Overshooting cloud tops (OTs) form in deep convective storms when strong updrafts overshoot the tropopause. An OT is a well-known indicator of convective updrafts and severe weather conditions. Here, we develop an OT detection algorithm using thermal infrared (IR) channels and apply this algorithm to about 20 years' worth of MODIS data from both Terra and Aqua satellites to form an extensive, near-global climatology of OT occurrences. The algorithm is based on a logistic model which is trained using A-Train observations. We demonstrate that the overall accuracy of our approach is about 0.9 when the probability of the OT candidates is larger than 0.9. The OT climatology reveals a pattern that follows the climatology of deep convection and shallow convection over the midlatitude oceans during winter cold-air outbreaks. OTs appear most frequently over the Intertropical Convergence Zone (ITCZ), central and southeastern North America, tropical and subtropical South America, southeastern and southern Asia, tropical and subtropical Africa, and northern middle–high latitudes. OT spatial distributions show strong seasonal and diurnal variabilities. Seasonal OT variations shift with large-scale climate systems such as the ITCZ and local monsoonal systems, including the South Asian monsoon, North American monsoon, and West African monsoon. OT diurnal variations agree with the known diurnal cycle of convection. Maximum OT occurrences are in the afternoon over most land areas and around midnight over ocean, and the OT diurnal cycle is stronger and more varied over land than over ocean. OTs over land are usually colder than over ocean, except at around 10:30 LT (Equator-crossing time). The top 10 coldest OTs from both Terra and Aqua mostly occur over land and at night. This study provides OT climatology for the first time, as derived from 2 decades of MODIS data, that represents the longest and stable satellite records.
An overshooting cloud top (OT) forms when a convective storm updraft penetrates the level of neutral buoyancy and thus extends into the upper troposphere and lower stratosphere (UTLS). OTs and their associated strong updrafts have been found to be an important transport mechanism for water vapor and other atmospheric constituents into the stratosphere, thus impacting the chemical composition and radiation budget of the UTLS (e.g., Gettelman et al., 2002, 2004). They are often used as indicators of hazardous weather conditions such as strong winds, large hail, flooding, and tornadoes at the Earth's surface (Bedka et al., 2018; Dworak et al., 2012; Marion et al., 2019). More generally, the characteristics of OTs express information about the characteristics of the related updrafts well below cloud top, including the convective mass flux through the troposphere, which is an important parameterized quantity used in global climate models.
In addition to the expectation of a connection between updraft strength and OT depth (Heymsfield et al., 2010), Trapp et al. (2017) have shown a strong link between updraft core area and OT area (OTA), indicating that a relatively intense and wide mid-tropospheric updraft core area will tend to have a large OTA. Given that the direct measurements of updrafts within intense convective environments are either from a few ground-based radars or several field campaigns, these studies suggest a pathway for characterizing global updraft and updraft size distributions by quantifying the global OT distributions and characteristics from space.
To this end, the first step is to detect OTs. Geostationary satellite
imagery provides the opportunity to study OT occurrence over a wide region
with fine spatial and temporal resolutions. A series of OT detection
algorithms have been developed based on geostationary satellite
observations. A commonly used OT detection method utilizes the brightness
temperature (
Another commonly used OT detection method is the IRW texture approach (Bedka et al., 2010). This method uses a threshold of 215 K
Observations from spaceborne active sensors have also been used for exploring OT detections. For instance, the cloud-profiling radar (CPR) on CloudSat (Stephens et al., 2008) was used for validating the passive satellite-based OT detection methods (Bedka et al., 2010; Dworak et al., 2012; Rysman et al., 2017), calculating the heights of OTs (Griffin et al., 2016), and understanding WV-IRW BTD variability in OT regions (Setvák et al., 2013). The combined CloudSat–CALIPSO (Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observation) data were also used for detecting OTs, which led to the creation of a 12-year OT database (Li et al., 2022). As demonstrated by these studies, the CloudSat–CALIPSO observations are powerful in detecting OTs and gauging OT depths, but they are only available in a narrow swath that leads to a lack of knowledge of three-dimensional (3-D) OT structures and large uncertainties in their coverage (Astin et al., 2001). The precipitation radar on the Tropical Rainfall Measuring Mission (TRMM) or Global Precipitation Measurement (GPM) mission can provide 3-D depictions of storm structures. The precipitation radar observations have been used to investigate OT climatology, including their geodistributions, area, and diurnal cycles, in the tropical regions (20
In addition, using three water vapor channels of the Advanced Microwave Sounding Unit-B (AMSU-B), the convective overshooting detection method was developed through the microwave technique (Hong et al., 2005). A 7-year OT climatology based on AMSU-B was derived in the tropical and subtropical areas that shows OT diurnal and interannual variations (Hong et al., 2008).
