Journal of Geophysical Research: Atmospheres

Diurnal variations in convective storm activity over contiguous North China during the warm season based on radar mosaic climatology

Authors


Corresponding author: M. Chen, Institute of Urban Meteorology, China Meteorological Administration, 55 Beiwaxili St., Haidian, Beijing 100089, China. (mxchen@ium.cn)

Abstract

[1] Diurnal variations in the formation and development of convective storms over contiguous North China during the warm season were studied using reflectivity from six China Next Generation Weather Radars between 2008 and 2011. Our results, including temporal and spatial analysis of hourly storm frequency through the warm season, and inter-month comparisons during June, July and August, indicate that most storms initiate over the northwestern mountains in the afternoon as a result of solar heating, with the highest frequency in June, and the lowest in August. Storms propagate from the mountains to the southeastern foothills and plains, with the highest rates occurring in June, and the lowest in August. In the late afternoon, there is a remarkably high storm frequency over the foothills and plains, which indicates a significant topographic control on the southeastward propagation and intensification of storms during the warm season. Storm activity occurs mainly on the plains through the night, with the highest frequency in July and the lowest in June, as a result of a favorable nocturnal convection mechanism. The region-averaged hourly storm frequencies for the warm season, and also for each month in JJA, are all bimodally distributed, with peak frequencies occurring in late afternoon and during the night, with the highest frequency recorded in the late afternoon during June and July, but at night in August. In general, the mean storm frequency is highest during the day and night in July, and lowest from afternoon to evening in August, but from nighttime to the next morning in June.

1. Introduction

[2] Both weather forecasting and climate prediction are heavily dependent on the adequate representation of convective precipitation and storm processes, including the diurnal cycle of convective precipitation and storm life. However, this diurnal cycle is poorly represented in today's weather and climate models. Analysis of diurnal variations in warm-season precipitation and storms is of considerable importance for several reasons, namely: (1) to develop a better understanding of the mechanism of both warm-season rain formation and local convective storm climatology [e.g.,Carbone et al., 2002; Parker and Knievel, 2005; Carbone and Tuttle, 2008; Chen et al., 2009; Overeem et al., 2009; P.-F. Lin et al., 2011]; (2) to evaluate and improve model parameterization schemes and physics, and serve as a benchmark for the ongoing development of numerical models [e.g., Dai et al., 1999; Lin et al., 2000; Davis et al., 2003; Zhang, 2003; Liang et al., 2004]; and (3) to improve the accuracy of forecasts of the timing and location of convective precipitation and storms [e.g., Shafer and Fuelberg, 2006; Saxen et al., 2008].

[3] Over the past several decades, climatological studies of diurnal variations in global and regional warm-season precipitation and storms have been based on observations from surface stations [e.g.,Wallace, 1975; Steiner et al., 1995], weather radar [e.g., Kuo and Orville, 1973; Steiner et al., 1995; Carbone et al., 2002; Hsu et al., 2006; Carbone and Tuttle, 2008; Weckwerth et al., 2011], satellite imagery [e.g., Levizzani et al., 2006; Laing et al., 2008], cloud-to-ground lightning [e.g.,Reap and MacGorman, 1989; Watson et al., 1994; Livingston et al., 1996; Murphy and Konrad, 2005], and also on numerical modeling [e.g., Davis et al., 2003; Koo and Hong, 2010; Surcel et al., 2010]. Most of these studies have shown that the spatial and temporal patterns of diurnal variations in convective precipitation or storms vary markedly from place to place. Some of these studies have also indicated that there are climatological differences in the diurnal variations of convective storms and precipitation in midlatitude areas from month to month in the summer or warm season [see Carbone et al., 2002; Wang et al., 2004; Hsu et al., 2006; Levizzani et al., 2006; Yin et al., 2009].

[4] Several regions of China form an important part of the global convective precipitation and storm activity system. In recent years, climatological studies of diurnal variations in warm-season precipitation over China have focused on rainfall over the east, center, or south of large regions of China, and used mainly surface and satellite observations [e.g.,Wang et al., 2004; Zhai et al., 2005; Yu et al., 2007a, 2007b; Chen et al., 2009; Yin et al., 2009; Zhang and Zhai, 2011; Chen et al., 2012]. However, these studies did not include contiguous North China, which covers mainly the Beijing area, the Tianjin area, and Hebei province, and is a very important warm-season rainfall region, with a complex topography [Zhai et al., 2005; He and Zhang, 2010]. In addition, these studies did not use continuous high-resolution radar data to investigate detailed diurnal variations in convective storms that are a well-known primary source of heavy rain over this region during the warm season.

