The ability to understand and predict key life history events for insect pests plays a vital role in their management (Ham & Hertel 1984; Clarke et al. 2016). For instance, the pales weevil, Hylobius pales (Herbst) (Coleoptera: Curculionidae) is attracted to fresh cut pine stands. If a stand is replanted too early, their brood can decimate the newly planted pine seedlings. Forest managers may employ a silvicultural control tactic of cutting early in the yr (before Jul) and planting the following winter, allowing the pales weevil to complete its life cycle and leave the stand before the seedlings are planted (Nord et al. 1984). The Nantucket pine tip moth, Rhyacionia frustrana (Comstock) (Lepidoptera: Tortricidae) can be a pest of young pine and Christmas tree plantations. Insecticide sprays may be used to control their populations (Berisford et al. 1984). However, effective control requires nursery managers to be able to predict R. frustrana egg hatch and larval development in order to time the application of insecticides to coincide with these life history events (Douce et al. 2002). The aforementioned examples express why researchers have developed predictive models (Gargiullo et al. 1985; Kumral et al. 2007; Knutson & Muegge 2010; Akotsen-Mensah et al. 2011; Haavik et al. 2013) or observational cues (Mussey & Potter 1997; Herms 2004; Reding et al. 2013; Hartshorn et al. 2016) to predict key life history events for many serious insect pests.

The southern pine beetle, Dendroctonus frontalis (Zimmerman) (Coleoptera: Curculionidae), can be a severe pest of all southern pine species, but most notably loblolly pine (Pinus taeda L.), shortleaf pine (Pinus echinata Mill.), slash pine (Pinus elliottii Englm.), and longleaf pine (Pinus palustris Mill.) (all Pinaceae) (Payne 1980; Blanche et al. 1983). Approximately 84% of total pine timber losses across 10 southeastern states of the US between 1977 and 2004 can be attributed to 4 large D. frontalis outbreaks (Pye et al. 2004).

A regional annual D. frontalis risk assessment survey was developed in the 1980s and became an important part of the integrated pest management strategy developed for D. frontalis (Billings 1988; Billings & Upton 2010). The survey is an early warning system used to predict the D. frontalis population status and infestation trends for the current yr. This prediction allows forest managers to appropriate adequate resources to address potential D. frontalis outbreaks (Billings & Upton 2010). The survey is conducted currently over a consecutive 4 wk period in the spring using Lindgren 12-unit funnel traps (Lindgren 1983) baited with polyethylene bags containing 70% α and 30% β pinene (released at about 5 g per d) along with D. frontalis aggregation pheromones frontalin (released at about 5 mg per d) and endo-brevicomin (Billings 2011; Sullivan 2016). State and federal forest agencies deploy traps in the host pine forests throughout each of the 16 participating states (Billings 2011). Trap catches are collected weekly, and numbers of D. frontalis and their most significant invertebrate predator, the checkered clerid beetle, Thanasimus dubius (Fabricius) (Coleoptera: Cleridae), are tallied. The mean number of D. frontalis per trap per d and the ratio of D. frontalis to T. dubius are used to derive a prediction for a given locality. The predictions were initially obtained from a chart developed and revised by the Texas Forest Service (Billings & Upton 2010).

The survey's ability to accurately assess D. frontalis population levels has been variable in recent yr. In Mississippi, outbreaks occurred on the Homochitto National Forest in 2012, the Tombigbee National Forest in 2014, and the Bienville National Forest in 2015 (Asaro et al. 2017); however, the survey projected population trend/levels to be static/low, decreasing/moderate, or increasing/low, respectively, for each outbreak occurrence (Table 1). A variety of factors may affect the predictive power of the survey. The chemistry of lures used was changed in 2007, because polyethylene bags of (70% α-pinene to 30% β-pinene) replaced steam-distilled turpentine volatilized from a wicked bottle as the host compound component (Billings 2011). This change was due to a lack of commercially available sources of turpentine (Sullivan 2016). Endo-brevicomin, which synergizes the attractiveness of the frontalin lure (Sullivan & Mori 2009), is now also included (Billings 2017). In addition, trap placement recommendations have changed, because traps must be placed 20+ m from the nearest host pine to reduce the risk of spillover attacks on adjacent pines now that endo-brevicomin is used (Stephen Clarke, personal communication).

Table 1.

Dendroctonus frontalis outbreaks in Mississippi and the prediction results from the annual spring survey.