While many OT detection algorithms have been developed using either passive
or active remote sensing techniques, their use toward quantifying OT
occurrences and attributes from space is mostly from datasets with coarse
spatial resolutions, e.g.,
The MODIS instrument (King et al., 1992) acquires data at a high spatial resolution (
To utilize these climate records, the main objective of this study is to show a near-global climatology of OT occurrence derived from about 20 years of Aqua and Terra MODIS data. Owing to the relatively high spatial resolution of MODIS, this climatology includes OTs in a small size that are missed by GPM radar. It includes both the tropical and midlatitude regions and thus complements the climatology by Liu and Zipser (2005) that was only focused on tropical and subtropical regions. It also provides OT diurnal information at four observation times. To achieve these objectives, we first develop an OT detection algorithm that is specifically designed for MODIS, works for both daytime and nighttime data, and is more flexible to thresholds compared to those used in Bedka et al. (2010) and Li et al. (2022). In Sect. 2, we will present the details of data and methods used for developing the OT detection algorithm. Validation of the algorithm will be discussed in Sect. 3. Section 4 discusses the results produced from our OT detection algorithm. Finally, in Sect. 5, we conclude with the findings of this study.
In order to develop a method that can detect OTs during both daytime and nighttime, this study uses observations from multiple sensors on board multiple platforms in addition to a machine learning method called logistic regression. The OT detection algorithm is developed in two main steps. First, we manually identified a number of OT candidates from the combined CloudSat–CALIPSO data. The infrared radiative characteristics of these OTs extracted from the combined Aqua MODIS infrared data serve as inputs to train the logistic regression. Second, we applied the regressed model to the Terra and Aqua MODIS data for automatic OT detection. We call this method an IR algorithm.
The CloudSat and CALIPSO satellites are two members of the afternoon
constellation in a sun-synchronous orbit with an Equator-crossing time at
01:30 and 13:30. The cloud-profiling radar (CPR) on board
CloudSat is a near-nadir-view radar operated at 94 GHz (
The CALIPSO flew about 15 s after CloudSat during the time period of observations used in this work. The lidar on board CALIPSO operates at 532 nm, having a vertical resolution of 30 m below 8.2 km and 60 m above 8.2 km (Winker et al., 2003). The lidar is sensitive to optically thin clouds and aerosols. The 2B-CLDCLASS-LIDAR product, provided by the CloudSat Data Processing Center, reports cloud-top and cloud-base heights for up to five layers (Wang et al., 2012). This product utilizes the complementary features of the CloudSat radar and the CALIPSO lidar and thus includes thin cirrus clouds. The cloud-top height of the topmost layer was used to aid in the identification of OTs. A total of 2 years of 2B-GEOPROF and 2B-CLDCLASS-LIDAR data (2007–2008) were used in this study.
