Volume 120, Issue 22 p. 11,519-11,535
Research Article
Free Access

Synoptic conditions related to soil moisture-atmosphere interactions and unorganized convection in Oklahoma

Trent W. Ford

Corresponding Author

Trent W. Ford

Department of Geography and Environmental Resources, Southern Illinois University, Carbondale, Illinois, USA

Correspondence to: T. W. Ford,

[email protected]

Search for more papers by this author
Steven M. Quiring

Steven M. Quiring

Department of Geography, Texas A&M University, College Station, Texas, USA

Search for more papers by this author
Oliver W. Frauenfeld

Oliver W. Frauenfeld

Department of Geography, Texas A&M University, College Station, Texas, USA

Search for more papers by this author
Anita D. Rapp

Anita D. Rapp

Department of Atmospheric Sciences, Texas A&M University, College Station, Texas, USA

Search for more papers by this author
First published: 26 October 2015
Citations: 20


Atmospheric modification by anomalously dry or wet soils can both enhance and suppress convective activity. However, the local-scale and mesoscale feedback governing soil moisture-precipitation coupling are embedded within the larger synoptic-scale environment. Despite their importance, synoptic-scale atmospheric conditions are rarely considered in studies examining soil moisture-atmosphere interactions. We combine self-organizing maps of 500 hPa geopotential height, spatial synoptic classification, and Hybrid Single-Particle Lagrangian Integrated Trajectory model air mass trajectories to determine if the synoptic-scale environment affects the ability of the land surface to force unorganized convection in Oklahoma. We identify several synoptic patterns that significantly impact the frequency of unorganized convection. Synoptic patterns characterized by midlevel troughs over the Southern Great Plains are less frequently associated with unorganized convective events. These patterns exhibit cool air advection in the midlevel and lower level of the atmosphere and are linked to suppression of convective activity. The synoptic patterns characterized by 500 hPa ridging over the study region are more frequently associated with unorganized convective events. These patterns likely result in increased net radiation, vapor pressure deficit, and more homogenously dry soils. Unorganized convective events that occur during these synoptic conditions initiate preferentially over dry soils. We present evidence that the synoptic-scale environment can influence whether and how the land surface has an impact on convection.

Key Points

  • Synoptic-scale conditions affect soil moisture feedback to convective precipitation in Oklahoma
  • Midlevel trough to the north diminishes the likelihood of unorganized convection
  • Midlevel ridging over Oklahoma is favorable for convective initiation over dry soils

1 Introduction

1.1 Soil Moisture-Precipitation Feedback

Anomalies of land surface moisture and energy have the potential to impact regional climate on daily to seasonal time scales [Dirmeyer et al., 2009; Meng and Quiring, 2010; Gentine et al., 2013]. One of the more fascinating and widely studied interactions is that between soil moisture and precipitation. In semiarid regions such as the United States Southern Great Plains, soil moisture has a significant influence on evapotranspiration rates [Basara and Crawford, 2002; Teuling et al., 2006] and latent and sensible heat fluxes [Guillod et al., 2014; Ford et al., 2014]. Modification of evapotranspiration and moisture flux to the atmosphere can potentially enhance or inhibit convective processes [Gentine et al., 2013; Ford et al., 2015]. This soil moisture-precipitation coupling has been demonstrated in both observational and modeling studies [Koster et al., 2004; Hohenegger et al., 2009; Santanello et al., 2009; Frye and Mote, 2010].

Soil moisture heterogeneity and corresponding surface heat flux gradients can also enhance instability and force mesoscale circulation features leading to precipitation [Weaver, 2004; Frye and Mote, 2010]. Taylor et al. [2011], for example, found a preference for convective cloud development on the dry side of wet-to-dry soil gradients in the Sahel region of Africa. Model and observational studies at a relatively fine spatial resolution have shown that precipitation can be triggered by wet and dry soils as well as by surface heterogeneity [Alfieri et al., 2008; Taylor et al., 2011; Santanello and Peters-Lidard, 2013]. Ford et al. [2015] demonstrate both wet soil and dry soil regimes, in which convection in the Southern Great Plains is initiated preferentially over dry or wet soils, primarily dependent on boundary layer humidity and the stability of the free troposphere. Meanwhile, global climate models show a consistent positive (wet) soil moisture–precipitation feedback [Koster et al., 2004], sometimes in direct contrast to observation-based analyses. For example, Taylor et al. [2012] found a strong atmospheric response to relatively dry soils and soil moisture heterogeneity, which lead to an increased probability of afternoon precipitation over dry soils. This is in contrast to the increased probability of precipitation over wet soils that is shown in general circulation models (GCMs). The spatial resolution of these GCMs has a major impact on their ability to accurately simulate these interactions [Taylor et al., 2013] because cloud cover and precipitation responses to surface flux anomalies are on the order of tens of kilometers [Weaver and Avissar, 2001; Allard and Carleton, 2010; Taylor et al., 2013].