In addition to the factors detailed above, the trap timing is important in ensuring an accurate assessment of existing spring D. frontalis population levels. Trap deployment must coincide with the peak of D. frontalis spring flight activity (Billings & Upton 2010). Spring D. frontalis flight activity generally occurs within a 3 mo time frame, usually with a 3 to 6 wk peak period (Friedenberg et al. 2007). Predicting the peak is difficult because all life stages of D. frontalis overwinter (Lombardero et al. 2000). Further development or even emergence can occur during periods of favorable winter temperatures (Moser & Dell 1979). Climate change may also affect the timing of bark beetle spring dispersal flight (Jönsson et al. 2009; Milton & Ferrenberg 2012). Multiple emergence peaks due to variable spring temperatures may influence population levels in subsequent mo because they may affect the ability of D. frontalis to allocate a sufficient number of beetles to mass attack pines and initiate an infestation (Friedenberg et al. 2007). Therefore, it is crucial that forest managers have a practical means of predicting the peak or peaks of spring dispersal by D. frontalis.

Peak spring dispersal of D. frontalis has been anecdotally associated with the blooming phenology of various indigenous tree species (Hopkins 1909). Flowering of eastern redbud (Cercis canadensis L.; Fabaceae) (St. George & Beal 1929), pollen release of loblolly pine (P. taeda) (Billings 1988), and flowering of flowering dogwood (Cornus florida L.; Cornaceae) (Thatcher & Barry 1982; Billings 1988) all have been suggested as indicators for the onset of peak D. frontalis spring dispersal. The onset of flowering dogwood bloom was the protocol for trap deployment of the annual spring D. frontalis risk assessment survey for several decades (Billings 1988). Because the actual flowers of flowering dogwood are inconspicuous, the onset of bloom refers to the white bracts that open before the flower buds. In 2017, regional spring trapping guidelines from the USDA Forest Service were revised to suggest the use of the bloom of redbuds instead of dogwoods as a phenological cue for peak D. frontalis spring dispersal (Billings 2017).

The synchronicity between peak spring D. frontalis dispersal and the phenology of local tree species have been based solely on observations, and analyses to assess these claims are lacking. Sub-optimal timing of trapping may have contributed to the recent failures of the annual survey to accurately predict local outbreaks. Therefore, we conducted studies to (1) quantify if recent survey dates were optimally timed to encompass the peak in D. frontalis spring dispersal, (2) determine if the onset of flowering dogwood bloom is a good predictor of peak D. frontalis flight activity, (3) compare bloom phenology of redbud and dogwood trees, and (4) evaluate the viability of a 4 to 6 wk survey for describing D. frontalis spring flight activity.

Materials and Methods

DENDROCTONUS FRONTALIS TRAPPING

We conducted yr-round D. frontalis trapping to monitor flight activity to identify when peak spring dispersal occurred for a given location. Two Mississippi locations (Oktibbeha County and Homochitto National Forest) and 1 in Alachua County, Florida, USA, were surveyed; henceforth these trapping locations will be referred to as Oktibbeha, Homochitto, and Alachua. There were 2 trap sites in 2014 and 5 trap sites in 2015 in Oktibbeha (Table 2). The number of trapping sites remained constant throughout the study period for both Alachua and Homochitto, with 1 and 3 sites, respectively. The traps remained deployed for the entirety of 2014 and 2015 except in 2014 on the Homochitto, when the traps were taken down on 6 May 2014 and redeployed 1 Jan 2015. The extended trap deployment required lures to be changed every 4 wk to ensure the baits remained attractive to D. frontalis. All Lindgren 12-funnel traps were baited identically to the annual spring survey traps. All lures were purchased from Synergy Semiochemicals Corporation (Vancouver, British Columbia, Canada). The frontalin and pinene lures were affixed to the trap while the endo-brevicomin was attached to a twig approximately 4 m from the trap (Sullivan & Mori 2009). All trap locations were 20+ m from the nearest pine in hardwood bottom lands adjacent to pine stands. Traps were hung so that the collection cups were approximately 1 to 2 m above the ground, maintaining uniformity with the spring survey trapping. All trap captures were collected weekly.

For the purposes of this study, we considered the potential spring flight season of D. frontalis to occur between 1 Jan and 31 May of each yr. This 5 mo range was early enough to capture the earliest late winter flights (Moser & Dell 1979), and long enough to allow the spring dispersal flight to conclude. The weekly D. frontalis and T. dubius captures from yr-round traps were tallied and recorded by location and trap. Then all traps in a location were summed to provide the weekly total of D. frontalis and T. dubius trap captures for each location. The peak D. frontalis and T. dubius spring dispersal period was defined as the continuous 4 wk period that the observed trap captures were greater than any other continuous 4 wk period. Weekly totals also were converted to a percentage of the 5 mo total spring trap captures to evaluate the yearly variation in population size (Akotsen-Mensah et al. 2011). The percentage of all beetles collected during the peak period was calculated.

Table 2.

Trap site coordinates for monitoring 2014 and 2015 yr-round Dendroctonus frontalis flight activity.