MODIS, on board both the Aqua and Terra platforms, has 36 discrete spectral bands between 0.415 and 14.235
To obtain OT radiative characteristics, the MODIS Collection 6.1 Level-1B calibrated radiance data, with MYD021KM from Aqua and MOD021KM from Terra, were used. In this study, the bands selected have a center wavelength at 6.715 and 11.03
The Global Precipitation Measurement (GPM) Core Observatory, launched in February 2014, carries the first space-borne Dual-frequency Precipitation Radar (DPR) that includes a Ka-band (35.5 GHz) radar (KaPR) and a Ku-band (13.6 GHz) radar (KuPR; Hou et al., 2014). The KuPR measures 3-D structures of convective systems, with a vertical resolution of 250 m and a footprint of 5 km over a swath of 245 km. The GPM KuPR echoes have been demonstrated to be effective in the study of deep convection reaching to the tropopause (Liu et al., 2020; Liu and Liu, 2016). To utilize the GPM as an independent detection of OTs, we colocated the Ku-band echoes with the linear structure the OT candidates identified from Terra MODIS as a validation of our IR algorithm (Sect. 2.2). About 6 years (March 2014–December 2020) of GPM data (2A.GPM.DPR.V8) were used.
Tropopause temperature is needed for our IR algorithm. We used the
tropopause information output from the Modern-Era Retrospective analysis for
Research and Applications, Version 2 (MERRA-2), instantaneous, two-dimensional collection, hourly, and single-level diagnostics (MERRA2_400.inst1_2d_asm_Nx) product (Bosilovich et al., 2016). The
MERRA-2 parameter of TROPT is a blended estimate of tropopause temperature
(
The first step in the IR algorithm is to generate an OT training dataset. We
manually selected OT candidates from around the world between 2007 and 2008 by visualizing the CPR reflectivity factor from 2B-GEOPROF, topmost cloud-top
height from 2B-CLDCLASS-LIDAR, tropopause information from MERRA-2, and the
colocated
Parallax correction was examined but not employed in this work. We found
that a parallax correction produced
OTs were selected by visually inspecting the visualization rather than using
a fix criterion. The OT selection basically followed four principles, namely
The total of 287 samples were randomly distributed over four seasons and in different locations on Earth. The data are available in the Supplement (File S2).
An OT case occurring at night over the Indian Ocean on 1 June 2007.
Once an OT was manually selected from the A-Train data, OT edges were
determined using the method described in Marion et al. (2019). Briefly, the
local minimum
The cirrus (Ci) anvil in this work was searched within 20 pixels around the
OT center but with the OT area excluded. Pixels starting from the OT edge
and having
For the 209 OT candidates, all of them have diameters of less than 25 km, 180 OTs (86 %) have diameters of less than 15 km, and the peak in the OT diameter distribution is about 10 km (Fig. 2a), which is in agreement with Bedka and Khlopenkov (2016), who stated that OTs are typically less than 15 km in diameter. The
Similar to Bedka and Khlopenkov (2016), a probability was generated for an OT candidate. The 209 OTs and 78 NOTs selected from A-Train observations served as inputs for the logistic model. The logistic regression is a statistical model that is used to model a certain event through assigning a probability between 0 and 1, such as for the classification of OT and NOT. The logistic model depends on several variables or predictors, which are shown as follows:
Three MODIS-based variables were settled on after a series of tests to
optimize the accuracy. They are
A summary of the regressed coefficients for the variables selected for OT detection used in Eq. (1).
The logistic regression in Sect. 2.2.3 forms the basis of our IR algorithm, which aims to automatically identify OTs from Terra and Aqua MODIS in the daytime and at nighttime. The application of the IR algorithm starts from the pixel search, with
A window size of 41 km was adopted, considering that 98 % of the OTs
(Fig. 2) have diameters of less than 20 km, according to A-Train
observations (Sect. 2.2.2). This window makes sure that two OT centers are
at least 20 km apart and that enough pixels contribute to the cirrus anvils. If multiple OTs occurred in the same window, then the one with the coldest
Flowchart for the application of IR algorithm to MODIS data. The
GPM has been demonstrated to be an effective tool in studying intense storms and overshooting top events (Hourngir et al., 2021; Liu et al., 2020; Liu and Liu, 2016). Here, we used the GPM observations for two purposes, namely to compare the performance of OT detection between GPM KuPR and Terra MODIS and to investigate the cloud structure of detected OTs. The colocation between GPM KuPR and MODIS data was achieved when the time difference between them was within 5 min and the spatial difference between them was less than 10 km. A 5 min time window was used because the life cycle of OTs can be as small as several minutes (Setvák et al., 2013). The colocating process was performed only when OT candidates were identified from Terra MODIS. We obtained 6949 colocations for the period of March 2014–December 2020.
