Advanced Search
Article Contents

Favorable Environments for the Occurrence of Overshooting Tops in Tropical Cyclones


doi: 10.1007/s00376-016-6122-y

Get Citation+

Export:  

Share Article

Manuscript History

Manuscript received: 30 April 2016
Manuscript revised: 03 October 2016
Manuscript accepted: 04 November 2016
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Favorable Environments for the Occurrence of Overshooting Tops in Tropical Cyclones

  • 1. School of Atmospheric Sciences and Key Laboratory of Mesoscale Severe Weather of Ministry of Education, Nanjing University, Nanjing 210023, China

Abstract: Based on Multifunctional Transport Satellite data and the infrared window-texture detection algorithm, the level of overshooting top (OT) activity within a tropical cyclone (TC), which is defined as the hourly mean number of OT occurrence, was statistically investigated in the western North Pacific basin for the period 2005-12. Based on the level of OT activity, the samples were divided into OT and non-OT cases or high-activity-OT (HA-OT) and low-activity-OT (LA-OT) cases. The differences in large-scale environmental variables between OT (HA-OT) and non-OT (LA-OT) cases were examined 12 hours prior to the OT occurrence. Statistical analysis showed that environmental differences did exist between the OT and non-OT cases. The OTs were more skewed towards the early stage of the TC life cycle, and mostly concentrated in low latitudes. Meanwhile, a sufficiently deep warm-water layer, large temperature difference between the upper- and lower-level troposphere, large humidity at the middle and upper levels, and large atmospheric instability, were favorable for OT occurrence. The differences in large-scale environmental characteristics between HA-OTs and LA-OTs were not as significant as those between OTs and non-OTs, but the HA-OT samples tended to occur when the vertical shear was weak and the TC intensity was low. Finally, statistical models were designed to predict the OT and HA-OT. When at least three OT (HA-OT) predictor thresholds were satisfied, the Peirce skill score reached a maximum value of 0.49 (0.30).

1. Introduction
  • Hot towers (HTs) are towering cumulonimbus clouds with intense rapidly ascending cores that reach the level of the tropopause or higher. They are characterized by a horizontally small scale (10 km; Houze et al., 2009), short duration (<30 mins; Heymsfield et al., 2001) and a large amount of latent heat released above the freezing level (Zipser, 2003; Houze et al., 2009). Many studies (Heymsfield et al., 2001; Guimond et al., 2010; Rogers, 2010; Rogers et al., 2013) also named HTs as "convective bursts" (CBs), which represent the phenomenon of multiple HTs gathered in groups. Over the past several decades, many researchers have demonstrated that HTs play a significant role in the transportation of moisture (Chemel et al., 2009; Hassim and Lane, 2010) and energy (Riehl and Malkus, 1958; Riehl and Simpson, 1979) from the atmospheric boundary layer into the low-level stratosphere, which imposes a major impact on the energy balance and global climate (Dessler, 2002; Solomon et al., 2010). In addition, the relationship between HTs and tropical cyclones (TCs) have attracted the attention of many researchers (e.g., Montgomery et al., 2006; Houze et al., 2009; Guimond et al., 2010; Zhuge et al., 2015a, 2015b).

    For instance, in the process of tropical cyclogenesis, HTs possessing strong vertical vorticity in their cores (Vortical Hot Towers; Hendricks et al., 2004) play a key role in strengthening the TC vortex through the upscale vorticity growth mechanism, and thus promote the formation of the TC. This mechanism has been confirmed by numerical studies (Montgomery et al., 2006) and multi-source observations (Houze et al., 2009). During the development of a TC, HTs can affect TC intensity through the mechanism of subsidence warming (Heymsfield et al., 2001; Guimond et al., 2010; Zhang and Chen, 2012; Chen and Zhang, 2013). When the outflow from the cloud top forces air to sink near the TC eye, the downdraft adiabatic heating leads to a warm core in the upper troposphere of the TC center. As a result, surface pressure over the TC eye region decreases during the process of hydrostatic adjustment while the tangential wind speed increases, and the TC intensity increases significantly. (Kelley et al., 2004) analyzed six years of Tropical Rainfall Measuring Mission (TRMM) data and found that TCs tend to intensify when one or more extremely tall convective tower (HT-like convection) occurs in the eyewall area of the TC. (Zhuge et al., 2015a) also demonstrated that using TRMM-detected HTs in conjunction with other large-scale environmental indicators can improve the predictability of TC rapid intensification.