1.2 Role of the Synoptic Environment

Despite the many studies that have examined soil moisture-precipitation feedback in this region, few have focused on the role of the synoptic-scale environment. Those that have considered forcings at this scale found synoptic-scale atmospheric conditions conducive to mesoscale convection driven by land surface heterogeneity [Dixon and Mote, 2003; Carleton et al., 2008; Allard and Carleton, 2010]. Here we focus on unorganized convective precipitation events, defined by Carleton et al. [2008] to be isolated convective events occurring in the absence of larger-scale systems. Warm season precipitation in the Southern Great Plains region is often generated by deep convection, driven in part by Gulf of Mexico moisture advected by the Great Plains Low Level Jet [Higgins et al., 1997; Frye and Mote, 2010]. Advancing low-pressure systems, squall lines, and the occasional tropical cyclone are other means of precipitation influencing the region [Raddatz and Hanesiak, 2008; Carleton et al., 2008; Knight and Davis, 2009]. Although preexisting soil moisture conditions can modify these large-scale systems [Kellner et al., 2012], it is difficult to observe. Typically, any soil moisture-precipitation coupling signal is dampened by the passing of these systems. Therefore, we focus here on locally initiated unorganized convective precipitation events to clearly connect soil moisture-precipitation interactions to the synoptic-scale environment.

Synoptic-scale atmospheric conditions can have a tremendous influence on the strength of soil moisture-atmosphere interactions in generating precipitation [Dixon and Mote, 2003; Carleton et al., 2008]. The warm season in the Southern Great Plains is characterized by a variety of synoptic-scale circulation patterns; some of which have been shown to be strongly associated with local-scale drivers of convection [Weaver, 2004]. Determining the atmospheric conditions most and least frequently associated with unorganized convection aids forecasting of such events and contributes to understanding the mechanisms that impact soil moisture-atmosphere interactions. The focus of this study is to identify the synoptic-scale atmospheric conditions associated with unorganized convection in Oklahoma and to relate these conditions to soil moisture–precipitation coupling. Our study has three objectives: (1) determine the synoptic patterns most and least frequency associated with unorganized convective events in Oklahoma, (2) identify atmospheric conditions during these synoptic patterns that may influence unorganized precipitation event frequency, and (3) document how soil moisture-precipitation coupling occurs during these synoptic patterns.

2 Data and Methods

2.1 Soil Moisture

Our analysis of synoptic-scale atmospheric patterns is separated into two categories: (1) convective precipitation that initiates over drier than normal soils and (2) convective precipitation that initiates over wetter than normal soils. This classification is based on daily soil moisture observations from the Oklahoma Mesonet [http://www.mesonet.org; Illston et al., 2008], a state-wide monitoring network composed of over 100 stations. For this study, we use in situ volumetric water content (cm3 cm−3) from 113 Oklahoma Mesonet stations. Volumetric water content of the soil is estimated using the thermal matric potential measured by Campbell 229-L heat dissipation sensors at 5, 25, and 60 cm. Physical soil properties were recently updated at each Oklahoma Mesonet site, and soil moisture observations from the 229-L sensors were evaluated with gravimetric samples [Scott et al., 2013]. Root-mean-square difference between direct measurements of volumetric water content and those reported by the Mesonet sensors varied from 0.08 to 0.05 (cm3 cm−3), with an overall network average of 0.05 [Scott et al., 2013]. Daily, midmorning (09:00 local standard time (LST)) soil moisture observations are taken for this study to best represent morning soil moisture conditions prior to convective initiation. We convert daily volumetric water content observations to percentiles, which are then gridded to a 0.25° resolution using the average of all stations within each grid cell. The gridded soil moisture data are used to determine whether convective precipitation initiated over dry or wet soils. Soil moisture at the point of precipitation initiation less than or equal to the 25th percentile of volumetric water content is considered dry, while values greater than or equal to the 75th percentile of volumetric water content are categorized as wet. Synoptic atmospheric conditions that are more or less frequently associated with convective precipitation over dry or wet soils, and the conditions forcing the associations, are easier to detect when separating events into those occurring over dry and wet soils.

In addition to soil moisture percentiles, we use the 09:00 LST volumetric water content measurements to characterize state-wide soil moisture patterns during days with and without convective events. Station-based measurements of daily maximum temperature, dew point temperature, relative humidity, vapor pressure deficit, and solar radiation are used to identify how the near-surface atmosphere is modified by soil moisture conditions.

2.2 Precipitation Events

We focus on unorganized convective precipitation events as defined by Carleton et al. [2008] to be those isolated convective events that occur in the absence of larger-scale systems. Because moisture and energy from larger-scale phenomena such as low-pressure systems and squall lines are remotely sourced, it is difficult to detect the impact of local soil moisture conditions. However, properly identifying unorganized convective precipitation events requires high spatial and temporal resolution data sets. We utilize the unorganized convective precipitation events identified by Ford et al. [2015]. Events in this data set were identified using the National Weather Service Stage IV precipitation product, based on the system of Next-Generation radar network. The Stage IV precipitation product is available at an hourly temporal and 4 km spatial resolution and captures nearly all of the contiguous United States.