Ku-band radar, with reflectivity factor (Ze) in an area with a radius of less than 40 km around the colocated radar pixel, was collected to construct the contoured frequency by altitude diagram (CFAD; Yuter and Houze, 1995). The parallax error between KuPR and MODIS could be more than 20 km, according to the method described in Wang et al. (2011). Also, the OT diameter is likely less than 20 km. An area with a 40 km radius for the colocated KuPR data is likely able to encompass the OT event identified by MODIS. Figure 4 shows the CFADs contributed by all (6949) colocated OT cases. The CFADs were segregated into five OT probability intervals for the tropical and midlatitude areas. As shown, the largest frequency occurs above 5 km in tropical areas (Fig. 4a1–a5). As the OT probability increases, the frequency increases for large Ze (
Contoured frequency by altitude diagram, showing the frequency
normalized by the maximum bin of radar reflectivity. Data were binned at
1 dBZ intervals at each level. The upper panels are for the tropics (within
25
To compare the performance of OT detection between GPM and MODIS, we need to
determine when GPM detects an OT. If the maximum altitude of 15 dBZ in the
40 km radius area was higher than 2 km below the MERRA-2 tropopause, an OT
flag was assigned to the colocated GPM pixel. Previous studies also adopted
a level below the tropopause as the OT reference, considering the tropopause
height variability (Sun et al., 2019; Zhuge et al., 2015), such as the noted double tropopause observed in deep convection (Vergados et al., 2014). Here, 2 km was selected due to an agreement of 67 % between MERRA-2 and ERA5 tropopause height (from ECMWF-AUX; Partain, 2007) for the 287 OT and NOT cases used in Sect. 2. Once OT flags were assigned to the colocated GPM cases, an agreement of OT detection between MODIS and GPM was calculated for a wide range of OT probability generated by the IR algorithm. The agreement is expressed as follows:
Figure 5 shows the agreement in the OT detection between MODIS and GPM, which
increases with OT probability. In the tropics, the agreement is about 70 % when
Comparison of OT detection between GPM and Terra MODIS. Curves
represent the agreement of OT detection between MODIS and GPM in various
probability intervals (red for the tropics and blue for the midlatitudes). The numbers of the potential OT candidates are shown in bars.
As a complement to the GPM-MODIS comparison for assessing IR algorithm accuracy, we manually checked 1158 daytime OT candidates (selected randomly across the year) from Terra MODIS from 2018–2020 (data are available in the Supplement in File S3). These OT candidates have a wide range of probability. OT and NOT flags were assigned to the candidates by visually inspecting the IR and visible images from the NASA Worldview website (
Fraction of OT candidates with a wide range of probability in the
tropics
Overall, we choose a
In this section, we show an OT climatology of those OT candidates with
Before showing the climatology, we first show four cases, including all OT candidates with a variety of probabilities, for a detailed view of the performance of our IR algorithm in different storm environments.
Figure 7 shows the visible reflectance overlapped with OT centers, which are colored by OT probability.