[5] Continuous weather radar observations over the region have been made since 2008. The radar data set has a high spatial and temporal resolution, and there is good coverage from China's operational CINRADs (China Next Generation Weather Radar) in the region. In this study, we use this radar data set to complete a climatological analysis of diurnal variations in convective storms over the region, and consider both the warm-season mean patterns and monthly differences during June, July, and August (JJA).

[6] Data collection and our analytical methods are described in the following section. The characteristics of the mean warm-season diurnal variations in convective storm frequency are presented insection 3, which includes hourly spatial distributions of diurnal storm frequency, hourly region-averaged storm frequencies, and a consideration of diurnal storm evolution and propagation and the effects of terrain. Inter-month differences in these diurnal variations in convective storm activity during JJA are discussed insection 4. The final section contains the closing discussion and summary.

2. Data Sources and Analytical Methods

[7] Carbone et al. [2002], Carbone and Tuttle [2008], and P.-F. Lin et al. [2011]show that radar climatology can be used to identify significant variations in the characteristics of warm-season convective storms and precipitation using four years (or more) of weather radar data. In this paper, we analyzed composite reflectivity mosaic (CRM) data based on observations from six operational CINRAD radars comprising four S-band radars (BJRS near Beijing City, TJRS near Tianjin City, SJZRS near Shijiazhuang City, and QHDRS near Qinhuangdao City) and two C-band radars (ZBRC near Zhangbei County, and CDRC near Chengde City) in contiguous North China between 2008 and 2011 (Figure 1). All radars have been controlled for synchronization scanning under the radar Volume Coverage Pattern 21 (VCP21) mode, with an interval of approximately six minutes, using a remote radar command system at Beijing Meteorological Service since May 2008.

Figure 1.

Surface elevation (m) around the study area in contiguous North China showing the radar climatology analysis domain of 600 × 600 km (solid square), the subdomain shown in Figure 4c (the solid diagonal rectangle and line A–B), and the subdomain used to calculate the mean storm percentage over the Beijing area (the small dashed square). The six radar sites of the operational CINRAD network are indicated (plus symbols), as is the urban area of Beijing close to the BJRS radar site. The blue thick lines indicate the coastlines.

[8] The CRM data set was produced as follows. First, reflectivity data from individual radar polar coordinates were interpolated to produce 3D reflectivity with a Cartesian resolution of 1 km. To reduce clutter contamination before the interpolation was performed, quality control of radar reflectivity observations was implemented to remove abnormal propagation, ground clutter, and particle clutter, using fuzzy logic and clutter map methods [Chen et al., 2010]. The complex mountainous terrain, which reaches elevations of over 2400 m around the two C-band radars (ZBRC and CDRC,Figure 1), caused beam blockage and ground clutter, and it was difficult to completely eliminate them without affecting the true storm echoes. Therefore, some ground clutters from ZBRC were still retained in the northwest of the domain to maintain the integrity of the storm echoes (cf. Figure 2). Second, the quality-controlled 3D Cartesian reflectivity data from the individual radar were combined to form 3D reflectivity mosaic grids [Chen et al., 2010]. Consequently, the mosaic data also have a spatial resolution of 1 km on Cartesian coordinates, and an updating interval of approximately six-minute. Third, bright bands that frequently cause artificially high reflectivity near the freezing layer [Fabry and Zawadzki, 1995] were removed from the 3D reflectivity mosaic using an automatic identification algorithm [Chen and Gao, 2006]. Finally, the 2D CRM data were generated from the 3D reflectivity mosaic grids using the maximum reflectivity in the column of each 1 × 1 km grid in a square analysis domain of 600 × 600 km that was centered on the BJRS radar site (Figure 1).

Figure 2.

Spatial distributions of the warm-season mean number of observations of reflectivity ≥40 dBZ for each hour of the day. The thick solid lines indicate an elevation of 200 m.

