.65-GHz channel) to 7.2 km (85.5-GHz channel), with a cross-track sampling resolution of 4.6 km for the 85.5-GHz channel and 9.1 km for the other channels. The 85.5-GHz channel is referred to as the 85-GHz channel for simplicity. The vertically polarized brightness temperatures for the TMI 85-GHz (TB85V) data (1B11) are used, because the TMI rain retrieval algorithm uses 85-GHz brightness temperatures for land.
The PR (an active radar operating at 13.8 GHz with a 220-km swath) measures radar reflectivity with a vertical range resolution of 0.25 km and a horizontal sampling resolution of 4.3 km. The equivalent reflectivity (Ze) profiles (TRMM 2A25) from the PR, PR-derived rain rates, and storm heights (TRMM 2A23) are obtained from NASA Goddard Space Flight Center (http://mirador.gsfc.nasa.gov/). Here, the storm height is defined as the echo-top height at a given rain target, which is defined using PR reflectivity of 15 dBZ as a criterion (Awaka et al. 1998).
The combined TMI and PR data (collected only after the orbit boost in August 2001) are used to identify cloud–rain features representing Korea and US-OK during the summer. The PR and TMI 85-GHz measurements are at resolutions of 5 km × 5 km and 8 km × 5 km, respectively. When collocating these two sets of measurements, the PR pixel located nearest to a TMI pixel is selected only if the PR pixel is found to be a rain pixel. Collocations are made for the Korea and US-OK regions (two solid boxes in Fig. 1) for summer (June–August) over a 10-yr period (2002–11). Note the slight difference in spatial resolutions from the original setup; this is due to the orbit boost.
c. CloudSat/CPR data
The Cloud Profiling Radar (CPR) on board the CloudSat is an active sensor operating at 94 GHz, providing the vertical structure of cloud with a pixel size of about 1.7 km × 1.3 km. It measures reflectivity for 125 vertical bins (each about 240 m thick) up to 30-km altitude (Stephens et al. 2008). In this study, CloudSat ice water content profiles from the Radar–Visible Optical Depth Cloud Water Content Product (2B-CWC-RVOD) are used to compare the vertical structures of clouds found in Korea and US-OK during the four summers of 2007–10.
d. ERA-Interim data
To understand large-scale features that might be linked to rain characteristics over Korea, ERA-Interim data are collected for the same 10 summers (2002–11). The parameters used include water vapor and wind profiles and geopotential height at the 850-hPa level. In this study we use 6-hourly 1.5° × 1.5° gridded data.
3. TMI–PR rain characteristics over the Korean peninsula
To examine the characteristic rainfall features noted over Korea and understand why the TRMM rain retrieval algorithm severely underestimates the rain rate over the Korean peninsula, we compare climatological features of the rain–cloud system using TRMM data with those over US-OK. A total of 94 070 and 40 803 TMI–PR collocated pairs are obtained for Korea and US-OK, respectively. Note that detectable rain clouds are more frequent over Korea, in association with the progress of the East Asian monsoon during summer.
To examine how rain characteristics differ between the two regions, probability distribution functions (PDFs) obtained from PR-derived rain rates are produced for both regions (Fig. 2). Although there are more rain samples over Korea, the general frequency distributions are found to be similar. It is shown that light rain (i.e., that weaker than 10 mm h−1) is dominant in both regions, comprising around 90% of all cases. Despite a general similarity between the two PDFs in terms of instantaneous PR rain rate, US-OK has a larger percentage of occurrences of moderate to heavy rain (10 < rain rate < 40 mm h−1).
Despite the similarities in these PDF distributions, the PDF distributions of PR-derived storm height and collocated TMI 85-GHz Tbs (TB85V) are found to be substantially different (Fig. 3). The top two panels in Fig. 3 illustrate the PDF and cumulative distribution function (CDF) of PR storm height for Korea (left panels) and US-OK (right panels). The storm height distribution over Korea (Fig. 3a) exhibits a skewed pattern with a sharp peak at altitudes of 5–6 km. The frequency of storm heights greater than 6 km decreases sharply and most rain-producing storm heights are lower than 10 km. It is also of interest to note about 20% of precipitating clouds are found to be produced from storm heights of less than about 4 km, probably representing stratiform-type precipitation.