    Given the crucial role of HTs, it is necessary to analyze the favorable environments associated with their activities and explore the predictability of HTs to make weather warnings more effective and strengthen the capacity of communities to prepare for meteorological disasters in real time. (Rogers, 2010) indicated that the favorable environments and the prediction of CB initiation inside a TC are very worthy of research, from the perspective of forecasting rapid intensification. Nevertheless, it remains controversial as to whether HTs are predictable. Many researchers believe that HT occurrence is random and undetectable (Shin and Smith, 2008; Van Sang et al., 2008). Moreover, due to the large spatial and temporal variation of updrafts and their interactions with each other (such as merging and splitting), it is hard to trace the evolution of a specific HT (Chen and Zhang, 2013), let alone predict it. However, existing studies (Chen and Zhang, 2013; Jiang and Tao, 2014) also show that large-scale environmental information provides good indicators for the overall trend of HT activities within TCs. Large vertical wind shear, for example, can make it very difficult for HTs to occur in a TC (Jiang and Tao, 2014). On the contrary, warm sea surface temperature (SST) tends to increase convective instability and convective available potential energy, which are favorable for the development of deep convection that can penetrate the tropopause (i.e., HTs) (Chen and Zhang, 2013).

    On the basis of previous work, in this study we systematically examined the favorable environments associated with the HT activities in TCs. For this purpose, two types of data needed to be utilized: the environmental meteorological information of TC systems, and sufficient samples of HT observations. For the first type of data, the Statistical Hurricane Intensity Prediction Scheme (SHIPS; DeMaria et al., 2005) database was utilized. This database provides various synoptic weather indicators that describe the environmental conditions in areas surrounding TCs over major ocean basins. Regarding the second type of data, a key issue was to select sufficient samples of HT observations collected by reliable detection methods. The Cloud Profiling Radar (CPR) onboard CloudSat, the Precipitation Radar (PR) onboard the TRMM satellite, and the Dual-Frequency Precipitation Radar (DPR) onboard the Global Precipitation Measurement core satellite, can provide high-resolution vertical sounding and, as such, have been used in many studies related to HTs (e.g., Jiang and Tao, 2014; Zhuge et al., 2015a). However, as polar orbiters, CPR, PR and DPR do not produce sufficiently fine-scale temporal information necessary for monitoring and forecasting rapidly changing phenomena such as HTs. In addition, these polar orbiters have orbital gaps that sometimes provide no data at the location of TC development (Schmit et al., 2009). The temporal and spatial resolution of a geostationary satellite is less than 0.5 h and 5 km, respectively, which makes it able to collect spatially continuous observations and accurately capture the active state of short-lived systems such as HTs. The main disadvantage of geostationary satellite detection is the lack of vertical sounding ability. This is because cloud is opaque for the visible and infrared channels, which are the standard channels in operation on current geostationary satellites. Thus, geostationary satellites can only provide cloud-top information. Nevertheless, because the updrafts within HTs are so strong that a distinct dome-like protrusion called an overshooting top (OT; Bedka et al., 2010) will form above its cumulonimbus anvil (Glickman, 2000), OTs can be deemed as a proxy in satellite detection for the occurrence of HTs (Monette et al., 2012). The OT is colder than surrounding cirrus clouds and has a large texture. Considering these features, more recently, (Bedka et al., 2010) proposed an OT detection algorithm based on observations of the geostationary satellite infrared window (IRW) channel, called the "IRW-texture" algorithm. Originally, this method could only be applied over midlatitudes. However, it was later modified by (Monette et al., 2012) and can now be applied over tropical regions and to TC systems. The development of the algorithm that detects OTs in TCs based on observations of geostationary satellites, provided an important tool for the present study in producing a dataset of sufficient OT samples.

    This paper is organized as follows. Section 2 introduces the OT detection algorithm. The data and preprocessing methods are also introduced. The large-scale environmental characteristics associated with the occurrence of OTs are described in section 3. The results from two types of forecasting experiments and evaluation related to OTs are presented in section 4. Section 5 shows the environmental differences between OT activities and TC development. A summary and discussion is presented in section 6.

2. Data and methods
  • This study focuses on TCs that occurred from August 2005 to December 2012 over the western North Pacific (WNP) Ocean. This time period was covered by Multifunctional Transport Satellite (MTSAT)-1R and MTSAT-2. The data of the MTSAT IRW channel (10.8 μm) were obtained from Kochi University, Japan (via http://weather.is.kochi-u.ac.jp/sat), and had been remapped to a regular latitude-longitude grid with a spatial resolution of 0.05\r (5 km) at hourly intervals. In addition, the central position and intensity (1-min maximum wind speed) of these TCs were obtained from the 6-h interval best-track data provided by the JTWC (Joint Typhoon Warning Center) and then interpolated to an hourly resolution. Note that the TC samples with a current intensity lower than 34 kt (1 kt = 0.5144 m s-1), and those with a central position located over land, are not included in this study.

