HTML
--> --> -->The first study to examine the intraseasonal variation of DSD was conducted by (Kozu et al., 2005). The authors took advantage of 2D video disdrometer (2DVD) observation during the first campaign of Coupling Processes in the Equatorial Atmosphere (CPEA-I) on mainland Sumatra, particularly at Kototabang. (Marzuki et al., 2010a) followed up the study conducted by (Kozu et al., 2005) by carefully classifying the data into several rain types. Both studies found that the DSD during the inactive MJO phase is broader than that indicated by the Marshall-Palmer model (Marshall and Palmer, 1948) for the active MJO phase. The active MJO phase is characterized by the occurrence of a large-scale, eastward moving center of strong deep convection and precipitation, while the inactive phase is characterized by weak deep convection and precipitation (suppressed phases). The difference in the DSD between the active and inactive MJO phases leads to differences in the radar reflectivity-rainfall rate relation (Z-R relation), which is commonly incorporated into meteorological radars. Recently, (Marzuki et al., 2016) conducted a more comprehensive investigation of the intraseasonal variation of DSD over Sumatra. The authors analyzed 2DVD data over eight consecutive years (2003-10) at Kototabang. In general, their research is consistent with previous studies indicating that the DSD during the inactive MJO phase is broader than that during the active phase. The intraseasonal variation of DSD can not only be observed for the surface DSD, but also for the vertical structure of DSD, as determined from Equatorial Atmosphere Radar (EAR) data (Kozu et al., 2005).
The difference in the strength of convection during the active and inactive MJO phases may influence the DSD characteristics in Sumatra. (Morita et al., 2006) found stronger convection during the inactive MJO phase, which is indicated by a higher rain top height. Recently, (Marzuki et al., 2016) also found stronger convection during the MJO inactive phase in Sumatra, characterized by a higher rain top height, larger reflectivity radar aloft, and less stratiform rain. Correlatively, lightning frequency in the Indonesian Maritime Continent is higher during the inactive than active MJO periods (Virts et al., 2013). More unstable environments during the inactive period, which is indicated by a larger lifted index, are conducive to stronger convective updrafts, liquid water content (LWC), graupel and, hence, lightning (Virts and Houze, 2015). (Marzuki et al., 2016) found a stronger convective updraft during the inactive MJO phase in Sumatra, as determined from EAR observation. They also found that cloud drops varied between the two MJO phases, as determined from Moderate Resolution Imaging Spectroradiometer (MODIS) data.
While studies in Sumatra have shown significant intraseasonal variations of DSD, the same condition is less distinct in Darwin. The DSD and Z-R equations formulated based on Joss-Waldvogel disdrometer (JWD) observations for each MJO phase in Darwin are similar (Giangrande et al., 2014). (Thompson et al., 2015) analyzed 2DVD data in the equatorial Indian and western Pacific oceans, at the islands of Gan and Manus, respectively. While rain intensity and accumulation vary with MJO phase, the mean DSD variabilities over these open ocean locations are similar and less related to the MJO phase.
The above review of the literature indicates a lack of consensus pertaining to the intraseasonal variation of DSD in response to the MJO. To date, research has been carried in only four locations; namely, Sumatra, Darwin, the northern Indian Ocean (Gan Island) and the western Pacific Ocean (Manus Island). Therefore, in this paper, we present the intraseasonal variation of DSD in response to the MJO at another location; namely, in the southeastern tropical Indian Ocean. The DSD data over this region during the Cooperative Indian Ocean experiment on Intraseasonal Variability in the Year 2011 (CINDY 2011) project were analyzed. An overview of the CINDY project can be found in (Yoneyama et al., 2013). During the CINDY 2011 field campaign, the active and inactive MJO phases were clearly observed (Xu and Rutledge, 2014); thus, it is possible to investigate the DSD characteristics during the two different MJO phases. Furthermore, previous studies regarding the effect of the MJO on DSD were conducted around coastal areas (Sumatra and Darwin) and were thus significantly affected by land-sea interaction, as indicated by a significant diurnal variation of DSD. During the CINDY 2011 campaign, observations were made in and around the tropical Indian Ocean; thus, precipitation is considered to be purely maritime and not influenced by local factors like the land-sea breeze. Although (Thompson et al., 2015) did not find a significant intraseasonal variation of DSD at Gan Island during the CINDY 2011 campaign, other researchers have reported variations in aerosol evolution (DeWitt et al., 2013), storm intensity and lightning activity in the Indian Ocean (Xu and Rutledge, 2014; Virts and Houze, 2015), that could affect the resultant DSDs. The data and methods used in this study are described in section 2, followed by a description and discussion of the rainfall and DSD analysis results in section 3. Section 4 provides a summary of the study and states the main conclusions.