In contrast to the sharp peak at 5–6 km noted over Korea, US-OK exhibits a weak peak at about 5 km, with frequency decreasing gradually with increasing storm height. This pattern results in about 60% of the selected PR-based rain pixels exhibiting storm heights greater than 6 km, a striking contrast to the 35% for Korea. Therefore, these results lead to the conclusion that rain-producing clouds in US-OK are generally taller than those found in Korea. In particular, the proportion of high cloud (above 10 km storm height) appears more abundant in US-OK: nearly 10% of the samples are higher than 10 km, compared to the small percentage of high storm height clouds (>10 km) over Korea.
TB85V distributions over Korea exhibit a peak around 265 K and decrease rapidly for colder TBs (Fig. 3c), yielding approximately 60% of collocated TMI pixels warmer than 250 K. Conversely, TB85V distribution over US-OK indicates that about 50% of all cases are warmer than 250 K. The percentage of cases exhibiting colder TB85V is comparatively small over Korea, with only a small percentage of TB85V colder than 200 K. In contrast, about 7%–8% of all cases are found to be colder than 200 K over US-OK. It is clear that rain clouds with much colder temperatures (e.g., lower than 150 K), are found over US-OK, while almost no instances of such cold temperatures are found over Korea. Based on the analysis of storm height and collocated TB85V, it is reasonable to conclude that rain clouds over Korea are generally at lower heights and exhibit warmer TB85V than those over the US-OK region. In particular, the rain clouds with warmer TB85Vs and lower cloud tops over Korea suggest that the rain-producing clouds over Korea contain a smaller amount of ice crystals. This is consistent with the finding that heavy rainfall over Korea is often associated with relatively warmer TB85V (Ryu et al. 2012).
To examine how storm height varies with rain intensity, PDF distributions of PR storm height are classified by rain intensity (here, rain rate < 10 mm h−1, 10 < rain rate < 20 mm h−1, 20 < rain rate < 40 mm h−1, and rain rate > 40 mm h−1); the results are presented in Fig. 4. The rain classes are referred to as light, medium, heavy, and extreme, respectively. Light rain (rain rate < 10 mm h−1) is dominant in both regions, making up around 92% and 88% of total cases for Korea and US-OK, respectively. The corresponding mean PR rain rates are 2.31 and 2.56 mm h−1. The second class, representing moderate rain (i.e., 10 < rain rate < 20 mm h−1), constitutes around 5.2% and 7.8% of the total for Korea and US-OK, respectively; their respective means are 13.7 and 13.8 mm h−1 (Fig. 4b). The third class, representing heavy rain (20 < rain rate < 40 mm h−1), accounts for 2.0% and 3.3%, with mean rain rates of 27.4 and 27.2 mm h−1, over Korea and US-OK, respectively (Fig. 4c). The extreme rain class (rain rate > 40 mm h−1) accounts for only 0.88% and 0.97% of the totals in Korea and US-OK, respectively, with respective mean rain rates of 59.8 and 56.7 mm h−1, although such rain rates determined from PR measurements tend to be associated with greater uncertainty (T. Iguchi 2012, personal communication). More quantitative statistics are provided in Table 1.
The PDF distributions of storm height for the light rain (Fig. 4a) exhibit similar features in Korea and US-OK, although a peak around 5 km is much more pronounced in Korea. It is noted that storm heights over US-OK shift rapidly to higher altitude with intensified rain rates. In particular, the peak shown at around 5 km for rain rate < 10 mm h−1 is shifted quickly to much higher altitudes when rain intensity becomes stronger. In contrast, the peak shown in the Korean case is much less variable; the peak for the light rain at 5 km moves slightly to 7–8 km for the other three classes. Despite the tendency for an increasing percentage of the storm height above 10 km with increasing rain intensity, the occurring frequency is much smaller than that observed over US-OK. Only 5.4%, 10.9%, and 21.5% of medium, heavy, and extreme rain-rate classes, respectively, exhibit storm heights above 10 km in Korea, whereas 24.4%, 44.3%, and 71.1%, respectively, are above 10 km for US-OK. Because of these substantially different characteristics, we presume that the cloud system producing heavy rain over US-OK likely contains more ice crystals than the cloud system producing a similar rain rate over Korea.
The PDF distributions of TB85V for the four rain-rate classes are also examined (Fig. 5). In Korea, the TB peaks move from 265 K for rain rate < 10 mm h−1 to lower temperatures with increasing rain rate. Although colder temperatures are more frequent for heavier rain cases, patterns shown for light rain are maintained for the moderate to extreme rain classes. Compared to the Korean case, as illustrated by the PR storm height distribution, TB85Vs over US-OK are found to be much colder in spite of a similar feature found for the light rain. The US-OK region also shows a much broader temperature distribution in particular when rain gets stronger. Interestingly TB85V patterns for heavy rains appear to be quite similar to those found in Fig. 2 of Hanna et al. (2008), in which infrared cloud-top temperatures for rain areas during the winter over the United States were characterized as a bimodal distribution.