Ford et al. [2015] identify 477 unorganized convective precipitation events initiating between 12:00 LST and 20:00 LST over Oklahoma from 2002 to 2012. This identification is manual and uses a classification scheme based on six decisions, including the location, size, intensity, shape, and propagation of the precipitating system. Unorganized convective events are identified as circular or semicircular systems initiating in Oklahoma and evolving and propagating independently of other precipitating systems. More detailed information about the event identification procedure, including the reproducibility and reliability of events identified, can be found in Ford et al. [2015]. It should be noted that the afternoon unorganized convective precipitation events identified by Ford et al. [2015] contribute a relatively small proportion of total warm season precipitation in the Southern Great Plains, likely due to the stringent event identification criteria. Nonetheless, between 2002 and 2012, unorganized convective events accounted for, on average, 5% of total warm season precipitation in Oklahoma and range from as little as 3% in 2008, to as much as 8% in 2010 and 2011.

2.3 Atmospheric Conditions

The North American Regional Reanalysis (NARR) climate data set [Mesinger et al., 2006] is used to characterize low-level and midlevel atmospheric conditions. NARR is a relatively high resolution (32 km) atmospheric and land surface data set that covers the entirety of North America. Meteorological and hydrological variables are available for 1979–2015 at 3-hourly and monthly resolutions. Because of the short time scales at which unorganized convective events operate, we use the 3 h data. NARR assimilates observations from several sources including radiosondes, aircraft, geostationary satellites, and model simulations.

We use NARR to characterize synoptic-scale atmospheric conditions associated with unorganized convective events. The variables analyzed include geopotential height anomalies, zonal and meridional wind velocity, integrated moisture flux, and total column precipitable water. These variables are composited for (1) all convective events, (2) convective events over dry soils (<25th percentile), and (3) over wet soils (>75th percentile). The composites associated with events are then compared with variable composites over all days (event and nonevent), in which a specific synoptic pattern occurs.

In addition, NARR is used to characterize atmospheric stability and convective energy for different synoptic patterns. Atmospheric soundings from Lamont, Oklahoma, are used to validate NARR's representation of convective available potential energy (CAPE), convective inhibition (CIN), and surface temperature at 06:00 and 12:00 LST, as well as the 12:00–06:00 LST change in each of three components (Table 1). Correlations between NARR and Lamont observations are consistently strongest for CIN and surface temperature and slightly lower for CAPE. Interestingly, NARR is able to better capture the variability of all three parameters at 06:00 and 12:00 LST but shows smaller agreement with the 12:00–06:00 LST change, arguably most important for demonstrating soil moisture impact on convective initiation [Ford et al., 2015]. However, in this study we only use NARR to characterize the overall environmental conditions prior to convective initiation, and therefore, the relatively good correspondence between NARR and the observations provides confidence in NARR's ability to estimate these parameters.

Table 1. Correlations Between NARR and Sounding Observations of Convective Available Potential Energy (CAPE), Convective Inhibition (CIN), and Surface Temperature at 06:00 and 12:00 LST, as Well as the 12:00–06:00 LST changea
06:00 LST 12:00 LST 12:00–06:00 LST
CAPE (J kg−2) 0.43 0.48 0.38
CIN (J kg−2) 0.65 0.71 0.50
Surface Temp (°C) 0.63 0.66 0.61
  • a Atmospheric sounding observations are taken from Lamont, Oklahoma.

2.4 Self-Organizing Maps

Classification of regional patterns of atmospheric conditions is common in synoptic climatology. Synoptic classification is useful for data reduction as well as evaluating the impact of synoptic-scale patterns on surface conditions [e.g., Kalkstein et al., 1996; Frauenfeld and Davis, 2002; Hanna et al., 2011; Dayan et al., 2012; Vanos et al., 2014]. Procedures range from manual classification of atmospheric pressure maps [Frakes and Yarnal, 1997] to neural networks and machine learning [Mihalakakou et al., 2002]. One methodology for synoptic classification that has been gaining popularity in climate science is self-organizing maps (SOMs), originally introduced by Kohonen [1990]. SOMs provide a useful method for visualizing nonlinear associations throughout a continuum of atmospheric conditions [Hewiston and Crane, 2002]. Sheridan and Lee [2011] provide an exhaustive review of the advantages and disadvantages of SOMs, and there are several examples of SOM applications in climate research [e.g., Cassano et al., 2015].