[9] In this study, the warm season is defined as 15 May to 15 September. Based on preliminary statistics from radar observations, and the operational experience of local forecasters, 98% of convective storms that occur over contiguous North China do so during this period of every year, and 95% of these storms occur in JJA, the main warm-season months. The radar climatology analysis was based on CRM data from the four warm seasons between 2008 and 2011. The monthly CRM data from JJA in the four warm seasons was also analyzed. The number of echo occurrences in each hour of each day with reflectivity observations indicating the presence of convective activity was recorded for each 1 × 1 km grid square in the 600 × 600 km domain. These data were then averaged in the each 1 × 1 km grid square over both the whole warm season, and for each month in JJA of the four years. These average hourly occurrence frequencies of reflectivity above the set threshold were then expressed either as the number of observations above the relevant threshold, or as a percentage of the total number of observations for the each 1 × 1 km grid square, which is similar to the approach ofKuo and Orville [1973]. Previous studies have taken 40 dBZ as a reasonable reflectivity threshold criterion for convective activity [e.g., Reap and MacGorman, 1989; Livingston et al., 1996; P.-F. Lin et al., 2011], and used this value to discriminate between convective storms and stratiform rain [e.g., Falconer, 1984; Rickenbach and Rutledge, 1998; Parker and Knievel, 2005]. Most of organized convection often generates widespread areas of quasi stratiform rain, typically below 40 dBZ, that is mostly of convective origin late in the organized convective lifecycle. Here, the average number of hourly observations ≥40 dBZ is used to follow diurnal variations in the occurrence frequency of convective storms in the warm season, and every month of JJA.

[10] Unless otherwise stated, the times quoted in this paper refer to universal time coordinates (UTC), which is eight hours behind local standard time (LST).

[11] In addition, calling a 4–year period as “climatology,” we need to warrant for atmospheric circulation of the four warm seasons was not unusual. Based on the motive, the mean 500 hPa geopotential heights of the four warm seasons (JJA of 2008–2011) were compared to counterparts of the long-term climatological average using the NCEP/NCAR Reanalysis Monthly Means Data set [Kalnay et al., 1996]. We used the two long-term means of 500 hPa geopotential height in JJA of 1968–1996 and 1981–2010 provided by the reanalysis data set to regard as the benchmark of climatological average, and found atmospheric circulation over northeast hemisphere in the four warm seasons were not distinct atypical (figures omitted).

3. Diurnal Variations During the Warm Season

3.1. Hourly Spatial Distributions

[12] The spatial distribution of the average number of hourly observations of reflectivity ≥40 dBZ was analyzed to determine the pattern of diurnal storm initiation and intensification over contiguous North China in the warm season (Figure 2). The thick solid line in Figure 2 indicates an elevation of 200 m, and marks the boundary between the mountains and the foothills or plains. The extremely large, and scattered, number of observations in the northwest of the region resulted from the ground clutter associated with ZBRC, as explained above.

[13] From around 04:00 (12:00 LST, local noon), storm frequency clearly increases over the mountainous area in the northwest of contiguous North China (part of the Yanshan Mountains), especially between 06:00 and 10:00 (local afternoon), and this results from thermal forcing in the mountainous terrain caused by solar heating. The mountainous area acts as a heat source, and is an important storm genesis zone in the local afternoon as has been reported elsewhere [Barker Schaaf et al., 1988; Carbone et al., 2002; P.-F. Lin et al., 2011; Weckwerth et al., 2011]. Thereafter, storm frequency decreases rapidly over the mountains from 11:00. During other periods, most storms are initiated and develop over the foothills or the southeast plains, with a distinct increase in storm frequency from 08:00. Between 09:00 and 11:00 (local late afternoon), Figure 2 shows that a direct relationship exists between storm enhancement activity and topography over the analysis region. In particular, the number of reflectivity observations ≥40 dBZ increases to over 60, and reaches a maximum of almost 100, in front of the mountains of the Beijing area between 10:00 and 11:00; these reflectivity observations correspond to storm probabilities of approximately 2.5% and 3.6%, respectively. These results indicate that not only do most storms strengthen as they move to the southeast (and downhill), but also that storm initiation is frequent along the foothills, especially in the Beijing area and its vicinity. Previous studies [Wilson et al., 2007; Chen et al., 2010; Wilson et al., 2010] show that storm intensification and initiation over the foothills nearly correlated with the favorable situation of a southerly airflow rising over the side hills, and air mass instability over the plains, which occurs frequently during the warm season. Between 11:00 and 13:00 (local evening), the radar climatology indicates that storm cells move continuously southeastward, and have a high probability of evolving into a long-lived storm system (e.g., a squall line) after they reach the southeast plains near Beijing and Tianjin. From 13:00 to 22:00 (local nighttime), most nocturnal convective storms occur over the plains, and storm frequency decreases.Zheng et al. [2007] use satellite data to describe the similar characteristics of nocturnal convection. The study given by He and Zhang [2010]using satellite and model data show that the diurnal convective precipitation cycle over the North China region is related to the diurnal variation of a regional scale, mountain–plains solenoid (MPS) circulation induced by differential heating between the mountains and the plains over the region. There are at least two possible mechanisms responsible for the nocturnal convective precipitation over the plains, related to the MPS circulation: (1) the initiation or enhancement of convection over the mountains and slopes of the mountains in the afternoon that subsequently propagate southeastward to the plains, following the mid-tropospheric mean flow, and (2) the local initiation or enhancement of convection following the upward branch of a MPS circulation induced by differential heating between the mountains and plains, which is further facilitated by the nocturnal low level jet (LLJ) over the plains that contributes to a transport of southerly warm and moist air to the area during nighttime. The large number of isolated storm cells near the Beijing urban area between 13:00 and 14:00 may be related to the urban heat island effect and urban land surface heterogeneity [Guo et al., 2006; C.-Y. Lin et al., 2011; Miao et al., 2011; Niyogi et al., 2011], and requires further investigation. The majority of storms occur over the southeast plains, with a much lower frequency through the morning (00:00–04:00) than at night (13:00–22:00). In general, the lowest frequency of convective storms over the region occurred between 02:00 and 04:00 (near local morning to noon) in the warm season, and this pattern can also be seen in the following analysis.