2.1. JWD measurements
Between 25 September and 30 November 2011, DSDs were measured over the Indian Ocean (8°S, 80.5°E) by a JWD. The measurements were conducted under the CINDY 2011 project. The general meteorological and oceanic conditions during the CINDY 2011 field campaign are described in (Yoneyama et al., 2013). The disdrometer was installed on the roof of the anti-rolling system of the R/V Mirai, as shown in Fig. 1. The JWD is an RD-80 manufactured by Distromet Ltd., Switzerland. The instrument is designed to count and measure the fall velocity and precipitation particle size simultaneously in 20 size intervals ranging from 0.3 to 5 mm by converting the measurement of the impact energy of falling drops to the size of the diameter of each drop. The JWD employs a non-uniform bin size that increases with raindrop size, ranging from 0.092 to 0.572 mm (20 bin sizes).
Figure1. Field view of the whole of R/V Mirai. The white arrow indicates the position of the JWD installed on the roof of the anti-rolling system of R/V Mirai. The photograph was taken by Dr. Djamal KHELIF.Although the JWD is widely used to observe raindrops, there are some shortcomings in JWD measurements, such as system sampling limitations, dead time, atmospheric noise (Sheppard and Joe, 1994), quantization (Marzuki et al., 2010b), and underestimation of very small (D≤ 0.5 mm) and large (D≥ 5 mm) drop sizes (Tokay et al., 2001). The limitations of the JWD may influence the DSD for very heavy rainfall. During very heavy rainfall (R>20 mm h-1), the concentration of small drops (D<0.5 mm) and very large drops from JWD measurements is smaller than that from an optical disdrometer (2DVD) (Tokay et al., 2001). However, the difference in the concentration of small drop does not have a significant impact on integral rainfall parameters (IRPs), such as radar reflectivity and rainfall. Furthermore, the number of very large drops (D>5 mm) that cannot be detected by the JWD is very small, so this too does not significantly influence IRPs (Tokay et al., 2001). Nevertheless, the JWD is still useful today to study precipitation microphysics if the data are appropriately characterized and aggregated. To reduce the effect of the instrument's limitations, we conducted several filtering processes. First, we sampled the disdrometer data at a time interval of 1 min and disregarded very light rain (R<0.1 mm h-1). The minutes with small drop numbers (<10 drops) or R<0.1 mm h-1 are considered to be within the noise level (Tokay et al., 2001). Second, DSDs possessing fewer than four consecutive bins with non-zero values were also disregarded to reduce statistical and quantization errors.