We implement SOMs to classify synoptic-scale patterns of 500 hPa geopotential height anomalies from NARR for all days between May and September 2002–2012 (1683 days in total). The study area classified extends from 25°N–50°N and 82°W–112°W. To test the sensitivity of the analysis to domain size, we also conducted the SOM analysis over a geographic regions that was 10° smaller in both latitudinal and longitudinal extent. No appreciable differences in the resulting SOM patterns arose from changes to the study region size (not shown). Investigation of midlevel (i.e., 500 hPa) atmospheric circulation provides valuable information for diagnosing and predicting weather conditions at the surface and in the low-level atmosphere [Michaelides et al., 2010; Sheridan and Lee, 2011]. The SOM procedure used here is similar to that described in Michaelides et al. [2010]. For our classification, input of daily geopotential height anomalies was randomized and the network was trained for 5000 iterations or “epochs.” By using geopotential height anomalies, the seasonal cycle between spring, summer, and autumn was removed. The learning-rate parameter of the SOM analysis was set at 0.5. This was chosen as a compromise between high values that result in a faster, but potentially unstable result, and low values that generate more stable results, but are time-consuming [Gutiérrez et al., 2005]. For each SOM process iteration, the neighbors of the selected node are modified through a distance decay function [Sheridan and Lee, 2011]. This is the primary difference between the SOM procedure and a k-means method of classification [Kohonen, 2001]. The neighborhood distance function used is the “linkdist” built-in MATLAB function with an initial neighborhood size of eigth grid cells. Despite the subjectivity of the learning-rate, distance function, and neighborhood size decisions, variations in these parameters result in minimal differences in the final classification [Cassano et al., 2006; Johnson et al., 2008].

The SOM procedure's output is a series of map patterns, the number of which is subjectively chosen based on the study needs. For example, a large number of patterns (36 or 48) will provide more detailed information about each set of atmospheric conditions; however, it will be much more difficult to interpret. Fewer patterns (4 or 8), conversely, will be more straightforward to interpret, but may overgeneralize diverse atmospheric conditions. We tested 8, 12, 16, and 20 synoptic patterns (not shown), and no visual differences were discernible, particularly between the 12 and 16 pattern solutions. Based on results of previous studies using fewer than 20 patterns [Crimmins, 2006; Coleman and Rogers, 2007; Guèye et al., 2011], and given our focus on just the warm season months, we concluded that a 12-pattern solution is ideal. The 12 synoptic patterns are related to the frequency of convective events to determine if the synoptic-scale atmospheric circulation has an influence on soil moisture-atmosphere interactions (Figure 1). We tested the sensitivity of the SOM classification to variable selection by repeating the analysis with 200 hPa geopotential heights anomalies and 2 m air temperature anomalies. The resulting SOM patterns of the two variables were very similar to the 500 hPa patterns (not shown), which gives us confidence in the robustness of our results.

Details are in the caption following the image
Mean 500 hPa geopotential height (m) composites associated with each of 12 self-organizing map synoptic patterns.

2.5 Spatial Synoptic Classification

Another method of synoptic classification is the Spatial Synoptic Classification (SSC), which classifies weather types based on surface observations at individual stations [Sheridan, 2002]. Temperature, dew point, wind, pressure, and cloud cover are used for the procedure, making the SSC a representation of surface conditions [Sheridan, 2002]. The SSC classifies surface variables into eight weather types, distinguished by the temperature and moisture content of the near-surface air. These types include dry polar (DP), dry moderate (DM), dry tropical (DT), moist polar (MP), moist moderate (MM), moist tropical (MT), and transitional (T). The transitional days are those in which large changes in pressure, dew point, and wind occur over the course of a single day. The final weather type is moist tropical plus (MT+), which represents a more intense moist tropical weather type. For the purposes of this study, all MT+ and MT days are grouped together.

Each calendar day is classified at individual surface observing stations. We use the daily SSC for Oklahoma City from 2002 to 2012 to represent the dominant weather type over the region. The nearest SSC station to Oklahoma City is Tulsa in northeast Oklahoma. All SSC days over the study period were compared between Tulsa and Oklahoma City, with generally good correspondence. The two cities had the same classification over 75% of the time for MT and DP weather types, over 70% of the time for DM and MM weather types, and over 60% of the time for DT, MP, and T weather types. Although not identical, the similarities between Oklahoma City and Tulsa, along with the typical geographic size of these weather types, suggest that the SSC for Oklahoma City is representative of conditions across the region.


The NOAA Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model [http://ready.arl.noaa.gov/HYSPLIT.php, Draxler and Rolph, 2003] was used to identify the source region of air masses associated with unorganized convective events. The HYSPLIT model provides three-dimensional movement of air parcels in time. HYSPLIT was run using data from NARR to maintain consistency with the SOM output.