Because colder TB85Vs for rain clouds imply more ice-induced scattering, the general shift to colder temperatures for US-OK suggests that precipitating clouds in US-OK are taller and contain more ice particles than their Korean counterparts. Thus, rain clouds over Korea should be associated with less ice crystals in the upper part of cloud while more liquid water clouds in the lower layers. Such discrepancies become more prominent when rain gets stronger because deep convective clouds are more abundant over US-OK, as also demonstrated in the storm height distribution (Fig. 4). These TRMM-based findings strongly suggest that the rain mechanism over the Korean peninsula is closely associated with warmer clouds whose storm heights are located at lower altitudes.
4. Contrasting vertical distribution of PR reflectivity and ice water content
To compare the vertical structures of the rain-producing cloud system of Korea with that of US-OK, contoured frequency by altitude diagrams (CFADs) of PR reflectivity are constructed for the same four rain-rate classes. In the CFADs for light rain for Korea (Fig. 6a), the highest frequencies are found within the lower layer, below 5 km, where the reflectivity is primarily 20–35 dBZ. A high-frequency area is also found above 5 km but below about 7 km in altitude, with a weaker reflectivity of around 20 dBZ. The abrupt increase of the reflectivity just below 5 km appears to correspond to the melting layer. Thus, we interpret the light rain class over Korea to be associated primarily with stratiform-type precipitation (Fu and Liu 2003; Takayabu and Hirosaka 2009).
The CFADs for heavier rain intensities in Korea (Figs. 6b–d) are characterized by higher reflectivity below 5 km (i.e., 35–45 dBZ for 10 < rain rate < 20 mm h−1, 40–50 dBZ for 20 < rain rate < 40 mm h−1, and 45–55 dBZ for rain rate > 40 mm h−1 near the surface). Also noted is the increase in reflectivity closer to the surface, resulting in a reflectivity slope below 5 km. This slope becomes steeper for heavier rainfall. In contrast, reflectivity intensity and contour pattern above 5 km vary little, despite changes in rain intensity.
Vertical profile of radar reflectivity slope above the melting layer has been known as an indicator of storm intensity with lightning. More vertically aligned reflectivity profile with larger reflectivity just above the melting layer is likely associated with strong updraft, and with more frequent lightening and hail occurrences (Zipser and Lutz 1994; Toracinta et al. 1996). Unlike those observations related to the deep convection, the smaller reflectivity just above the melting layer around 5 km and lower storm height in Korea (Fig. 6) indicate a lack of large ice particles aloft in the cloud, suggesting that the cloud system over Korea may have weaker vertical motion and produce small lightning flash counts. The CFAD features showing increased reflectivity slope with increased rain intensity in Fig. 6 further suggest that rain drop sizes below the melting layer grow larger during the falling, probably via collision and coalescence processes. And the growth rate of drop size is thought to be higher for the heavier rain. These findings further lead to a conjecture that a large portion of the heavy rain cases may be attributed to rain drops growing below the melting layer. Lower cloud layer may contain plenty of small cloud droplets formed under the humid environment surrounding Korea during the summer, which can keep supplying water to rain drops and thus produce heavy rainfall. These whole features are evidently contrasting with ones interpreted from larger reflectivities above the melting layer and more vertically aligned reflectivities noted over US-OK.
The CFADs over US-OK for the same four rain classes are presented in Fig. 7. The CFAD for the light rain case (Fig. 7a) exhibits a similar pattern to that of Korea, despite slightly higher reflectivities above 5 km and taller storm heights. However, substantial differences to the Korean pattern are evident in CFADs when rain is stronger. The reflectivity patterns in US-OK remain nearly vertically aligned and are considerably different to the reflectivity slopes noted for Korea below 5 km, indicating that raindrop sizes (or rainfall amount) in the lower layer (below 5 km) vary little with altitude.