A total of 4123 1-min DSD measurements were collected during the CINDY 2011 campaign. The following normalized gamma distribution was used to fit the DSDs, as reported in previous studies (e.g., Testud et al., 2001; Bringi et al., 2003): \begin{equation} \label{eq1} N(D)=N_{\rm w} f(D/D_{\rm m}), \ \ (1)\end{equation} with \begin{equation} \label{eq2} f(D/D_{\rm m})=\dfrac{6}{4^4}\dfrac{(4+\mu)^{(\mu +4)}}{\Gamma(\mu+4)}\left(\dfrac{D}{D_{\rm m}}\right)^\mu e^{-(\mu +4)\left(\frac{D}{D_{\rm m}}\right)} , \ \ (2)\end{equation} where D is the raindrop diameter, $\varGamma(x)$ is the complete gamma function and N w is the normalized intercept parameter indicating raindrop concentrations with exponential distributions featuring the same median volume diameter and LWC. The mass-weighted mean diameter (D m), shape parameter (μ) and N w are calculated by first estimating the LWC, as follows: \begin{eqnarray} \label{eq3} {\rm LWC}&=&\dfrac{\pi10^{-3}}{6}E[D^3] ;\\ \label{eq4} D_{\rm m}&=&\dfrac{E[D^4]}{E[D^3]} ;\\ \label{eq5} N_{\rm w}&=&\dfrac{4^4}{\pi\rho_{\rm w}}\dfrac{LWC}{D_{\rm m}^4} . \end{eqnarray} Here, ρ w is the density of water and E represents the expectation in which E[D3] and E[D4] are defined as: \begin{eqnarray} \label{eq6} E[D^3]&=&\int_0^\infty D^3N(D)dD ; \ \ (6)\\ E[D^4]&=&\int_0^\infty D^4N(D)dD . \ \ (7) \end{eqnarray} (Bringi et al., 2003) used a fitting procedure to estimate μ by minimizing the absolute deviation between the normalized DSD data and the scaled gamma form. In this work, μ is estimated as follows (Leinonen et al., 2012): \begin{equation} \label{eq8} \mu=\left(\dfrac{\sigma_{\rm m}}{D_{\rm m}}\right)^2-4 , \ \ (8)\end{equation} where σ m is defined as \begin{equation} \label{eq9} \sigma_{\rm m}=\sqrt{\dfrac{\int_0^{D_{\max}}{(D-D_{\rm m})^2D^3N(D)dD}}{\int_0^{D_{\max}}{D^3N(D)dD}}} . \ \ (9)\end{equation}
The rainfall rate R (units: mm h-1), in terms of the measured DSD, is defined as \begin{equation} \label{eq10} R=6\pi10^{-6}\int_0^\infty{D^3v(D)N(D)dD} , \ \ (10)\end{equation} where π is mathematical constant (3.14), and v(D) is the raindrop fall speed given by (Atlas et al., 1973) \begin{equation} \label{eq11} v(D)=9.65-10.3{\it e}^{-0.6D} . \ \ (11)\end{equation} Like R, the radar reflectivity factor Z (mm6 m-3) is calculated using the following equation: \begin{equation} \label{eq12} Z=\int_0^\infty {D^6N(D)dD} . \ \ (12)\end{equation}
2
2.2. Satellite data
As indicated in Eq. (11), radar reflectivity is strongly influenced by large-sized drops, because it is proportional to the raindrop diameter to the sixth power (Z≈ D6). Therefore, the vertical profile of radar reflectivity is closely related to the evolution of raindrop size, particularly for large-sized drops (e.g., Schumacher and Houze, 2003; Thurai et al., 2003). To investigate the vertical gradients of the individual radar reflectivity profiles, attenuation-corrected radar reflectivity data gathered from Tropical Rainfall Measuring Mission (TRMM)-Precipitation Radar (PR) observations were used. The datasets used for the current study are from the latest, TRMM 2A25, products (version 7). Because this study focuses on vertical structure, only profiles with an incidence angle less than 7° on either side of the nadir are included. We only analyzed the data within the region (7°-9°S, 79.5°-81.5°E), around R/V Mirai. The vertical profile of the radar reflectivity gradient is used as a proxy for raindrop evolution from the melting layer to the surface. The gradient is calculated by using the linear regression as a function of reflectivity and height (Hirose and Nakamura, 2004). In addition, the freezing height and storm height were analyzed. In this work, the storm height is assumed to occur at the highest altitude with Z≥ 18 dBZ for at least two consecutive range bins, as in (Marzuki et al., 2018).The characteristics of precipitation microphysics are related to the properties of clouds and the environmental conditions (May et al., 2011; Munchak et al., 2012). We took advantage of daily MODIS data to study the daily cloud effective radii (CER) of ice and liquid water, and the aerosol optical depth (AOD) over the research area. In addition, the convective available potential energy (CAPE) determined from ERA-Interim data were also used, to investigate the potential of storm growth.