Air mass back trajectories were run in two different setups (Figure 2). First, we used the model to develop a climatology of air mass back trajectories for every day during the study period. A back trajectory was determined from each location in a 50-point grid over the study area. The trajectories were initiated over each grid point at 850 hPa and 12:00 LST, and run 48 h back in time. This was repeated for each day during the study period. The culmination of these trajectories allow us to identify the typical air mass origin for each day. An example of one day's output from this HYSPLIT run is shown in Figure 2a. A second set of trajectories was calculated for the specific convective events. These trajectories were also initiated at 850 hPa at the location of the precipitation initiation, and run 48 h back in time from 12:00 LST. Figure 2b shows an example of a single precipitation event trajectory. Composites of atmospheric variables, SSC weather types, and air mass back trajectories were calculated for each synoptic pattern to test whether the synoptic-scale environment has an impact on unorganized convective precipitation in Oklahoma.

Details are in the caption following the image
Examples of HYSPLIT air mass back trajectories run (a) as a 1° × 1° matrix over the study region, and (b) from an individual convective precipitation event. All trajectories are initiated at 12:00 LST from 850 hPa and run 48 h back in time.

3 Results

3.1 SOM Patterns

Daily composites of morning (0:600–12:00 LST) geopotential height anomalies over the Southern Great Plains were used to define the synoptic patterns. The result is 12 patterns of 500 hPa geopotential height maps (Figure 1). The maps are numbered from 1 to 12 and will be referred to by their pattern number. The first pattern shows an upper air trough to the northwest of Oklahoma. This is contrasted by the 12th pattern, which shows a midlevel ridge centered over the Southern Great Plains. Patterns 7 and 8 correspond with strong to very strong upper air ridges and centers of high pressure at the surface, while patterns 3 and 6 show upper air troughs to the north and northeast; both are common features during the warm season months. The 500 hPa vector wind composites for each synoptic pattern (Figure 3) closely follow the geopotential heights. The patterns with strong ridging exhibit weak synoptic flow, generally from the south or west. Table 2 shows the percent of the total number of 1683 classified days that are within each SOM pattern. Patterns 3, 7, and 8 occur at the highest frequency, representing 14%, 12%, and 13% of all days classified, respectively. These patterns are all characterized by midlevel ridging over the study region, a common synoptic pattern in the Southern Great Plains during the warm season.

Details are in the caption following the image
Mean 500 hPa vector winds (ms−1) associated with each synoptic pattern.
Table 2. Total Number of Days and the Overall Percent of Days Classified Into Each SOM Patterna
SOM Pattern Total Days Percent of Days
1 37 2.2%
2 148 8.8%
3 237 14.1%
4 109 6.5%
5 128 7.6%
6 170 10.1%
7 201 11.9%
8 218 12.9%
9 59 3.5%
10 80 4.8%
11 152 9.0%
12 144 8.6%
  • a The total number of days classified in the study was 1683.

The contribution of unorganized convective event precipitation to total (May–September) precipitation was evaluated using observations from 120 Oklahoma Mesonet (Figure 4). The blue line in Figure 4 represents the mean precipitation contribution (%) for each synoptic pattern, calculated over all 11 years in the study period. The black intervals extend from the maximum contribution to the minimum contribution between 2002 and 2012. In general, the patterns occurring most often (3, 7, 12) are associated with the highest percent contribution. The overall average May–September precipitation contribution of any of the synoptic patterns is less than 10%, with the maximum contributions all less than 25%.

Details are in the caption following the image
May–September precipitation contribution percentage from each synoptic pattern, calculated for each year between 2002 and 2012. The blue line represents the 11 year mean contribution (% total), while the black intervals extend from the minimum to the maximum contribution.

Certain synoptic patterns tend to be more or less frequently associated with convective events because they are associated with conditions that are favorable for unorganized, deep convection to initiate. The set of 477 unorganized convective events identified by Ford et al. [2015] was used to determine if any of the 12 synoptic patterns occurred more or less frequently on unorganized convective event days. This test was implemented by selecting a group of 477 random event days and nonevent days (matching the number of unorganized convective events). We then calculated the frequency of each synoptic pattern within the group of 477 randomly chosen days. A bootstrapping procedure was employed so that the random selection of days is repeated 10,000 times. This produced a distribution of the “typical” frequency associated with each synoptic pattern (Figure 5). The mean of the distribution represents how frequently a particular pattern should occur. The actual frequency of each synoptic pattern over the 477 convective event days (black line) is then compared to these distributions to test if the frequency is statistically significantly (95% confidence level) different than what would be expected by chance (Figure 5). If the frequency of a synoptic pattern is less than the 2.5 percentile or larger than the 97.5 percentile of the frequency distribution, the pattern occurs less or more frequently on days with unorganized convection than expected by chance.

Details are in the caption following the image
Frequency distributions of synoptic patterns based on 10,000 iterations of randomly chosen sets of 477 days. The black line represents the observed frequency of each pattern for the 477 convective events. Patterns occurring at a statistically significantly (95% confidence level) different rate are denoted with an asterisk.