Overshooting tops in tropical cyclones (TCs) are common. They are found to be closely linked to intense convection and rapid intensification in TCs (Griffin, 2017; Monette et al., 2012; Tao and Jiang, 2013). Figure 7a1–a4 display a tropical cyclone over the northern Indian Ocean on 8 November 2019. OTs are detected in the area, with very cold
In the mesoscale convective system case (Fig. 7b1–b4), OTs are detected in
the clusters that are associated with cold
Cold-air outbreaks can produce shallow convection when cold air blows from
frozen surfaces to the warmer ocean. The Cold-Air Outbreaks in the Marine
Boundary Layer Experiment (COMBLE) found that these convective clouds are
commonly lower than 5 km associated with updrafts of 4–5 m s
Midlatitude winter cyclones are associated with mostly stratiform cloud
systems (Stewart et al., 1998), as also demonstrated by the GPM rain type that shows mostly stratiform precipitation (Fig. 7d1–d4). The tops of the stratiform clouds associated with the fronts usually reach the tropopause without strong convective cores. However, they can be associated with lightning and heavy precipitation when fueled by potential instability, with updrafts of 6–8 m s
Four selected cloud systems with OTs detected by our IR algorithm. The first column shows the reflectance at 0.65
Figure 8 shows the seasonal distributions of OT occurrences contributed by
those OT candidates with
Aqua MODIS also shows frequent OT occurrences over the southeastern United States, which is associated with the afternoon convection. In regions over the southwestern USA and northwestern Mexico, OTs are detected, which are associated with the summer North American monsoon (Adams and Comrie, 1997).
During December–February (DJF; Fig. 8d), OT occurrences are about 44 % at 10:30 LT (Terra Equator-crossing time) and 36 % at 01:30 LT (Aqua Equator-crossing time) less than that in JJA. OTs are primarily located over the Southern Hemisphere as the ITCZ moves to the south of the Equator. A large number of OTs are detected by Aqua MODIS over tropical and subtropical South America and Africa. In the Northern Hemisphere, OTs become infrequent over land. Note that ice clouds have an occurrence frequency of about 70 % over middle- and high-latitude Asia during winter (e.g., Hong and Liu, 2015), which often poses challenges for OT identification. These cold ice clouds are rarely classified as OTs in our analysis, demonstrating the ability of our IR method to avoid the misclassification of cold ice clouds as OTs. In contrast, over the midlatitude ocean in winter, we see some OT occurrences. These OTs are associated with isolated convective clouds occurring in the cold-air outbreaks, as discussed in Sect. 4.1. These OTs are also observed over Southern Ocean during JJA (austral winter). We also notice a small number of OTs extending from northwestern to southeastern North America in DJF. These OTs are associated with the convection in winter midlatitude cyclones, as discussed in Sect. 4.1.
Convective activity over land is weak at the Terra overpass time in the morning (
The global distributions of OT occurrences derived from Terra and
Aqua MODIS in four seasons.
This section discusses OT diurnal cycles based on the four observation times
by Aqua and Terra MODIS. The OT occurrences in the daytime (
To better view the OT diurnal cycles, Fig. 10 shows when maximum and minimum OT occurrences occur within the four observation times. The diurnal cycle intensity, defined by the difference in maximum and minimum OT numbers normalized by the mean, is shown in Fig. 10e and f. As expected (Fig. 10a and b), the largest OT occurrences over land occur at about
The diurnal cycles of OT occurrences over ocean are generally weak (Fig. 10e and f) and are consistent with previous convection diurnal cycle analysis (Alcala and Dessler, 2002; Liu and Zipser, 2005; Nesbitt and Zipser, 2003). In contrast, the OT diurnal cycles over land are much stronger than over ocean. Strong regional variations are also discovered over land areas. Relatively strong OT diurnal cycles are found during JJA over southwestern North America, southeastern United States, Tibetan Plateau, and tropical South America and during DJF over southeast Australia, tropical and subtropical South America, and subtropical Africa. Relatively weak diurnal cycles over land are observed in central North America and West Africa in JJA. Strong regional variations in the OT diurnal cycle over land are consistent with previous studies based on convection and precipitation that demonstrate that the diurnal cycles are complicatedly modulated by the land–sea contrast, topography, coastline curvature, and response to solar heating to surface (Janiga and Thorncroft, 2014; Tian et al., 2005).
The global distributions of OTs at four observation times. Grids
with the OT number
Panels
Panels
Summary of the top 10 coldest OTs from Terra and Aqua, respectively.