Above 5 km, in contrast to the sharp decrease in reflectivity evident in the Korea case, the decrease is slower and the vertical extent of detectable reflectivity reaches much higher values in US-OK. A greater spread in reflectivity is also evident for US-OK. As discussed for the storm height distributions in Fig. 4, stronger reflectivities are noted at higher altitudes. For example, in US-OK, reflectivities of 30–40 dBZ are found at 10 km for 20 < rain rate < 40 mm h−1, while such intensities are rarely found above 7 km in Korea. Because of the linear relationship between radar reflectivity and ice water contents above the freezing level (Donaldson 1961), the ice water content must be much lower in Korea than in the US-OK region. This interpretation is consistent with a finding that clouds producing heavy rainfall over Korea have much warmer brightness temperature at 85 GHz than clouds in US-OK producing similar rain intensity (as shown in Fig. 5).
The much lower ice water content in Korea is clearly depicted in the vertical frequency distributions of ice water content estimated by CloudSat (Fig. 8). In contrast to the analysis of TRMM measurements, all CloudSat-derived ice water profiles observed over Korea and US-OK are taken for the average, regardless of precipitation status, because of the limited number of samples. Thus, ice water content from nonprecipitating cirrus clouds is included in the analysis. Ice water content for Korea is generally less than 400 mg m−3, which is much lower than the ice water content found in the US-OK region. This lower ice water content is interesting because the monsoon activity over East Asia, which favors the generation of greater amounts of cloud and precipitation over Korea, cannot help produce ice water in amounts as large as those in US-OK. These results are sufficient to conclude that the ice generation mechanism may be less important over Korea to cause heavy rainfall than over US-OK.
5. Warm-type versus cold-type rain over Korea
Over the humid East China Sea near Okinawa Island, it has been suggested that convection cells in association with the Japanese baiu front can exhibit low echo-top heights in radar measurements (Hashimoto and Harimaya 2005; Kato et al. 2007; Oue et al. 2010; Oue et al. 2011). Zhang et al. (2006) also reported that convection cells with echo-top heights (using reflectivity of 15 dBZ as a criterion) below 8 km are characteristic features found in the mei-yu front area in China, suggesting so-called medium depth convection. We believe that the concept of heavy rainfall from lower clouds in this study is consistent with the aforementioned medium depth convection observed by radar under humid environments near the East Asian monsoon front. Extending the concept of medium-depth convection, the CFADs for Korea described in Fig. 6 are further divided into two groups, with storm heights above or below 8 km (Fig. 9). In this study, however, instead of referring to medium depth convection, rainfall from clouds with storm height lower than 8 km is referred to as “warm type” rainfall and that from clouds with storm height higher than 8 km is referred to as “cold type” convection. Since warm-type heavy rain cases show that reflectivity larger than 30 dBZ is confined within about 1 km above the melting layer (Fig. 10a), the warm-type rain is thought to be induced largely by warm rain-forming processes in the lower layer.
In Fig. 9, it is shown that about 90.8% of the total rain cases are associated with clouds with storm height lower than 8 km. If this 90.8% is grouped according to rain intensity, light rain (weaker than 10 mm h−1) is dominant, occupying 85.25% of the total. That is, only 5.5% of cases exhibit rain intensity greater than 10 mm h−1. It is noted that the reflectivity slope below 5 km becomes stronger with increasing rain intensity when the rain becomes heavier (e.g., >10 mm h−1). In particular, the rapid increase in reflectivity from 8- to 5-km altitude is interesting, because such a pattern suggests rapidly forming precipitation features above the melting layer.
The light rain associated with storm heights greater than 8 km amounts to a sizable 6.46% of the total cases, although this is much smaller than the 85.25% of cases grouped as light rain at lower storm heights. This light rain is probably associated with anvil-type precipitation occurring at a later stage of convection. Here 1.42% and 0.83% of totals, falling into cold-type medium to heavy rain cases, appear to be slightly weaker in percentage than corresponding warm-type cases showing 3.88% and 1.29%. These indicate that even heavy rain formation over the Korean peninsula in summer is more frequently associated with the warm-type convection. It is surprising to note that extreme rain events are not dominated by any of types (i.e., 0.41% vs 0.47%). Again, for medium to extreme rain classes, the vertical distributions of reflectivity exhibit an increasing pattern but with a slightly weaker slope than that of warm-type cases. It is also noted that the precipitation frequency above the melting layer seems to be same as that in the lower layer, indicating that raindrops falling on the surface are mostly initiated high above the melting layer.
Different features of the vertical distribution of reflectivity and layer precipitation occurrence are summarized with mean reflectivity profiles and precipitation occurrence frequency relative to the surface for both warm-type and cold-type heavy rain cases (20 < rain rate < 40 mm h−1; Fig. 10). For the warm-type rain (Fig. 10a), the reflectivity increases rapidly until reaching the melting layer; then, it continues to increase below the melting layer but at a lower rate. In conclusion, combined reflectivity and precipitation occurrence strongly suggest that snow/ice particles falling from the upper parts of clouds grow into detectable sizes primarily before hitting the melting layer; then, liquid water particles below the melting layer must keep growing. Raindrops are large enough to cause heavy rainfall by the time they fall to the surface, since its reflectivity show around 40–45 dBZ.