    Large-scale environmental variables were extracted from the SHIPS database (via http://rammb.cira.colostate.edu/research/tropical_cyclones/ships/developmental_data.asp), which contains a variety of weather information over TC surroundings at 6-h intervals. In SHIPS, atmospheric variables are calculated based on the NCEP (National Centers for Environmental Prediction) analysis fields. The oceanic variables were determined from the weekly SST analysis field (Reynolds and Marsico, 1993) and satellite altimetry observations (Mainelli-Huber, 2000).

  • The geostationary satellite-based OT detection algorithm, also known as the IRW-texture algorithm (Bedka et al., 2010; Monette et al., 2012), includes three major steps, given as follows:

    (1) Identification of pixels where the brightness temperature (BT) is below 215 K within a given IRW image, and the distance between any two of these pixels must be greater than three pixels (15 km). Otherwise, the warmer pixel will be ruled out. This step is conducted to ensure the same OT will not be identified twice. Those pixels identified during this step are referred to as the candidate OT center.

    (2) Sampling of the anvil cloud pixels (defined as BT < 225 K) surrounding a candidate OT center in a circular region with a radius of two pixels (10 km). A candidate OT that has at least nine anvil cloud pixels in 16 directions in the circular region will be retained. Otherwise, it will be discarded.

    (3) Calculation of the difference between the candidate OT central BT and the average anvil cloud BT (BT). If BT is greater than a certain threshold then the candidate OT will be finally identified as an OT. In (Bedka et al., 2010) and (Monette et al., 2012), the value of the BT threshold was 6.5 K and 9 K, respectively. Considering the fact that the tropopause in the tropics is relatively cold and high, we chose 9 K as the threshold.

    The IRW-texture algorithm was first applied in OT identification using observations from the Geostationary Operational Environmental Satellites (GOES). However, the method is also appropriate for the European MeteoSat Second Generation (MSG; Proud, 2015) and Japanese MTSAT (Bedka et al., 2012) satellites. According to the statistics calculated by (Bedka et al., 2012), the probability of detection (POD) and false alarm rate (FAR) of this detection algorithm when applied to MTSAT observations were 33.3% and 9.1%, respectively——slightly lower than when applied to GOES and MSG. There are two major reasons for the lower POD for MTSAT: first, the lowest BT of the OT in most cases is higher than the tropopause temperature obtained from numerical weather prediction (Bedka et al., 2012), due to the relatively coarse spatial resolution of MTSAT observations (i.e., an OT can exist at the sub-pixel scale); and second, the temporal resolution of MTSAT observations is 0.5 h, which is much longer than the life cycle of an OT. As a result, those OTs that occurred between two consecutive observations of MTSAT are more likely to be missed (Monette et al., 2012). Nevertheless, spatial and temporal mean values are used in the statistical analysis of this study (as described in section 2.3), which to the greatest extent eliminates the uncertainty due to the observational resolution of MTSAT, and thus ensures the credibility of this research.

  • Geostationary satellite IRW images with hourly resolution were matched with the best-track data that had been interpolated to hourly intervals. Using the information on TC central position, the IRW-texture detection method described in the previous section was applied to calculate the number of OT occurrences within a radius of 300 km from the TC center. The radius of 300 km is an empirical parameter for average TC size, as suggested by (Monette et al., 2012).

    In order to avoid observational errors at individual time levels, as well as ensure the temporal resolution was consistent with that of the SHIPS database (6 h), the number of OT occurrences three hours before and after every synoptic hour (i.e., 0000, 0600, 1200 and 1800 UTC) was averaged and then rounded down to produce the levels of OT activity at these synoptic hours. For example, if nine OTs occurred in a TC within the 6-h period during 0900-1500 UTC, then the level of OT activity at 1200 UTC was 1 (9/6) per hour. Ultimately, a dataset that includes the centers and intensities of TCs and the levels of OT activity in the WNP basin for eight continuous years was produced for further analysis. A total of 163 TCs and 2780 samples are involved in the dataset. Figure 1a depicts the frequency distribution of the level of OT activity. It was found that the proportion of samples of various OT activity levels decreased exponentially with the increase in its level. In particular, a level of OT activity greater than 1 corresponded to the 52nd percentile of overall activities, while those equal to or greater than 4 corresponded to the 90th percentile. Figure 1b shows the spatial distribution of the mean of these statistical samples of eight years (2005-12) in each 1°× 1° bin. It can be seen that the samples with a higher level of OT activity mainly occurred in the low latitudes (south of 15°N). This aspect is discussed in the next section.