From the diurnal cycle analysis in Sect. 4.3, we have noticed some land–sea
contrast in OT characteristics. For instance, OTs occur more frequently in
the afternoon over land, whereas they are more frequent at midnight over
ocean, and the OT occurrence diurnal cycle is stronger over land than over
ocean. In this section, attention is given to the OT center
Particularly, the first 10 coldest OTs (marked with red triangles in Fig. 11 and summarized in Table 2) from Aqua and Terra MODIS nearly occur in the Southern Hemisphere, with more cases over land than over ocean. The top 10 coldest OTs from Aqua are colder than 167 K, with the coldest OT of 165.6 K over the east of Papua New Guinea, whereas Terra shows the coldest OT of 167.2 K occurring in northern Australia. This finding agrees with the cold OT distributions discussed in Proud and Bachmeier (2021), who state that an extremely cold tropopause coupled to an energetic overshooting top produced such a cloud-top temperature.
Additionally, Fig. 11a and b reveal colder OTs over land than over ocean at the same latitudes. By checking the probability density distributions (PDFs) of the OT center
Our findings indicate that OTs over land are more intense than over ocean, except for the early morning (
To utilize about 2 decades' worth of MODIS records in the study of convective overshooting tops, we developed an IR algorithm to detect OTs from MODIS. The resultant OT climatology was used to understand OT regional and seasonal distributions, OT diurnal cycles, and OT land–sea contrast.
The approach to detect OTs uses IR radiances from MODIS water vapor (6.7
The global and seasonal distributions of OT occurrences follow the expected pattern, based on the known climatology of deep convection and precipitation, which shifts with the ITCZ and monsoonal systems. Frequent OTs are also observed over central North American, Europe, northern Asia, and the northwestern Atlantic Ocean in summer. Our OT climatology also includes those OTs observed in the shallow convection over the midlatitude ocean during spring–winter cold-air outbreaks.
MODIS observations at four different times were used to derive part of the
OT diurnal cycle. The diurnal cycle follows the known diurnal cycle of convection. The most OT occurrences are observed at about 13:30 over most land areas, including tropical and subtropical South America, tropical and subtropical Africa, southeastern North America, the foot of Himalayas, and the Maritime Continent. Over ocean, maximum OT occurrences are usually at around midnight (
Jeyaratnam et al. (2021) indicated that tropical convection is deeper than midlatitude convection. This is also revealed by the midlatitude tropic contrast in OT center
This study has displayed a comprehensive analysis of OT occurrences over the globe for the first time, using MODIS data. As MODIS has a fine spatial resolution (1 km) and provides about 2 decades of stable climate records, results in this study are an important complement to the current OT climatology in the literature derived from GPM, the Geostationary Operational Environmental Satellite, and AMSU-B (Bedka et al., 2018; Hong et al., 2008; Liu et al., 2020). This study also lays a foundation for understanding near-global climatological distributions of hazardous thunderstorms, leading to valuable insights into intense updraft size distributions in deep convection over the globe.
CloudSat data, including 2B-GEOPROF, 2B-CLDLASS-LIDAR, and ECMWF-AUX, were downloaded from
The supplement related to this article is available online at:
YH, JT, SN, and LDL conceived this study. YH performed the analysis, collected data, and wrote the paper. SN collected the data, helped with the data analysis, and edited the paper. JT helped with interpretation of the results and edited the paper. LD joined in with result discussions and edited the paper.
The contact author has declared that none of the authors has any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This work has mainly been supported by NASA (grant no. 80NSSC20K0902). The authors would like to acknowledge Guangyu Zhao for his help in downloading the Terra MODIS data. We thank the CloudSat Data Processing Center for providing CloudSat products, including 2B-GEOPROF, 2B-CLDCLASS-LIDAR, and ECMWF-AUX. We thank the Level-1 and Atmosphere Archive and Distribution System (LAADS) Distributed Active Archive Center (DAAC) for offering MODIS data (LAADS DAAC, 2022). We also acknowledge NASA's Goddard Earth Sciences Data and Information Services Center (GES DISC) for archiving MERRA-2 data and GPM data (Iguchi and Meneghini, 2021).
This research has been supported by the National Aeronautics and Space Administration (grant no. 80NSSC20K0902).
This paper was edited by Pawan K. Bhartia and reviewed by two anonymous referees.