A full MJO cycle is divided into eight phases, which correspond to the position of the active convective center originating from the western Indian Ocean (phase 1), to its decay over the central Pacific (phase 8). The method used to estimate the amplitude and phase of the MJO is described in (Wheeler and Hendon, 2004). Of the eight MJO phases, the period over which the JWD, TRMM and other data were gathered were classified into two categories, i.e., active and inactive (suppressed) MJO phases. The definition of active and inactive MJO phases for the Indian Ocean and nearby locations used in this study is the same as employed in (Xu and Rutledge, 2014). The active convective phase occurs in phases 1, 2 and 3, while the inactive convective phase of the MJO occurs in phases 5, 6, 7 and 8.
3.1. General features of DSD and rainfall
Figure 2 shows the normalized DSD for certain rain categories. The lower the rainfall intensity is, the larger the variability of the values become. At an intensity of R<1 mm h-1, values vary from -3 to 59; and at an intensity of R>20 mm h-1, values vary from 2 to 20. This value is greater than that previously reported. For example, (Bringi et al., 2003) also analyzed JWD data and obtained a μ range of -3 to 15 for R>0.5 mm h-1; and (Marzuki et al., 2013a) analyzed 2DVD data and obtained a μ range of -3 to 30 for raindrops at R>0.1 mm h-1 in Sumatra. For a normalized gamma distribution, positive (negative) values of the shape parameter (μ) indicate a concave downward (upward) shape of the raindrop spectrum (Kliche et al., 2008). Furthermore, negative values of μ in the gamma distribution indicate that the number of small raindrops exceed that of an exponential distribution. Thus, the oceanic raindrop spectrum during the CINDY 2011 campaign is more concave downward than the spectra reported in certain previous studies. The shape of the raindrop spectrum for very heavy rain (Fig. 2f) may be influenced by the limitations of the JWD in measuring very small and very large drops. However, such influence cannot be quantified in the current work because there are no another disdrometer observations available as a comparison.
Figure2. The scaled DSD N(D)/N w versus the normalized diameter D/D m for the data from the JWD disdrometer acquired during the CINDY campaign for several rain categories.Figure 3a shows the time series of the MJO index and amplitude during the CINDY campaign. Two inactive MJO periods were observed between 25 September and 14 October and between 6 November and 16 November. Rainfall events were observed almost every day during the first inactive MJO period. The largest amount of rainfall (59 mm d-1) was observed on 1 and 4 October. A relatively large amount of rainfall (≈ 14 mm d-1) was also observed on 29 September and 10 October. During the second inactive MJO period, the largest amount of rainfall was observed on 13 November (≈ 3.5 mm d-1). Heavy rainfall during the inactive MJO phase was derived from isolated convective cells (Xu and Rutledge, 2014; Xu et al., 2015), as also previously observed in Sumatra (Alexander et al., 2006).
Figure3. Time series of (a) MJO phase, (b) daily rainfall, (c) the frequency of occurrence of rain data, and (d) rainfall on a diurnal basis. X indicates missing data or observations.Two active MJO periods were also observed during the CINDY campaign, i.e., between 15 October to 3 November, and during 17-30 November. The largest amount of daily rainfall during the active MJO phase was observed on 29 October, with an intensity of 39 mm d-1. A relatively large amount of rainfall (≈ 23 mm d-1) was observed during the second active MJO phase (28 November). Thus, throughout the entire campaign period, the rain intensity during the first inactive MJO period was higher than that during the active MJO phase. The precipitation during the active MJO phase is associated with organized mesoscale convective systems (Xu et al., 2015).