3.2. Region-Averaged Hourly Storm Frequency

[14] Figure 3ashows the mean warm-season percentage of observations for each hour of the day having thresholds of reflectivity ≥40 dBZ in the each 1 × 1 km grid square, averaged over the analysis domain of 600 × 600 km. Here, the time of the region-averaged hourly percentage is labeled as the end-point of each hour on the horizontal axis ofFigure 3a (and also in Figure 3b and Figures 10a and 10b). The region-averaged storm frequency is much less between 02:00 and 03:00, and 03:00 and 04:00, than at other times, and increases distinctly from about 07:00, which can also be seen inFigure 2. There are two comparable peaks in the distribution of these mean percentages (Figure 3a), with the larger late afternoon peak (09:00–10:00 and 10:00–11:00) driven by the close relationship between storm enhancement activity and topography, and the other at midnight (around 18:00) related to the nocturnal convection over the plains. The high nighttime peak indicates that the southeast plains are much more sensitive to the area of reflectivity ≥40 dBZ during the night than the day. As described later, long-lived and organized convective systems such as MCSs have much higher occurrence frequency over the southeast plains, especially in July. These organized convective storms can generate huge areas of widespread precipitation late in their lifecycles due to evolution from erect convection toward slantwise ascent while they propagate downstream. This is capable of radar reflectivity over 40 dBZ in downstream favored locations and results in the high nighttime storm frequency peak.

Figure 3.

Histogram of the warm-season mean percentage of hourly observations with a reflectivity ≥40 dBZ averaged over (a) the analysis domain for contiguous North China and (b) the dashed rectangle domain marked inFigure 1 for the Beijing area. The number of rain gauge stations in each hour with a 24 h rainfall total over 50 mm is shown by the line in Figure 3b. The data were collected from automatic and traditional surface rain gauges in the Beijing area from May to September between 2008 and 2011. The horizontal axis is the UTC time.

[15] Focusing on the Beijing area, Figure 3b shows the mean percentage in the each 1 × 1 km grid square averaged over the dashed rectangle marked in Figure 1. The trend in the storm frequency percentage over the contiguous Beijing area is similar to its counterpart over contiguous North China, including a much lower percentage between 02:00 and 04:00 than at other times, and two peaks, with the larger at about 10:00, and the other between 18:00 and 20:00. The line in Figure 3bshows the number of surface stations each hour where 24 h rainfall amounts exceeded 50 mm. The rain gauge data were collected from hundreds of automatic and traditional surface stations in the Beijing area from May to September between 2008 and 2011. The diurnal trend in surface rainfall is similar to the trend in the region-averaged storm percentage, and also has an analogous dual-peak, which indicates that heavy rainfall (>50 mm in 24 h) results mainly from convective storms over the Beijing area in the warm season.

[16] The diurnal trend and bimodal distribution of the region-averaged hourly storm frequency differs from the results obtained using conventional surface precipitation data over contiguous China byYu et al. [2007a, Figures 1, 2d, and 2e]. The first primary reason for this discrepancy is that contiguous North China was partitioned and incorporated into two other large regions (contiguous Northeast China and East China) by Yu et al. [2007a]. The second primary reason for this discrepancy is that the climatology of storm frequency based on radar observations does not correspond exactly to that derived from surface observations of rainfall [Carbone and Tuttle, 2008]. Based on Yu et al. [2007b, Figure 1], the occurrence frequency of warm-season precipitation events of 1–3 h duration was greater than 60%, but total surface rainfall were less than 30% over contiguous China and were mainly derived from short-lived convective storms. However, the occurrence frequency of precipitation events lasting more than 6 h was less than 15%, but total surface rainfall of more than 50% comprised rain from both long-lived convective storms (e.g., a squall line or supercell) and mixed stratiform and convective precipitation systems. The results further explain that short-lived convective storms over the region have higher occurrence frequency but less surface rainfall; whereas, long-lived convective systems have lower occurrence frequency but more surface rainfall.