    This paper mainly discusses the favorable environments for the occurrence of two issues based on the level of OT activity: whether an OT occurs, and how active OTs are. To address the first issue, the samples were divided into OT and non-OT cases according to whether the level of OT activity was equal to or greater than 1. To discuss the second issue, the samples were divided into high-activity-OT (abbreviated as HA-OT) and low-activity-OT (abbreviated as LA-OT) cases, based on whether the level of OT activity was equal to or greater than 4. The non-OT cases were excluded from the LA-OT cases. In order to study the predictability of OTs, the level of OT activity was only combined with SHIPS environmental variables 12 hours prior to the OT occurrence. This combination ensured that the average lead time of the forecast of the OT (HA-OT) case was up to 12 hours. Note that the forecast and evaluation in this study is for the overall trend of OT activities in TCs, not a particular OT.

    Figure 1.  The (a) frequency and (b) spatial distribution of the level of OT activity with 2780 samples. The level of OT activity is defined as the hourly mean number of OT occurrence during the three hours before and after the synoptic hour within the range of 300 km from the TC center. Values of the horizontal axis in (a) were obtained by rounding down. Different colors (b) represent the levels of OT activity and were averaged from all samples of eight years (2005-12) in each 1°× 1° bin. For example, if a total of three samples existed in a bin and the corresponding levels of OT activity are 3, 5 and 7, respectively, then the mean level of OT activity is 5 (15/3) in this bin.

    The SHIPS database contains more than 100 variables. It is impossible to take all the variables to establish a forecasting model. Therefore, it is necessary to screen the data first. In this study, the average magnitude of each variable in SHIPS was compared between the cases with and without OT (or, the cases with HA-OT and LA-OT) occurring within the years 2007-10. If the difference between the average magnitude was statistically significant at the 99.9% confidence level based on the two-sided Behrens-Fishers t-test (Dowdy and Wearden, 1991), this variable was saved for further analysis. The box difference index (BDI; Peng et al., 2012) was then used to sort the variables saved in the previous step. Those variables with large BDI values and independent and clear physical interpretations were determined to be the forecast factors. The BDI is calculated as follows: \begin{equation} \label{eq1} {\rm BDI}=\frac{M_{\rm A}-M_{\rm B}}{S_{\rm A}+S_{\rm B}} , (1)\end{equation} where the A and B subscripts represent OT (or HA-OT) and non-OT (or LA-OT) cases, respectively; and M and S represent the mean and standard deviation of a variable, respectively. The absolute value of the BDI represents the ability of a variable to distinguish two datasets. A larger value indicates a more distinct difference between the datasets and a greater potential for the corresponding variable to be a predictor. Ultimately, seven variables were determined for OT forecasting, and seven variables were also determined for HA-OT forecasting (Table 1). These variables are discussed in the next section.

3. Large-scale environmental conditions of the two forecasting issues
  • 3.1.1. Statistical differences in each predictor

    Table 2 shows the mean values of the seven predictors and their differences corresponding to OT and non-OT cases and the absolute value of the BDI. The samples from 2007 to 2010 are involved in the statistics.

    It can be seen that the temperature difference between 150 hPa and 1000 hPa (TD15) and the oceanic heat content (OHC) were significantly larger for OT cases than that for non-OT cases. This result indicates that the cases with OT were associated with an environment that had a deep enough layer of warm water at the bottom and an evident temperature difference between the upper and lower atmosphere. This finding is similar to results from statistical analysis in previous studies (e.g., Chen and Zhang, 2013), which suggested that a warm SST is favorable for the occurrence of OT-like deep convection. The optimal oceanic and atmospheric environments that favored the occurrence of deep convection appears in the afternoon (Chen and Houze, 1997). However, due to the enormous heat capacity of the ocean, the ocean cools more gradually after sunset, and thus enables deep convection throughout the night. The peak OT activity usually occurs at night and in the early morning (Liu and Zipser, 2005; Proud, 2015). The temporal difference between the optimal environmental condition and the peak OT occurrence is about 12 hours, consistent with the OT forecast hours attempted in this study.