The diurnal variation of convective activities is less pronounced over the oceans and is significant over the continents, large islands, and their adjacent seas (Nitta and Sekine, 1994). Figures 3c and d show the diurnal cycle of the occurrence frequency of rain data and rainfall. The diurnal variation of these data is not particularly significant. Two heavy rain events on 1 and 4 October were observed during 0500-1000 UTC and 1500-2300 UTC. Less diurnal variation was also observed based on the DSD spectra (figure not shown). This finding implies that the diurnal variation of DSD observed over land may be due to the local convection system, such as land-ocean interaction circulation. The effect of the oceanic nature of rainfall is dominant over the Indian Ocean; thus, the diurnal variation of DSD is not distinct. Less diurnal variation of oceanic DSD has also been previously observed over Palau (Ushiyama et al., 2009) and Singapore (Kozu et al., 2006). Therefore, in the next section, the diurnal variation of DSD is ignored.
2
3.2. Intraseasonal variation of DSD
Figure 4 shows the partitioning of the CINDY 2011 JWD DSD according to the MJO phase. Recall that over the Indian Ocean the active MJO phase corresponds to phases 1, 2 and 3, and the inactive phase corresponds to phases 5, 6, 7 and 8. At low to medium rainfall intensity, the difference in the DSDs during the active and inactive periods is not significant. The DSD in the active period of the MJO is slightly broader than that during the inactive period. This difference becomes more pronounced at higher rainfall intensity (R>10 mm h-1). The DSD in the active phase of the MJO is broader than that in the inactive phase. The concentration of large-sized drops is also higher during the active phase of the MJO than that during the inactive phase. This finding contrasts with the results of previous studies conducted in Sumatra (Kozu et al., 2005; Marzuki et al., 2010a, 2016). In Sumatra, the DSD during the inactive MJO phase is broader than that during the active phase.
Figure4. Averaged DSD for several rain categories during active and inactive MJO phases.Table 1 summarizes the parameters of the normalized gamma distributions for each MJO phase. All three parameters of the gamma distribution in all MJO phases increase with the rain rate, which is similar to the findings reported in previous studies (e.g., Tokay et al., 2001). The results clearly show that the normalized intercept parameter of gamma DSD (N w) during the active phase of the MJO is lower than that observed during the inactive phase of the MJO. Despite the smaller concentration, the DSD spectrum during the active MJO phase features more large-sized drops, which are characterized by a larger D m value and a smaller shape parameter (μ). At 10<R<20 mm h-1, the difference in D m between the active and inactive MJO phases reaches a value of 0.15 mm.
(Bringi et al., 2003) grouped DSDs into two groups; namely, maritime and continental DSDs, based on a scatterplot of mean D m versus $\log_10(N_\rm w)$. Figure 5 compares the mean values of D m and N w obtained during the CINDY 2011 campaign with the data gathered from various locations. All data are adopted from (Bringi et al., 2003), but for Sumatra they are adopted from (Marzuki et al., 2016). Convective and stratiform precipitation are divided following the method of (Bringi et al., 2003). The D m vs N w value obtained during the CINDY 2011 campaign matches the maritime convective cluster. This result is not surprising because the CINDY campaign was carried out in the tropical Indian Ocean. The average N w during the inactive MJO phase (3.97), which is indicated by the number 13, is slightly larger than that during the active phase (3.89), which is indicated by the number 12. The average value of N w during the inactive phase is also slightly greater than that indicated by the Marshall-Palmer model $[\log_{10}(N_w)=3.9]$. Thus, the number of raindrops during the active phase of CINDY is less than that during the inactive MJO phase, but the DSD contains more large-sized drops, indicated by the average DSD (Fig. 4) and D m value (Fig. 5). This finding contrasts with the results of a previous study performed in Sumatra, which suggested the average N w value during the active phase (4.16) is greater than that during the inactive MJO phase (3.91) (Marzuki et al., 2016).