3.3. Diurnal Propagation and the Effect of Topography

[17] The time-distance plot (i.e., Hovmöller diagram) can be used to further highlight the diurnal evolution and movement of convective storms, as well as the effect of topography on the storms [e.g.,Carbone et al., 2002; Wang et al., 2004; Hsu et al., 2006; Yu et al., 2007a; Carbone and Tuttle, 2008; He and Zhang, 2010; P.-F. Lin et al., 2011]. Hovmöller diagrams of the average number of hourly observations of reflectivity ≥40 dBZ in the warm season are shown in Figure 4, as is the averaged topographic profile in the corresponding direction.

Figure 4.

Hovmöller diagram showing the warm-season mean of the number of hourly observations with a reflectivity ≥40 dBZ for (a) latitudinal average and (b) longitudinal average over the analysis domain, and (c) approximate SW–NE orientation average along line A–B inFigure 1. The average topographic profile is also shown below each panel. The vertical axis is UTC time for the Hovmöller diagrams, and elevation (m) for the topographic profiles (denoted as TopoHgt). In Figure 4c the horizontal axis is approximate distance (km) between A and B.

[18] Figures 4a and 4b contain the Hovmöller diagrams of the latitudinal average and the longitudinal average, respectively, in the analysis domain of 600 × 600 km, which reveal that storms initiate over the mountainous area from about 04:00 (local noon), and then tend to move to the east and south. Between 10:00 and 11:00 (local late afternoon), the two Hovmöller diagrams both indicate that storm frequency increases in the region 116°E–117°E, 39.5°N–40.5°N. It can be seen that this region is close to the foothills of the Beijing area where storm frequency tends to be very high during this period. The next morning, the storms dissipate rapidly while moving toward the Bohai Sea.

[19] The dominant direction of propagation of convective storms is approximately northwest to southeast (NW–SE) through the diagonal rectangular subdomain (see Figure 1), and this can be inferred from Figure 2, and tracked using a sequential plot of the hourly distribution of the number of storms observed. Consequently, the average number of hourly observations of reflectivity ≥40 dBZ was averaged in the subdomain for the Hovmöller diagram (Figure 4c). The diagram was aligned along the long dimension of the rectangle marked by A and B that expresses an approximate NW–SE orientation, and averaged along the shorter (cross) dimension of this rectangle that runs approximately southwest to northeast (SW–NE). The diagram shows that the dominant propagation direction of convective storms is close to NW–SE, and indicates that storm evolution characteristics are similar to those revealed in both the latitudinal average and longitudinal average diagrams (Figures 4a and 4b), other than the distinct increase in storm frequency in the foothills near Beijing between 08:00 and 13:00. Our findings regarding the main direction of storm movement are similar to those of He and Zhang [2010, Figure 4]. However, obvious discrepancies exist in other aspects due to differences in the data, method, and domain used.

[20] Figures 4a–4csupport the direct relationship between storm frequency and topography: in general, the steeper the average terrain profile, the higher storm occurrence frequency over the foothills. Steeper topography favors the uplift of the low-level southerly warm and moist air over the foothills and plains, which induces storm intensification on moving from the mountains to the plains, or storm initiation along the terrain slopes and foothills [Wilson et al., 2007; Chen et al., 2010; Wilson et al., 2010].

[21] Note that some of the high observation numbers in the Hovmöller diagrams result from ground clutter around ZBRC over the mountains as outlined above and shown in Figure 2.

4. Inter-monthly Patterns of Storm Activity in JJA

4.1. Spatial Variations in Storm Frequency

[22] Analysis of the monthly averages of the number of hourly observations of reflectivity ≥40 dBZ over the region revealed variations in the spatial distribution of storm frequency across JJA. For brevity, results are only shown for several representative periods, and the primary differences can be summarized using the four periods described below.

[23] 1. Local early afternoon (04:00–07:00). A higher frequency of scattered storm cells occurs over the northwest mountains in June and July than in August due to solar heating. A few storms occur over the southeast plains in JJA. Figure 5 shows representative spatial distributions for 05:00–06:00 in each month of JJA.