For the cold-type rain, precipitation seems to start far above the melting layer and particles should be easily detectable by PR radar at around 8–9 km. Knowing that the reflectivity increases between 8 and 5 km, the layer should serve as a feeder zone and the feeding process may continue even after melting to explain continuously increasing reflectivity below the melting layer. Although the reflectivity slope in the lower layer is weaker than that for the warm-type case, it is suggested that the lower layer serves as a feeder zone, even for the cold-type rainfall.
One salient rainfall feature that should be found over the Korean peninsula is the heavy warm-type rainfall, which is summarized in Fig. 11. As explained previously, solid particles precipitating from the upper part of the cloud may grow as a result of the feeding process until reaching the melting layer. Subsequently, the raindrops tend to keep growing into larger particles, perhaps through collision and coalescence processes, until hitting the ground and causing heavy rainfall. Even though this is conjectural we consider the aforementioned feeding process to be a working hypothesis. This concept is particularly important because most of the water in the rainfall seems to be provided within the layer below 5 km, in which cloud liquid water should be sufficiently abundant to feed the raindrops, allowing them to grow and cause heavy rainfall. Thus, based on the current analysis, we may conclude that the more abundant warm-type rain causing heavy rainfall over Korea occurs through feeding processes in the lower layer, probably in combination with strong moisture convergence induced by the wind field in the humid atmospheric environment.
6. Large-scale features linked to rain types over the Korean peninsula
We wish to examine the hypothesis that the warm-type heavy rainfall over the Korean peninsula may be linked to the more humid environment and stronger water vapor transport along the northwestern periphery of the North Pacific high to the Korean peninsula. To achieve this, we used TRMM PR-derived storm height and rain rate to construct the associated large-scale atmospheric environments. A schematic diagram explaining the procedures for constructing the composite fields of meteorological parameters is presented in Fig. 12. Since we are interested primarily in heavy rainfall, we examine only the rain class of 20 < rain rate < 40 mm h−1. To construct the composite fields, TRMM PR rain rates are first tested to assess whether pixel-level rain rates over the Korean peninsula are in the 20 < rain rate < 40 mm h−1 class. The rain area pertaining to each rain type is obtained by separating the rain pixels into warm and cold types depending on storm height, and then by counting pixels for each type. Once a dominant rain type is determined based on the dominant rain area, ERA-Interim meteorological fields at the time of TRMM overpass are considered to be synoptic fields relating to the rain type over the Korean peninsula. Since ERA-Interim data at the TRMM overpass time are not generally available, two adjacent 6-hourly ERA-Interim data are interpolated to match the TRMM observation time. After repeating these procedures for the summers of 2002–11, two groups of synoptic fields are constructed by taking a simple average, from which large-scale atmospheric fields related to warm-type or cold-type heavy rainfall over the Korean peninsula can be examined.
The obtained large-scale environments that might affect dominant rain types in heavy rainfall events over the Korean peninsula are given in Fig. 13, in which the 850-hPa geopotential heights (solid lines), water vapor flux at the 850-hPa level (arrows), and atmospheric column-integrated water vapor flux convergence (color scales) are depicted. It is shown that large-scale atmospheric environments linked to heavy rainfall over the Korean peninsula tend to look similar in both cases: strong water vapor fluxes are present along the periphery of the North Pacific high, centered over the northwestern Pacific, and associated moisture convergences are found over the area from southern part of Korea to western Japan. This similarity is interesting, because expansion of the North Pacific high further into the continent and abundant water vapor transport seem to represent a large-scale environmental setup favorable for heavy rainfall events over the Korean peninsula.