Figure5. Mean $\log_{10}N_w$ (with ± 1 standard deviation) versus mean D m during the CINDY campaign, along with the data of (Bringi et al., 2003) and (Marzuki et al., 2016) for Sumatra. Panels (a) and (b) are for convective and stratiform rains, respectively. The dashed line indicates the Marshall-Palmer value $(\log_{10}N_w=3.9)$.One application of the results obtained for the DSD is the conversion of weather radar data to rainfall rate through the Z-R relation (Z=ARb). Figure 6 shows a scatterplot of the radar reflectivity factor versus rainfall rate. In general, at R<5 mm h-1, the measured radar reflectivity at the same rainfall rate is lower than that indicated by the Marshall-Palmer model, during both the active and inactive MJO phases. To further emphasize the clear differences in the Z-R relation characteristics between the active and inactive MJO, linear regressions of Z over R and R over Z, with a fixed exponent b=1.40, on log-transformed values, were used. The effect of the spurious variability of disdrometric data was minimized by averaging 10 DSD samples of sequential R values (Lee and Zawadzki, 2005). The equations for the active MJO phase obtained by the three methods are Z=272R1.25, Z=271R1.27 and Z=255R1.40, respectively. Furthermore, the equations for the inactive phase are Z=250R1.21, Z=250R1.23 and Z=245R1.40. The intercept and the exponent coefficient during the active MJO phase are slightly greater than those during the inactive phase. The intercept of the Z-R relations for the fixed exponent b is close to the value found in Sumatra and that recommended by (Rosenfeld et al., 1993) for tropical rainfall events. However, the values are smaller than the value proposed for the USA weather radar network; namely, Z=300R1.40 (Fulton et al., 1998).
(Thompson et al., 2015) argued that the Z-R relationship should not change as a function of the MJO. The intraseasonal variation of the Z-R relationship may be ignored for light and moderate rains because the difference in DSDs is very small (Fig. 4). However, under heavy rain, the intraseasonal variation of DSD may influence the accuracy of radar data conversion. For example, if we employ the DSD-based Z-Rs during active MJO phases, i.e., Z=272R1.25, Z=271R1.27 and Z=255R1.40, to convert a measured Z value of 40 dBZ to rainfall rate, the following rainfall rates are obtained: 17.9, 17.1 and 13.8 mm h-1, respectively. These values are greater than the value obtained by the Marshall-Palmer model (11.5 mm h-1). On the other hand, the DSD-based Z-Rs during inactive MJO phases, such as Z=250R1.21, Z=250R1.23 and Z=245R1.40, yield the following rainfall rates: 21.1, 20.1 and 14.1 mm h-1. The difference in the results of Z-R conversion becomes larger when the measured value of Z is greater, such as that measured under highly intense convective conditions. For a measured Z of 50 dBZ, the Z-R conversion using the DSD-based Z-Rs during active MJO phases provides rainfall rates of 112.8, 105.0 and 71.2 mm h-1. On the other hand, the Z-R relation obtained during the inactive MJO phase provides rainfall rates of 141.4, 130.5 and 73.3 mm h-1. These conversions are rough examples, and the accuracy of rainfall estimates using weather radar is determined not only by the selection of the Z-R relation, but also by multiple sources of error (Marzuki et al., 2018). Nevertheless, the effect of the intraseasonal variation of DSD on rainfall estimates can be appreciated. The same Z translates to a greater R in the inactive MJO phase compared with that in the active MJO phase.