Figure 5.

As for Figure 2, but for 05:00–06:00 in (a) June, (b) July, and (c) August.

[24] 2. Local afternoon (07:00–10:00). Storm frequency increases over the northwest mountains in June and July due to solar heating and these storms then propagate southeastward. Figure 6 shows that storm frequency between 09:00 and 10:00 over the northwest mountains is little higher in June than in July and much higher than in August, which indicates that solar heating is the most important factor in storm initiation in June. However, comparing Figure 6 with Figure 5, it is evident that the area where storm frequency increases the most is the foothills and the neighboring plains between 09:00 and 10:00 in July, which indicates a high terrain forcing effect on storm evolution. There is a much lower storm frequency over both the mountains and the plains in August than in June and July.

Figure 6.

As for Figure 2, but for 09:00–10:00 in (a) June, (b) July, and (c) August.

[25] 3. Local late afternoon to midnight (10:00–16:00).Storm frequency decreases rapidly over the northwest mountains, and is close to zero after 14:00 in JJA. In contrast, storm frequency increases noticeably over the foothills and the southeast plains, especially in July, and slightly less so in June. This suggests significant storm intensification and initiation in the vicinity of the foothills as storms propagate southeastward due to the impact of topographic forcing (a similar effect was observed in the warm-season mean), especially in the Beijing area between 10:00 and 11:00 (cf.Figure 7 and Figure 2). In addition, Figure 8 (12:00–13:00) suggests that the storms move most rapidly in June, and this is also evident from a comparison of the Hovmöller diagrams (see Figures 1113). Otherwise, we should bear in mind that successive initiation of stationary convective storms in different locations could also produce the similar pattern, but it is rare.

Figure 7.

As for Figure 2, but for 10:00–11:00 in (a) June, (b) July, and (c) August.

Figure 8.

As for Figure 2, but for 12:00–13:00 in (a) June, (b) July, and (c) August.

[26] 4. Midnight to the next morning (16:00–04:00). In general, storm frequency is close to zero over the northwest mountains, and also gradually diminishes over the southeast plains from midnight to the next morning in JJA. The lowest and highest frequency of nocturnal convective storms over the plains is in June and July, respectively. The spatial distribution of nocturnal storms over the plains differs in each month of JJA. Most nocturnal storms occur over the Beijing and Tianjin areas of the southeast plains in July, but over the southeast of the Beijing area and south of Hebei province in August, especially after midnight between 16:00 and 22:00 (Figure 9, 19:00–20:00). Storm frequency is much lower over the southeast plains in the morning than at night, which is similar to the pattern associated with the warm-season mean.

Figure 9.

As for Figure 2, but for 19:00–20:00 in (a) June, (b) July, and (c) August.

[27] Here, we need to illuminate that other factors, such as seasonal changes in the mean large-scale flow pattern could be also important to difference of storm frequencies in June, July and August, which is beyond the scope of this study.

[28] Based on the spatial distribution and variability of storm frequency, storm structure can be inferred from the radar climatology. Most of the storms over the northwest mountains can be inferred as single cells or multicell clusters in JJA. However, storms that occur over the southeast plains can be inferred as most of convective types (e.g., single cells, multicell clusters, multicell lines, or supercells). The occurrence frequency of long-lived storms (e.g., supercells, squall lines, and MCSs) over the southeast plains can also be deduced much higher in June and July than in August (Figure 8). The mechanism for forming and supporting long-lived episodes of propagating convective storms over the southeast plains may be similar to that over the central Great Plains of the United States [Trier et al., 2006, 2010], and this requires further research using numerical cloud-scale modeling.

4.2. Variations in Region-Averaged Diurnal Storm Frequency

[29] Using the same approach as for the warm-season mean, diurnal variations in the regional average of hourly storm frequency for every month of JJA were calculated; i.e., the percentage of hourly observations having thresholds of reflectivity ≥40 dBZ in the each 1 × 1 km grid square averaged over the region (Figure 10a). These data are also shown as a month-hour plot to highlight the frequency differences (Figure 10b). Figure 10reveals that the diurnal trends in region-averaged hourly storm frequency for June, July, and August are distinctly different, although there is some resemblance among the general trends. The primary differences are as follows.

Figure 10.

(a) Percentage of region-averaged hourly observations with a reflectivity ≥40 dBZ for June, July, and August (JJA). (b) Month-hour distribution of the same data. The horizontal axis is UTC time.

[30] 1. The lowest storm frequency was between 01:00 and 03:00 in June and July, but between 03:00 and 04:00 in August.