However, clear differences are present between the two types: vertically integrated water vapor fluxes are stronger for the warm-type dominant case, consistent with a stronger flow, expected from the stronger northwest–southeast gradient of 850-hPa geopotential height across the Korean peninsula. In effect, the pressure patterns appear to establish an “atmospheric river”-like channel through which much water vapor can flow. The more humid conditions over the East China Sea region (not shown) should also be taken into consideration. Consequently, much more water vapor is supplied due to the flux convergence to result in heavy rainfall over Korea with warm-type cloud development. Conversely, water vapor convergence is substantially weaker for the cold type. Simple explanation of rain processes may be given for relating synoptic conditions to heavy rain types. When the synoptic environment is set to continuously provide abundant water vapor to the peninsula, vigorous convection producing high cloud tops with abundant ice crystals may not be necessary to form the heavy rainfall. Instead, as shown in Figs. 10a and 11, raindrops may grow primarily below the melting layer at the expense of rich cloud liquid water produced under extremely humid conditions. Conversely, convectively unstable conditions may build up under relatively dry conditions and may induce high-altitude clouds via convective instability, with abundant ice crystals contributing to the occurrence of cold-type heavy rains. In conclusion, heavy rains over Korea can occur from lower/warmer clouds developed under very humid conditions through processes that are considerably different to the well-known cold-type rain mechanism prevailing over the Great Plains of the United States.
7. Summary and conclusions
This research first attempted to answer why the TRMM rainfall estimate, based on the use of scattering indices over the Korean peninsula, results in severe underestimation compared to surface rain gauge measurements. Then, characteristic features of rainfall were examined. It was assumed that convection features and associated cloud microphysics over Korea are considerably different to those noted in the Oklahoma region of the United States, where a scattering-based TRMM rain estimation algorithm for land has been calibrated. In other words, the lower availability of ice crystals over Korea was thought to be the main reason for the substantial underestimation of rain rate using the 85-GHz-based scattering method. To test this hypothesis, multiyear TRMM PR reflectivity, PR-derived rain rate and storm height, and TMI 85-GHz brightness temperature data were compared for the Korean peninsula and Oklahoma (US-OK) region.
It was found that clouds over Korea are not as high as the deep convective clouds found in the US-OK region, even when surface rain rates are equal. In general, clouds over Korea were found to be substantially lower than those over US-OK, even for heavy rainfall. Thus, 85-GHz Tbs can be much warmer because of the lower availability of ice crystals. Consistent with these findings, CloudSat measurements demonstrated that the ice crystals in Korea are much smaller than those found in clouds over US-OK. These cloud features are likely the main reason for significant underestimation of rain retrievals by passive microwave measurements over the Korean peninsula.
It was noted that the majority of summer rainfall over the Korean peninsula (about 90% of the total) occurred in association with relatively low clouds (so-called medium-depth convection) with PR storm height of less than 8 km. Moreover, the characteristics of warm-type rainfall were examined by separating the rain into warm-type and cold-type rains using the 8-km storm height as a criterion. Compared to rainfall associated with more conventional deep convective clouds, the reflectivity profile for the warm-type rain exhibited rapidly increasing reflectivity at two layers (i.e., one layer between 8 km and the melting layer at about 5 km, and another layer below the melting layer). The increasing reflectivity in the lower layer should be of more interest, because it is indicative of raindrops that grow while falling to the surface. We concluded that snow–ice particles in the upper layer grow mostly into detectable precipitation sizes before reaching the melting layer; subsequently, raindrops below the melting layer appear to keep growing. The growing processes in the lower layer should be closely associated with strong water vapor convergence over the Korean peninsula, which is caused by large-scale environmental conditions and appears to represent an important physical mechanism necessary for the production of heavy rainfall from low-level cloud over Korea. During summer, the intensifying North Pacific high leads to a pressure distribution that helps transport water vapor from the South and East China Seas to the Korean peninsula along the periphery of the high, establishing an atmospheric river–like water vapor transport channel across the Korean peninsula area. When vapor is abundant in the upwind region and pressure pattern provides a strong confluence region, water vapor should be much more widely available over the peninsula, where a slight ascending motion may result in the more immediate formation of clouds and warm-type rain there.
Finally, caution should be exercised when results are interpreted from the perspective of cloud's microphysical processes. It is because conclusions regarding microphysical processes were not fully supported by cloud microphysics itself. We much anticipate future studies, including modeling exercise, to test if precipitation formation by collision and coalescence below melting layer is a plausible mechanism to produce heavy rainfall under the meteorological conditions found over Korea.
The authors thank two anonymous reviewers and Dr. David Shultz for their constructive and valuable comments, which led to this improved version of the manuscript. Thanks are also given to Profs. K. Lee at Kyoungbuk National University and S.-Y. Hong at Yonsei University, South Korea, for valuable discussions. This work was funded by the Korea Meteorological Administration Research and Development Program under Grant CATER 2012–2092, and was supported by the “Research for the Meteorological Observation Technology and Its Application” project at the National Institute of Meteorological Research, Korea Meteorological Administration.