The frequencies of the patterns 5, 6, and 10, all characterized by upper level troughs to the north and northeast (Figure 1), are significantly lower than expected. In contrast, patterns 7, 8, and 12, all exhibiting upper level ridging over the region, occurred at a significantly higher rate than expected on days with unorganized convective precipitation. If we separate events occurring over very wet soils (>75th percentile) from those occurring over very dry soils (<25th percentile), we see that anomalous frequencies of these six patterns (5, 6, 7, 8, 10, and 12) are attributable to the occurrence, or lack thereof, of events over very dry soils (Figure 6). The trough patterns (5, 6, and 10) show a statistically significant preference for convective events to precipitate over very wet soils, with very few events initiating over dry soils. The opposite is true for the upper level ridge patterns (7, 8, and 12), which show a strong, significant preference for event precipitation to initiate over very dry soils. In general, we find that there is an association between the synoptic-scale environment and whether conditions are more or less conducive to unorganized convection. The subsequent analysis will only focus on the synoptic patterns exhibiting statistically significant changes in frequency on days with unorganized convection.

Details are in the caption following the image
Percent of convective events occurring over dry (red bars) and wet (blue bars) soils, reported for each synoptic pattern. Those exhibiting a statistically significant difference (95% confidence level) in the frequency of occurrence between wet and dry soils are denoted with a star.

3.2 Patterns Not Associated With Unorganized Convection

Synoptic patterns 5, 6, and 10 are all characterized by upper air troughs and occur less often than expected on days with unorganized convection (Figure 5). We examined HYSPLIT air mass trajectories to determine the typical origin of air masses associated with patterns 5 and 6 and compared them to the trajectories associated with convective events (Figure 7). The directions of origin for all pattern 5 and 6 event day air masses are displayed as trajectory roses (Figures 7a and 7b). These are contrasted by roses of air mass origin directions for pattern 5 and 6 days event days displayed in Figures 7c and 7d. The radial plots show the distribution of the trajectories as a percent of total days. Therefore, a trajectory-rose petal facing due south represents events, which originate from a southerly direction. For both patterns 5 and 6, a large proportion of typical trajectory origins are from southeasterly and northwesterly directions. In comparison, the trajectories associated with convective events show a clear preference for southerly and southeasterly origins; more so for pattern 5 than 6. This difference in the dominant direction of atmospheric circulation corresponds to an increase of MT weather types from just 28% of all pattern 5 days to over 48% for pattern 5 event days. A similar increase in MT weather types occurs for pattern 6, for which 24% of all days and 40% of event days are MT. In summary, we find a nearly 60% increase in the prevalence of MT weather types for patterns 5 and 6 during unorganized convective days as compared the prevalence during all days with these patterns.

Details are in the caption following the image
Trajectory roses displaying the direction of origin of HYSPLIT back trajectories for (a and b) for all days and (c and d) convective event days. The synoptic patterns shown are two of three which occur at a significantly less frequent rate during unorganized convection event days.

Changes in low-level atmospheric flow on days with unorganized convection suggest that synoptic patterns 5 and 6, combined with northerly low-level flow, are not conducive to unorganized convection. However, when southerly to southeasterly flow is combined with patterns 5 and 6, instability and corresponding likelihood of convection seemingly increases. The matrix of HYSPLIT back trajectories initiated from every pattern 5 and 6 event and nonevent day are used to determine if that day's low-level wind direction was primarily out of the north or south. For a day to be considered as having a northerly low-level wind, at least 75% of the HYSPTLIT air mass trajectories must have initiated north of the initiation point. Pattern 5 and 6 days with northerly flow (132 total) are compared against those with southerly flow (165 total) with respect to the conduciveness of atmospheric conditions to unorganized convection.

Total vertically integrated atmospheric moisture flux from northerly flow and southerly flow pattern 5 and 6 days (Figures 8a and 8b) show larger moisture flux associated with southerly flow, particularly over eastern Oklahoma. The patterns of total precipitable water follow those of moisture flux, with a much wetter atmosphere associated with southerly flow during synoptic patterns 5 and 6 (Figures 8c and 8d). Increased atmospheric moisture associated with southerly low-level winds results in significantly larger CAPE values across the entire state as compared to northerly flow days (Figure 9). This pattern is consistent for both 06:00 LST and 12:00 CAPE values, composited from NARR. Not surprisingly, southerly low-level winds advect moisture into the Southern Great Plains from the Gulf of Mexico, decreasing atmospheric stability and increasing energy available for deep convection. This likely explains why frequent northerly and northwesterly low-level winds during pattern 5 and 6 days are associated with fewer unorganized convective events.