[31] 2. In July, storm frequency was higher throughout the day than in the other two months, with an extremely high frequency between 10:00 and 12:00.

[32] 3. Two peaks in the region-averaged storm frequency are seen each month during June, July, and August, with the first at approximately 10:00–12:00, 10:00–12:00, and 08:00–10:00, respectively, and the second around 18:00–19:00, 22:00, and 18:00–20:00, respectively.

[33] 4. A much higher peak in storm frequency is seen during the day in June and July, but the larger peak was recorded during the night in August. The seasonal shift from an afternoon maximum in early summer (June, July) to nighttime maximum in late summer (August) is very striking.

[34] We suggest that the bimodal monthly distribution of storm frequency during JJA is the dominant influence on the pattern of storm activity observed over the whole warm season. The high daytime peak is related to the significant topographic impact on storm intensification and initiation as storms propagate from the northwest mountains to the southeast plains in June and July. The high nighttime peak indicates that convective activity is much more frequent over the southeast plains after dark than during the day in August, which may be connected with the above favorable nocturnal convection mechanism and northward movement of the West Pacific Subtropical High (WPSH) in August. The WPSH can influence the location and distribution of heavy rain over the southeast plains in contiguous North China in August [Liu et al., 2007]. The interpretation of the nocturnal peak in June and July can be followed the similar interpretation of the nocturnal peak in the warm season, as mentioned above.

4.3. Differences in Diurnal Propagation

[35] The latitudinal, longitudinal, and SW–NE averaged Hovmöller diagrams of the average number of hourly observations of reflectivity ≥40 dBZ for June, July, and August are shown in Figures 1113. As in Figure 4, the averaged topographic profile in the corresponding direction is also included. These diagrams indicate that convective storms over the region have propagation and development tendencies to the east, south, and NW–SE, with the latter dominating in every month of JJA. In general, there are much stronger propagation signals in June and July, than in August.

Figure 11.

As for Figure 4, but for the June mean.

Figure 12.

As for Figure 4, but for the July mean.

Figure 13.

As for Figure 4, but for the August mean.

[36] Comparing the Hovmöller diagrams from each month, the plots in all three directions are flatter for June and July than August, which suggests that storm propagation is much faster in June and July (Figures 11 and 12). Note that the limit of completely flat in the Hovmöller diagrams, which indicates widespread simultaneous precipitation with no propagation as opposed to infinite propagation speed of a coherent precipitation feature. It can be seen that most storms reach the southeast of the region, close to Bohai Sea, at night (around 19:00–20:00) in June and July, but the following morning (around 23:00–00:00) in August. As mentioned above, long-lived and organized convective storms have a high occurrence frequency in June and July, and can propagate to the southeast plains near the Bohai Sea before 20:00. However, there is a much higher frequency of short-lived and isolated (unorganized) storm cells initiated over the foothills and the southeast plains after 20:00 in July than in June, which further induces correspondingly slower storm propagation in July than in June. Most storms in August are isolated storm cells that have local initiation and short-lived characteristics, and which generate not only an indistinct propagation trace in the latitudinal and longitudinal average Hovmöller diagrams, but also the lowest propagation speeds in the SW-NE average Hovmöller diagram.

[37] In addition, the relationship between a steeper average terrain profile and increasing storm frequency is illustrated by comparing the Hovmöller diagrams for every month of JJA, and this was also evident in the previous analysis of the warm-season mean.

5. Discussion and Summary

[38] Radar mosaic climatology was used to investigate diurnal variations in convective storm frequency over contiguous North China using six-minute reflectivity observations from six CINRAD radars between 2008 and 2011. Our results comprise hourly warm-season mean storm frequencies for the period between 15 May and 15 September, and an analysis of the variations in diurnal patterns of storm activity in June, July, and August, which are the main warm-season months in this region.