    The vertical velocity averaged from 0 to 15 km (VVAC) was estimated based on the difference between a parcel lifted from the surface and its environment at each atmospheric level. A positive VVAC meant that the lifted air parcel was able to maintain upward motion. Therefore, VVAC represents the level of atmospheric instability. VVAC is actually a thermodynamic variable, although it seems like a dynamic one. The VVAC magnitude for the cases with OT was about 1.6 times that for the cases without OT (11.29 m s-1 versus 6.95 m s-1). Such an obvious difference suggests that the atmospheric instability is indeed much greater with OT cases.

    The relative humidity at the middle (RHMD) and upper troposphere (RHHI) was also significantly different between OT and non-OT cases. Although convective initiation favors an environment where the humidity in the middle troposphere is less than that in the lower troposphere, a large mid to upper tropospheric humidity can help shallow convection develop into deep convection (Zhang et al., 2010), even overshooting convection.

    The TC central latitude (LAT) was more southerly for the OT cases than that for the non-OT cases. This is consistent with the spatial distribution of the mean level of OT activity shown in Fig. 1b. LAT is a composite indicator that reflects the distribution of SST, atmospheric moisture, tropopause level, and other environmental variables that may related to OT activity.

    The duration of a TC with intensity greater than 35 kt (HIST) is also an important predictor. Table 2 shows that OT cases mostly occurred in the early stage of the life cycle of a TC. Because TCs often form at low latitudes and then move towards high latitudes, HIST may be correlated with LAT. Specifically, when a TC initially forms and begins to develop, it is likely to be situated in the low latitudes. At this time, the large-scale environment favors OT occurrence. When a TC matures or weakens, it is often located in the high latitudes. The environment no longer favors OT occurrence, and thus OT activity will be suppressed.

    Figure 2.  Frequency distribution of seven predictors for the OT and non-OT cases: (a) TD15; (b) OHC; (c) VVAC; (d) LAT; (e) RHHI; (f) RHMD; (g) HIST. See Table 1 for definitions of the predictors.

    3.1.2. Frequency distribution of each predictor

    In the above section we investigated the statistical differences in the seven variables listed in Table 2 between OT and non-OT cases in the 12 hours before OT occurrence. The frequency distributions of these seven variables for OT and non-OT cases are presented and compared in Fig. 2. Specifically, OTs were more likely connected with large values of TD15, considering the fact that approximately 88% of OT cases occurred when the TD15 was above 93°C. Furthermore, when TD15 was greater than 94.10°C, the frequency of OT cases was nearly 4.5 times higher than that of non-OT cases (Fig. 2a). As for OHC, taking 75 kJ cm-2 as a reference value, its left side corresponded to the majority (over 78%) of non-OT samples, while its right side corresponds to a high concentration (over 63%) of OT occurrence. The OT occurrence was particularly significant when the value of OHC was equal to or greater than 87.01 kJ cm-2 (Fig. 2b). It was also found that most VVAC values were equal to or larger than 10 m s-1 when OT cases occurred. The difference in the frequency between the OT cases and non-OT cases became more significant (44.5% versus 10.7%) when the VVAC was larger than 12 m s-1 (Fig. 2c). Besides, large LAT values were often found in non-OT cases, and about 61% of non-OT cases occurred in the area north of 20°. When LAT was less than 16°, the frequency of OT cases was threefold higher than that of non-OT cases (Fig. 2d). Differences in RHHI and RHMD between the OT and non-OT cases were also evident, with a large proportion of OT cases associated with large RHHI and RHMD (Figs. 2e and f). When RHHI was greater than 70%, the frequency of OT cases was much higher than that of non-OT cases (41.7% versus 16.5%). Similarly, when RHMD was greater than 70%, the frequency of OT (non-OT) cases was 64.6% (29.8%). In addition, the cases with OT were skewed towards low HIST. In particular, less than 5% of OT cases occurred when the HIST was greater than 6 days.

    Figure 3.  The probability of OT occurrence for each predictor when they met (gray) and did not meet (black) their specific OT thresholds. The OT thresholds of each of the predictors are also presented. OT samples during 2007-10 were used to obtain the OT thresholds of each predictor, and then determine the individual probability of OT occurrence. See Table1 for definition of the predictors.