Details are in the caption following the image
Composites of integrated moisture flux (kg m−2) for synoptic patterns 5 and 6 on (a) southerly flow days and (b) northerly flow days. (c and d) Composites of total precipitable water (mm) on the same southerly and northerly flow days, respectively.
Details are in the caption following the image
Convective available potential energy (J kg−2) at 12:00 LST from pattern 5 and 6 days with predominantly (a) northerly 850 hPa flow and (b) southerly 850 hPa flow.

Similar to patterns 5 and 6, pattern 10 occurs significantly less frequently than expected during unorganized convective events. Pattern 10 is also associated with a deep upper-level trough at 500 hPa over the Great Lakes and another trough over the eastern part of the study region. These features result in strong northwesterly to northerly winds, particularly over western Oklahoma. All of the low-level HYSPLIT trajectories for pattern 10 days (Figure 10a) and pattern 10 convective events (Figure 10b) show a predominantly northwesterly origin. Over 50% of all pattern 10 days exhibit a DM weather type, and this weather type is negatively correlated with the occurrence of unorganized convective events. An increase in DM conditions corresponds to a reduction in the number of unorganized convective events over dry and wet soils. DM days are associated with dry, cool air in the mid-to-upper atmosphere, which may inhibit instability over both wet and dry soils.

Details are in the caption following the image
Rose diagrams showing HYSPLIT back trajectory origin direction for (a) all pattern 10 days and (b) pattern 10 convective event days.

Tuttle and Davis [2006] attribute suppression of warm season convection in the Great Plains to northerly to northwesterly upper level flow, similar to this study. The northerly or northwesterly 500 hPa winds in the Southern Great Plains are driven by ridging to the southwest and a trough to the north or northeast [Tuttle and Davis, 2006]. Cool air advection, both in the midlevel and low-level atmosphere, likely reduces relative humidity near the top of the boundary layer [Zhang and Klein, 2010], while simultaneously decreasing vapor pressure deficit and boundary layer temperature (e.g., Figures 11a and 11b). The decreased vapor pressure deficit limits latent heating of the atmosphere over wet soils. Concurrently, limited surface radiation (e.g., Figure 11c) reduces sensible heating, thereby inhibiting the land surface's ability to modify the atmosphere through latent and sensible heat fluxes. These conditions are similar to the “energy-limited” evaporative regime, in which incoming solar radiation, not soil moisture, determines variations in surface evaporative fraction [Ford et al., 2014]. Examples of morning (06:00 LST) and afternoon (12:00 LST) atmospheric profiles from 22 June 2004 over Lamont, Oklahoma, demonstrate such conditions (Figure 12). Northwesterly midlevel winds and corresponding low temperatures throughout the profile result in a very stable boundary layer both in the morning and afternoon. Indeed, lifted index values at 06:00 LST and 12:00 LST on this day were 3.98 and 3.44, respectively, where positive values indicate low-level stability. This day was classified as pattern 5 and exemplifies the influence of synoptic-scale drivers suppressing instability and convection.

Details are in the caption following the image
Daily (a) vapor pressure deficit anomalies (hPa), (b) maximum temperature anomalies (°C), and (c) solar radiation anomalies (MJ m−2) averaged over all synoptic patterns 5 and 6 events.
Details are in the caption following the image
Profiles for Lamont Oklahoma from 06:00 LST (left) and 12:00 LST (right) on 22 June 2004.

3.3 Patterns Associated With Unorganized Convection

Synoptic patterns 7, 8, and 12 occur more frequently than expected on days with unorganized convection. Each of these patterns also exhibits a statistically significant preference for convective precipitation to initiate over very dry soils. Patterns 7, 8, and 12 are characterized by ridging over the study region at 500 hPa and high pressure at the surface. These conditions force weak synoptic flow at 500 hPa (Figure 3) and are typically associated with reduced cloud cover and strong evaporative demand. Vapor pressure deficit, maximum temperatures, and solar radiation across Oklahoma all show positive anomalies during pattern 7 days over dry soils (Figures 13a–13c). Similar conditions exist for synoptic patterns 8 and 12 (not shown).

Details are in the caption following the image
Daily (a) vapor pressure deficit anomalies (hPa), (b) maximum temperature anomalies (°C), and (c) solar radiation anomalies (MJ m−2) averaged over all synoptic patterns 7 and 8 events.

The mechanisms linking these synoptic patterns to convection over dry soils likely involve high pressure at the surface and an increase in net radiation, which results in increased vapor pressure deficit, decreased soil moisture content, and increased maximum temperatures. These factors lead to increased sensible heating of the near-surface atmosphere [Ford et al., 2015]. With sufficient uplift and a dry, weakly stratified free troposphere, increased surface heating can erode convective inhibition and the boundary layer can grow. As cooler free-tropospheric air is entrained, temperatures near the PBL top decrease, and the air reaches saturation [Ek and Holtslag, 2004]. The result is convective cloud formation and potential precipitation initiation over the dry soils. The atmospheric profiles over Lamont, Oklahoma, on 10 August 2006 at 06:00 LST and 12:00 LST provide an example of this pattern (Figure 14). Relatively weak midlevel flow and dry soils permit sufficient sensible heating to erode morning convective inhibition and grow the boundary layer. Dry, cool air entrainment allows for saturated or near-saturated conditions at the PBL top, and therefore convective cloud formation. This day was classified as a pattern 8, representing a synoptic environment conducive for convective initiation over dry soils.