[39] Analysis of the warm-season means revealed that the mountainous area in the northwest of the region is an important source of convective storm activity, and has a high storm frequency in the afternoon, but a very low frequency at other times. Most warm-season storms propagate from the northwest mountains to the foothills, or further to the southeast plains, with the main propagation direction being approximately NW–SE. In the late afternoon, there is a remarkably high storm frequency over the foothills, especially in the Beijing area and its vicinity, where storms propagate from the mountains to the foothills. This higher storm frequency correlates well with the steeper nature of the terrain. The warm-season mean results imply a close relationship between storm activity and topographic forcing, and also that the steeper topography facilitates uplift of the low-level southerly warm and moist airflow over the plains, which in turn encourages storm intensification and initiation along the terrain slopes and foothills. From the evening to the early afternoon of the next day a great majority of warm-season storms occur only over the southeast plains, and there is a general trend of occurrence frequency reducing gradually with time. The lowest storm frequency over the region occurs in the morning. Hourly mean storm frequency averaged over the region shows two peaks: a larger one in the late afternoon related to terrain forcing, and the other at night, which is connected to the nocturnal convective storms trigger mechanism given byHe and Zhang [2010] that related to both the regional scale, mountain–plains solenoid (MPS) circulation and the nocturnal low level jet (LLJ) over the plains. A similar diurnal trend, and bimodal distribution, is seen in rainfall measurements from surface stations around Beijing area in the warm season.

[40] Diurnal variations in convective storm activity over the region in every month of JJA were also analyzed. In the local afternoon, the occurrence frequency of storms over the northwest mountains is highest in June, followed by July. During the same period, the occurrence frequency of storms over the southeast plains is highest in July. The most noticeable storm intensification and initiation over the foothills occur in July, followed by June, from local late afternoon to evening. The overall trend of storm frequency over the region falls gradually from the nighttime to the next morning in JJA, and this is similar to the overall warm-season trend. The occurrence frequency of nocturnal convective storms over the southeast plains is higher in July and August than in June. The hourly storm frequency averaged over the region indicates that there are two prominent peaks every month in JJA. In general, the mean storm frequency is the highest in July compared with June and August. It can be drawn the conclusion that the most frequent warm-season storm activity over the region is in July. In addition, the mean frequency is higher from afternoon to midnight in June than in August, and this is due to the higher frequency of long-lived convective storm systems over the foothills and the southeast plains in June than in August. In contrast, between midnight and the next morning, the mean frequency is in August higher than in June because much more isolated and short-lived storm cells initiate over the southeast plains during this period in August. Much stronger signals, and much faster rates of diurnal storm propagation, were recorded in June and July than in August, while the dominant propagation direction was NW–SE throughout JJA, and the maximum speed occurred in June and the slowest in August. We suggest that the strong propagation signal, and rapid propagation rate, result from the high frequency of long-lived convective storm systems in June and July, in contrast to the high frequency of isolated storm cells in August. However, we should bear in mind that assessing propagation using Hovmöller diagrams of the storm frequency a threshold is exceeded plotted versus time of day (as in a diurnal frequency diagram in the paper) has its limitation. In particular, if the diurnal phase of the convection varies from day to day over the time period for which the diurnal frequency diagram is constructed, we might not see a propagation signal when individual storms could indeed propagate.

[41] In this paper, diurnal variations of warm-season storms over contiguous North China were climatologically analyzed using only radar reflectivity mosaic data. The effects of mountainous terrain forcing on convective storm evolution were also preliminarily revealed in the radar climatology study. However, the direct and indirect impacts of dynamical and thermal-dynamical forcing from terrain elevation, urban heat islands, urban land surface heterogeneity, and sea breezes on the diurnal initiation and development of convective storms requires further investigation using fine observation diagnosis and numerical simulation, which is beyond the scope of this study. In addition, the favorable nocturnal convection mechanism on the southeast plains must be further clarified using other data and methods.

[42] Some radar climatology studies have shown that diurnal variations in convective storms differ depending on the changing influence of steering-level winds and synoptic-scale forcing during the summer or warm season [Kuo and Orville, 1973; Saxen et al., 2008; Yeung et al., 2011]. The differences over contiguous North China will be discussed in detail in other papers. Climate statistics and experienced local forecasters suggest that primary synoptic-scale systems such as the Mongolian cold vortex, eastward moving troughs, and the WPSH have an important influence on convective storm initiation and evolution over the region in the warm season. However, the degree of influence from every synoptic-scale system on convective storms varies in each month of the warm season.

[43] In the near future, radar-based climatology may prove to be a very powerful and important tool for improving local analysis and forecasting of convective storms and precipitation, which is one of weaknesses in the realm of today's weather forecast. Results from radar-based climatology may be integrated into real-time auto-forecasting systems to enhance operational predictions of convective storms and precipitation [e.g.,Saxen et al., 2008]. However, the limitations of radar data must be better mitigated; e.g., the effects of ground clutter as demonstrated here, as well as beam blocking, poor calibration, and range dependence [Parker and Knievel, 2005].

Acknowledgments

[44] We are sincerely grateful for comments from Richard Carbone and two anonymous reviewers. This work was supported by the National Natural Science Foundation of China under grant 41075036.