2
3.3. Discussion
The differences in DSD characteristics during the CINDY 2011 campaign relative to those obtained over Sumatra may indicate differences in the microphysical processes affecting the DSD in each location. In Sumatra, the echo top height is higher during the inactive phase of the MJO (Kozu et al., 2005; Marzuki et al., 2010a, 2016). During the active MJO period, the TRMM-PR passed the location of the JWD on 16, 21 and 23 October; and on 1, 19, 23, 27 and 30 November. Furthermore, the TRMM-PR passed the JWD site on eight days: 28 September; 1, 5, 9 and 13 October; and 8, 11 and 15 November. Figure 7 shows the contoured frequency with altitude diagram of the radar reflectivity factor (Z) from TRMM 2A25 V7 over the region (7°-9°S, 79.5°-81.5°E) during the CINDY campaign. The echo top height is higher during the inactive phase of the MJO, as also observed in Sumatra. The same condition is also indicated by weather radar data (Xu et al., 2015). The isolated nature of precipitating clouds was observed during the inactive (suppressed) MJO periods of the CINDY campaign (Chen et al., 2015), as also observed over Sumatra during CPEA-I (Alexander et al., 2006). While the general features of vertical structures of precipitation during CINDY (ocean) and Sumatra (land) observations are similar, the intraseasonal variation of DSD in response to the MJO at each location is distinct.
Figure6. Reflectivity (Rayleigh regime) versus rain rate for the two MJO phases. The dashed-dotted line represents the Marshall-Palmer model (Z=200R1.6).Figure 8 shows the mean CAPE from ERA-Interim, and the freezing height and storm height from TRMM 2A25. The mean CAPE during the active MJO phase is slightly smaller than that during the inactive phase. The values are 283 J kg-1 and 287 J kg-1 during the active and inactive MJO phases, respectively. The standard deviation of the active MJO phase is slightly larger than that during the inactive phase because there are two days during the active phase (28-29 November) that are associated with larger CAPEs. A slightly higher storm height is also observed during these two days (Xu et al., 2015, Fig. 7c). The mean CAPE of these two days is 990 J kg-1. The intensity of rainfall on 28 and 29 November is 23 and 1.3 mm h-1, respectively. If the CAPEs of these two days are not included in the average, the mean CAPE during the active phase is 240 J kg-1 (gray line in Fig. 8a), which is much smaller than that during the inactive phase. Thus, in general, the characteristics of mean CAPE are consistent with the storm height. A higher CAPE during the inactive MJO phase supports more intense convection (Fig. 7) and, thus, a higher lightning frequency is observed (Xu et al., 2015), as also found in Sumatra.
Figure7. Contoured frequency with altitude diagram of radar reflectivity factor (Z) from TRMM 2A25 V7 over the region (7°-9°S, 79.5°-81.5°E) during the CINDY campaign. The bin size is 2 dBZ, and the plot is contoured at intervals of 2%. The vertical line is drawn at 30 dBZ.Precipitation with a higher freezing level during the active MJO phase (Fig. 8b) could result in larger drops at the ground level. Because of a higher freezing level height, aggregation and riming above the melting layer and drop sorting and collision-coalescence below the melting layer may result in larger drops at the ground level (Rosenfeld and Ulbrich, 2003). However, the difference in the freezing level height between the active and inactive MJO phases during the CINDY campaign is small (～ 200 m) and may not cause a significant difference in the DSD, as observed with light and moderate rains (Figs. 4a-d). Less intraseasonal variation of DSD for light and moderate rains has been previously reported in Sumatra (Marzuki et al., 2010a).