    3.1. Probability of OT occurrence

    Next, following the method described in (Kaplan and DeMaria, 2003), the aforementioned seven variables were taken as predictors to estimate the probability of OT occurrence. For each predictor, the OT threshold was defined as the sample mean of all OT cases. If the variable was either greater than or less than the OT threshold, whichever was suitable for OT occurrence, then the threshold was said to be met. For example, the OT samples had a mean OHC of 87.01 kJ cm-2, and thus the OHC OT threshold was met when it was greater than 87.01 kJ cm-2. Otherwise, it was not. Furthermore, among the 426 samples meeting the OHC OT threshold, 356 samples showed OT occurrence, and so the probability of OT occurrence was determined to be 83.6% (i.e., 356/426) when the OT threshold for OHC was met. The same approach was applied for all of the remaining predictors. Figure 3 shows the probability of OT occurrence for each predictor when they met (gray) and did not meet (black) their specific OT thresholds. The climatological (sample mean) probability of OT occurrence was 57.1% (710/1243), which was taken as a baseline. Therefore, from the results presented in Fig. 3, it was found that, for every predictor, the probability of OT occurrence was much higher when the predictor met its corresponding OT threshold, as compared to when the predictor did not meet the OT threshold. All predictors showed good forecasting ability.

  • 3.2.1. Statistical differences in each predictor

    Table 3 lists relevant information on the seven variables for the HA-OT and LA-OT cases. In contrast to Table 2, TD15, VVAC, LAT, HIST and RHHI were kept, but other large-scale thermal environmental variables (such as OHC and RHMD) were discarded, because of their small differences between the HA-OT and LA-OT cases. The vertical wind shear (SHRD) and TC wind intensity (VMAX) were newly-added. The occurrence of OT was mainly controlled by the thermal variables (i.e., OHC and RHMD). However, if these critical factors reached a certain degree of intensity, other seemingly unimportant variables (i.e., VMAX, SHRD) became important and determined whether the occurrence of OT was active. In addition, the absolute values of the BDI decreased significantly from 0.55 to 0.25, indicating that the differences in large-scale environmental characteristics between HA-OTs and LA-OTs were not as significant as those between OTs and non-OTs.

    The OT activity was partially controlled by SHRD. A weaker SHRD was more conducive to the eruption of OT than a strong SHRD. (Jiang and Tao, 2014) utilized 12 years of TRMM data to demonstrate that a large SHRD can cause a convective system to tilt severely when the convection is not strong enough, inducing an entrainment that is adverse for the updraft to reach or exceed the height of the tropopause.

    VMAX, a TC intensity parameter, was also an important predictor for the HA-OT cases. The statistical analysis in Table 3 demonstrates that there was a link between the HA-OT cases and a low VMAX. (Tao and Jiang, 2013) showed that the frequency of OT occurrence is negatively correlated with TC intensity. (Zhuge et al., 2015a) also indicated that the probability of OT occurrence decreases with an increase in TC initial intensity in the WNP. The underlying mechanism may be associated with the rapid filamentation process occurring outside the radius of maximum wind. Rapid filamentation is mainly characterized as strain-dominated flows. The rapid filamentation process is less likely to occur in weak tropical disturbances, but is almost always found in intense typhoons/hurricanes (Rozoff et al., 2006). It can suppress deep convection both in the spiral band and eyewall region (Wang, 2008; Kuo et al., 2012).

    Figure 4.  Frequency distribution of seven predictors for the HA-OT and LA-OT cases: (a) VVAC; (b) TD15; (c) LAT; (d) HIST; (e) RHHI; (f) SHRD; (g) VMAX. See Table 1 for definitions of the predictors.

    3.2.2. Frequency distribution of each predictor

    Figure 4 exhibits the frequency distributions of the variables listed in Table 3. The general patterns of VVAC, TD15, LAT, HIST and RHHI are identical to those shown in Fig. 2. However, the overall distributions of the data are shifted. This is because the values of related variables increased or decreased following the increase in the level of OT activity. For example, the probability of occurrence was about 74.5% (39.2%) for OT (non-OT) cases when LAT was lower than 20°, whereas it was about 83.7% (71.6%) for HA-OT (LA-OT) cases. The difference in the frequency between the HA-OT and LA-OT cases was more significant (57.0% versus 36.1%) when LAT was lower than 16° (Fig. 4c). In addition, HA-OT samples were associated with small SHRD, and the frequency of cases with LA-OT was 2.6 times higher than that of the cases with HA-OT when SHRD was above 9 m s-1. The HA-OT cases were also linked with small VMAX; the frequency of HA-OT (LA-OT) cases was about 68.0% (51.5%) when VMAX was below 30 m s-1.