Details are in the caption following the image
Profiles for Lamont Oklahoma from 06:00 LST (left) to 12:00 LST (right) on 10 August 2006.

4 Summary and Conclusions

One challenge in land-atmosphere interaction studies is the strong dependence of the results on the analysis scale [Jones and Brunsell, 2009; Taylor et al., 2013]. Many studies have focused on local-scale interactions [Findell and Eltahir, 2003; Ek and Holtslag, 2004; Santanello et al., 2009; Gentine et al., 2013; Tawfik and Dirmeyer, 2014]. These studies provide strong evidence that atmospheric modification by the land surface can lead to convection. Studies examining soil moisture-atmosphere interactions at the mesoscale (tens of kilometers) have shown that convective initiation occurs preferentially along surface energy and moisture gradients [Taylor and Ellis, 2006; Taylor et al., 2011; Couvreux et al., 2012] and land use/land cover boundaries [McPherson et al., 2004]. Continental-to-global scale studies have demonstrated the relationship between soil moisture and precipitation as a function of evaporative (moisture) regime [Seneviratne et al., 2010], with the strongest interactions in transition zones between arid and humid climates [Koster et al., 2004; Wei and Dirmeyer, 2012].

In this study, we combine self-organizing maps with the spatial synoptic classification and HYSPLIT air mass trajectories to determine if the synoptic-scale environment modulates the ability of the land surface to force unorganized convection in Oklahoma. We identify several synoptic patterns which are significantly more or less frequently associated with unorganized convection. Patterns less frequently associated with convective events show a significant preference for precipitation to initiate over wetter soils. These SOM patterns (5, 6, and 10) are characterized by upper air troughs to the north and northeast that drive northerly to northwesterly 500 hPa and 850 hPa winds, making for conditions less conducive (i.e., weaker CAPE) for convection to initiate [Zhang and Klein, 2010]. Those patterns, which are more frequently associated with convective events, are characterized by dominant 500 hPa ridging over the study region. These patterns are associated with increased net radiation and more homogenously dry soils. Consistent with these surface conditions, unorganized convective events that occur during these patterns (7, 8, and 12) have precipitation initiating preferentially over dry soils. The dry soil-precipitation feedback in these cases depends on synoptic-scale atmospheric conditions that allow sufficient surface heating through increased solar radiation and vapor pressure deficit. Based on the results presented here, we argue that the synoptic-scale environment can potentially force conditions that are more or less conducive to convection forced by the land surface.

Soil moisture anomalies modify the atmosphere primarily through control of surface evaporative fraction and local-scale feedback that act to suppress or enhance convective activity [Guillod et al., 2014]. However, the strength of this feedback depends on the overlying atmospheric conditions, namely, the slope of potential temperature and stratification of free tropospheric air [Westra et al., 2012; Gentine et al., 2013]. Mesoscale gradients in surface heat flux and moisture availability can also enhance instability and increase the probability of convection [Frye and Mote, 2010; Taylor et al., 2011]. Our results suggest the synoptic-scale environment can inhibit or enhance local-scale soil moisture feedback. It is therefore important to consider local-scale soil moisture-atmosphere interactions in the context of the overlying synoptic-scale atmosphere.

The synoptic-scale environment is also embedded within a larger (continental-scale) climatic context, and this, too, modulates the synoptic-scale, mesoscale, and local-scale feedback. For example, our study was conducted in a semiarid environment frequently identified as a “hot spot” of land-atmosphere interactions [Koster et al., 2004]. If this study were conducted in a more humid or arid environment, it is expected that the results would be considerably different. Overall, forcing at various spatial scales is integrated into land surface feedback to the atmosphere. Therefore, future studies of soil moisture-precipitation interactions should consider how this scale-dependency may impact their results.


We gratefully acknowledge the National Science Foundation (award number: BCS-1433881) for funding this work. We would also like to thank the Oklahoma Mesonet for providing soil moisture observations used in this work. Data files from the Mesonet can be obtained online at http://www.mesonet.org/index.php/weather/category/past_data_files. Data products of the North American Regional Reanalysis can be obtained online at http://www.esrl.noaa.gov/psd/data/gridded/data.narr.html. The HYSPLIT model can be run online or downloaded from http://ready.arl.noaa.gov/HYSPLIT.php. Atmospheric profiles are provided by the University of Wyoming Department of Atmospheric Science and can be found at http://weather.uwyo.edu/upperair/sounding.html. We thank the anonymous reviewers whose insightful comments improved this manuscript.