Figure8. Mean (a) CAPE, (b) freezing level height and (c) storm height. The error bar indicates the standard deviation. The gray line in (a) denotes the mean CAPE during the active MJO phase without the data on 28-29 November 2011.The storm height (Fig. 7 and Fig. 8c) during the active MJO phase is smaller than that during the inactive MJO phase, which indicates that shallow convective rain is more dominant during the active MJO phase. Dominant shallow convective rain during the active MJO phase could also be associated with larger drops at the ground level. The raindrop growth during shallow convective rains is more significant than that during deeper convection, which is indicated by the vertical profile of the radar reflectivity gradient, particularly over the ocean (Hirose and Nakamura, 2004; Marzuki et al., 2018). The gradient pattern is strongly coincident with the rain top height, where shallow convective rains (profiles with an echo top below the freezing level) mostly show a positive gradient while deep convective rains show both positive and negative slopes below the freezing level (Liu and Zipser, 2013). Thus, the downward increase in reflectivity gradient for shallow convective rains is much greater (more positive) than that associated with deep convective rains. In general, the majority of radar reflectivities below the freezing level in shallow convective rains over the ocean show a downward increase in the radar reflectivity gradient, which indicates dominant raindrop growth (Marzuki et al., 2018). However, during the active MJO phase, a more positive gradient was observed. The gradients of the mean radar reflectivity profile during the active phase of the MJO for R<8 mm h-1 and R>8 mm h-1 (Fig. 7) are 1.46 and 1.69 dBZ km-1, respectively; while they are 0.76 and 1.00 dBZ km-1, respectively, for the inactive phase.
For both Sumatra and the Indian Ocean, the storm height is lower during the active MJO phase than it is during the inactive MJO phase. However, the microphysics of precipitation at these locations is different. Such different microphysics may also be due to the strength of the updraft. The development of deep convection over large islands like Sumatra is influenced by land heating due to solar radiation (Marzuki et al., 2013b). Consequently, convective storms, particularly during the inactive MJO phase, are more intense on land, which is associated with a strong updraft, than those found in oceanic regions. A slightly strong updraft associated with deep convective rains during the inactive MJO phase over Sumatra was clearly observed based on EAR data (Marzuki et al., 2016). Strong updrafts carry small raindrops to higher altitudes, thereby allowing larger drops to precipitate locally (Huang and Cui, 2015; Seela et al., 2017), such that the concentration of large drops is higher during the inactive MJO phase in Sumatra. On the other hand, the updraft is generally weaker over ocean than over land. During the active MJO phase, smaller raindrops can fall and monotonically grow in size under weaker updraft conditions (Takahashi et al., 2017). Therefore, under the shallow convective conditions during the active MJO phase, the drop size increases with storm height, consistent with the longer path over which drop growth via collision occurs, such that a higher concentration of large drops is observed at the ground level (Munchak et al., 2012). Weak updraft during the inactive phase of the CINDY campaign is indicated by the radar reflectivity profile. The maximum echo top height of 30 dBZ, which indicates the depth of the precipitation system during the campaign, is lower than the heights observed over Sumatra. The maximum echo top height of 30 dBZ represents how high large hydrometeors reach (Liu and Zipser, 2013). In fact, only strong updrafts under intense convection can lift large hydrometeors to higher altitudes (Heymsfield et al., 2010).
Because the maritime atmosphere contains more water vapor than the atmosphere over land, the relative humidity over the ocean is higher than that over land. Humid environments are favorable for developing large-sized drops. Furthermore, evaporation can be negligible for high relative humidity (Liu and Zipser, 2013).
(May et al., 2011) and (Seela et al., 2017) observed a relationship between AOD and the characteristics of DSDs in Darwin, Taiwan and Palau, respectively. Thunderstorms involving high aerosol concentrations, which are indicated by large AOD values, feature a lower N w and larger D m compared with thunderstorms involving low aerosol concentrations. This condition was clearly observed during the CINDY campaign (Fig. 9a). The large mean AOD during the active MJO phase is coincident with small (large) N w (D m) values. Another cloud property that may influence the DSD is CER. The mean CER values of liquid during the active and inactive MJO phases are nearly the same (Fig. 9b). However, a large discrepancy was observed for ice: the mean CER value during the active MJO phase was much greater than that during the inactive MJO phase (Fig. 9c). (Seela et al., 2017) found that small CER values over Taiwan are associated with a large number of large-sized drops, which is indicated by a large D m. Similar results were obtained in this work.
Figure9. Mean (a) AOD, and (b, c) CER for (b) liquid and (c) ice. The error bar indicates the standard deviation.