    3.2.3. Probability of HA-OT occurrence

    The seven typical variables mentioned in Table 3 were used as predictors, and the method described in section 3.1.3 was applied to estimate the probability of HA-OT. Figure 5 shows the probability of HA-OT occurrence when each predictor's threshold was met or not met. The sample mean of HA-OT occurrence was 24.2% (172/710). Figure 5 suggests that the probability of HA-OT is generally small when all predictor thresholds are met, but it is far greater than the probability of HA-OT when predictor thresholds are not satisfied. It is also greater than the climatological probability of HA-OT. These predictive factors should exhibit relatively good forecasting ability.

    Figure 5.  As in Fig. 3, but for the HA-OT cases.

4. Forecasting experiments and evaluation
  • In this section, two types (i.e., OT and HA-OT) of forecasting experiments were implemented. Their forecasting skill was evaluated with four metrics (Wilks, 2006): POD, FAR, probability of false detection (POFD), and Peirce skill score (PSS). The definitions of the four metrics are as follows: \begin{eqnarray} {\rm POD}&=&\dfrac{h}{h+m} ;(2)\\ {\rm FAR}&=&\dfrac{f}{h+f} ;(3)\\ {\rm POFD}&=&\dfrac{f}{f+c} ;(4)\\ {\rm PSS}&=&{\rm POD}-{\rm POFD} ; (5)\end{eqnarray} where h(m) represents the number of OT/HA-OT cases that are correctly forecasted (missing), while c(f) is the number of non-OT/LA-OT cases that are correctly forecasted (false-alarmed). PSS was employed to evaluate the overall skill of the dichotomous forecast. The greater the PSS, the better the forecast. The PSS is 1 for a perfect forecast, 0 for a random forecast, and -1 for the worst forecast.

    The statistical forecasting model used was as follows. When at least N OT (HA-OT) predictor thresholds were satisfied, the OT (HA-OT) occurrence was determined, where N is the number of thresholds met, ranging from 0 to 7. The thresholds were obtained from samples for the period 2007-10. The samples within the years 2005-06 and 2011-12 were employed in the evaluation.

    Figure 6 shows the results of the OT forecasts. It can be seen that POD, FAR and POFD gradually reduced with the increase in the value of N, and the overall change was large. Note that PSS first increased and then decreased. When there were at least three OT predictor thresholds satisfied, PSS reached a maximum (0.49), while POD was 74.2% and FAR was 33.1%.

    As such, Fig. 7 shows the effect of the HA-OT forecasts. The overall performance of each metric was consistent with the results shown in Fig. 6. Note that FAR remained at a high level and the downward trend was very slow. When more HA-OT predictor thresholds were satisfied, POD decreased evidently. At the same time, the overall level of PSS was also relatively low. When there were at least three HA-OT predictor thresholds satisfied, PSS reaches a maximum value (0.30), while POD was 86.9% and FAR was 73.4%.

    Figure 6.  OT prediction statistics, where the horizontal axes indicate the minimum number of OT predictor thresholds satisfied: (a) POD (probability of detection); (b) FAR (false alarm rate); (c) POFD (probability of false detection); (d) PSS (Peirce skill score).

    Figure 7.  As in Fig. 6, but for the HA-OT cases.

5. Environmental differences between OT activity and TC development
  • It is worth noting that some of the predictors used in this study were also applied in a forecasting model of TC formation (Fu et al., 2012) and rapid intensification (Kaplan et al., 2010; Rozoff and Kossin, 2011; Shu et al., 2012). This may again draw the debate of whether the OTs are a precursor or a byproduct during the process of TC formation and rapid intensification. Because of this, we further examined the relationships between OT activity and TC intensity and intensity change. From Fig. 8, the OT activities were more influenced by the TC intensity 12 hours prior to OT occurrence (VMAX -12h; Figs. 8a and b) than the intensity at the time of OT occurrence (VMAX 0h; Figs. 8c and d). The BDI of VMAX 0h for the OT cases was 0.01, indicating that the relationship between OT and TC current intensity is not close (Zhuge et al., 2015b). In contrast, the BDI of VMAX -12h for the HA-OT cases was 0.18——ranked seventh out of the HA-OT factors listed in Table 3. Moreover, OT activity was less influenced by the intensification rate prior to OT occurrence. When the average magnitudes of the TC intensity change from -24 h to -12 h ( VMAX -12h- VMAX -24h) between the cases with and without OT (or, the cases with HA-OT and LA-OT) were compared, the difference was not as significant as for other variables. The lowest BDI for the OT predictors listed in Table 2 was 0.45, but the BDI value for VMAX -12h- VMAX -24h was only 0.35 (Fig. 9a). However, Figs. 9c and d show that a TC may undergo intensification after an OT erupts, since the BDIs of TC intensity change within the following 24 hours of OT occurrence ( VMAX +24h- VMAX 0h) increased to 0.42 and 0.22 for OT and HA-OT occurrence, respectively.

    Figure 8.  Frequency distribution of TC wind intensity (a, b) at the analysis time (VMAX -12h) and (c, d) at the time of OT occurrence (VMAX 0h) for (a, c) the OT and non-OT cases or (b, d) the HA-OT and LA-OT cases. The number of asterisks indicates the result is statistically significant at the 99.9% (***), 99% (**) and 90% (*) confidence level, based on the two-sided Behrens-Fishers t-test.

    Figure 9.  As in Fig. 8, but for TC intensity change (a, b) within the 12 hours before the analysis time ( VMAX -12h- VMAX -24h) and (c, d) within the following 24 hours of OT occurrence ( VMAX +24h- VMAX 0h).

    Besides, although some predictors are employed by forecasting models for both OT activity and TC development, their weightings are different. Our study indicates that thermodynamic variables are dominant in controlling the occurrence of OTs and HA-OTs. Note that at least five variables are thermodynamic (e.g., TD15, VVAC, OHC, RHHI, RHMD) while one variable is dynamic (SHRD) among the environmental factors listed in Table 1. Meanwhile, it can be seen from Tables 2 and 3 that the importance of thermodynamic factors, which is indicated by the absolute value of the BDI, is larger than that of dynamic factors. Therefore, OTs occurring within a TC keep a closer relationship with thermodynamic processes, and are linked to an environment where thermodynamic variables meet their thresholds easily. This is different from the favored environment for TC genesis and development. Fu et al. (2012) investigated the environmental conditions for TC genesis in the WNP and reveled that dynamic variables are more important (Fu et al., 2012, Table 3). Besides, (Shu et al., 2012) analyzed the large-scale environmental characteristics of TC rapid intensification in the WNP and suggested that the relative weights of dynamic factors (like SHRD) are larger than those of thermodynamic factors (like SST) (Shu et al., 2012, Fig. 7).

6. Summary and discussion
  • We studied the level of OT activity of all TCs in the WNP basin from 2005 to 2012 (within a radius of 300 km of the TC center). Based on the level of OT activity, all the samples were divided into OT and non-OT cases or HA-OT and LA-OT cases. At the same time, the samples within the four years from 2007 to 2010 were used to investigate the large-scale environmental characteristics 12 hours in advance of OT and HA-OT occurrence, and the samples for the remaining four years (2005-06 and 2011-12) were applied for forecasting experiments and evaluation.

    These preliminary results suggest that the state of the large-scale environment is important for two forecasting issues. For both OT and the HA-OT forecasts, their respective seven predictors identified in this study are strongly indicative of the overall possibility of their occurrence over certain areas, and that means the variables in Table 2 (Table 3) could represent the favorable environments for OT (HA-OT) occurrence. Further analysis indicated that OT occurrence is mostly embedded in an environment with a deep layer of warm water and a large temperature difference between the upper- and lower-level troposphere, while large atmospheric humidity and atmospheric instability are also prerequisites for the occurrence of OT. Furthermore, most OTs occur in the early period of a TC's life cycle and over low latitudes. The more favorable the environment, the higher the level of OT activity. Thereby, some environmental conditions for HA-OT appear to be more rigorous than for OT (such as TD15, VVAC, etc.). Note that the environmental conditions between HA-OT and LA-OT cases have certain similarities, which leads to the forecasting of HA-OT being slightly worse than that of OT. The final evaluation demonstrated that, when at least three OT predictor thresholds were satisfied, the maximum PSS was 0.49 and the corresponding POD and FAR were 74.2% and 33.1%, respectively. Similarly, when at least three predictor thresholds were satisfied, the PSS of HA-OT prediction reached a maximum (0.30), while POD and FAR were 86.9% and 73.4%, respectively.

    The statistical forecasting model developed in this study is relatively simple. To improve the forecasting of OT and HA-OT, the methods described in Kaplan et al. (2010, 2015) will be considered in a future study. These methods include calculating the weight of each predictor and establishing an objective statistical model (e.g., logistic regression model). Besides, we noted that there was interannual variability in the climatology of OT occurrence, and the sample size of the OT cases over 2005-08 was higher than that over 2009-12 (figures and tables omitted). The reasons for such problems are not yet known and will be discussed in our future research.

Reference

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return