Results

FLOWERING DOGWOOD BLOOM VS. EASTERN REDBUD BLOOM

Eastern redbud bloom occurred approximately 2.3 wk before flowering dogwood bloom in 2019 (Fig. 2). The mean (± SE) onset of eastern redbud bloom occurred 10.6 ± 0.13 wk after the first of the yr, vs. the mean onset of flowering dogwood bloom which occurred 12.9 ± 0.07 wk after the first of the yr. The variability of onset of bloom dates was over 2× greater in redbuds than in dogwoods, with 2.4 wk between the first and last eastern redbud bloom dates, and only 1.1 wk between first and last bloom dates for flowering dogwood trees (Fig. 2).

Table 3.

The timing of flowering dogwood bloom (DW) peak, Dendroctonus frontalis – southern pine beetle (SPB) spring dispersal, and initiation of annual Dendroctonus frontalis spring survey along with the corresponding percentage of Dendroctonus frontalis captured in nearby yr-round traps.

Discussion

The ability of the D. frontalis spring survey to predict outbreaks has been unreliable in recent yr (Table 1), potentially due in part to sub-optimal timing of trap deployment. Despite historical anecdotes to the contrary, dogwood bloom proved to not be an effective predictor of D. frontalis spring dispersal. Within the confines of this study, the best case scenario for using dogwood bloom as the phenological indicator of trap deployment resulted in missing 45% of the total D. frontalis spring dispersal, whereas in the worst case 92% of the spring dispersal passed before dogwood bloom. This suggests that historical timing of trap deployment could at least explain some of the recent inaccuracy with the annual spring survey. This also suggests that the survey's prediction model is based on deflated peak data, and improvements to timing of the survey will require the model to be recalibrated. The onset of eastern redbud bloom appeared a better phenological indicator of optimal D. frontalis spring survey timing. Our 2019 blooming survey indicated redbud bloom occurred 2.3 wk before dogwood bloom, narrowing the average 3 wk offset between D. frontalis peak dispersal and dogwood bloom measured in 2014 and 2015. However, eastern redbud bloom was variable highly among the trees monitored in 2019, creating the possibility of considerable asynchrony between onset of eastern redbud bloom and onset of peak D. frontalis dispersal.

Fig. 1.

Comparison of dogwood bloom dates, Dendroctonus frontalis (southern pine beetle) and clerid peak dispersal dates from yr-round traps, and dates of the actual southern pine beetle spring survey. The black bar represents the consecutive 4 wk period beginning with the median onset of dogwood bloom by yr and location. The dark gray bar represents the 4 consecutive wk that the annual southern pine beetle spring trapping survey was conducted. The light gray (southern pine beetle) and white (clerid) bars represent the consecutive 4 wk period during which the most beetles were trapped in yr-round traps. The percentages following the bars correspond to the percentage of southern pine beetle (black, dark gray, light gray) or clerid (white) spring trap captures from yr-round traps.

Fig. 2.

The 2019 blooming dates of flowering dogwood (DW) and eastern redbud (RB) trees (N = 25 each) in an unmanaged woodlot on the periphery of the Thad Cochran Research, Technology & Economic Development Park at Mississippi State University, in Starkville, Mississippi, USA (33.475129°N, 88.793119°W).

Dogwood bloom may have historically coincided with peak D. frontalis spring dispersal, but scientific studies are lacking. If a correlation did once exist, climate change may have altered the phenology of both species (Davis et al. 2010). Dogwoods must be exposed to sufficiently cold temperatures for a long enough period before growth resumption can be initiated within the bud (Hunter & Lechowicz 1992). Once the chilling requirement has been met, bud burst occurs after exposure to a specific amount of thermal time above a temperature threshold has accumulated (Cannell & Smith 1986), which can result in bud burst occurring later in the spring after atypically warm winters occur. Conversely, D. frontalis may have as many as 8 overlapping generations per yr (Hain et al. 2011). They overwinter in all life stages and continue to develop even at 0 °C (Lombardero et al. 2000). Adults may emerge and disperse after a few unseasonably warm d in the winter and early spring (Moser & Dell 1979). Cold winters should promote a synchronous spring emergence, particularly in the northern portion of the range of D. frontalis and higher elevation pine forests (Lombadero et al. 2018). Warmer winters result in staggered emergence as was observed in our study in southern states. Climate change likely will increase instances of multimodal spring emergence by D. frontalis, complicating the use of any phenological indicator for survey timing and continuing to impact the accuracy of the prediction model.

One method to overcome increased variability in D. frontalis spring emergence would be to extend the trapping period beyond the current 4 wk standard. Our results indicate that 6 wk of trapping, the maximum recommended life of the lures, only slightly would improve the chances of capturing peak emergence or the presence of multimodal emergence. Trapping for longer than 6 wk would add additional cost and require increased labor for trap collection and beetle counting. Therefore, the practicality of using longer trapping periods is minimal. Another technique would be the development of a robust model designed to determine optimal survey start dates based on weather conditions and thresholds of D. frontalis developmental and flight temperatures. Degree d models usually are good candidates for such models because they can successfully predict key life history events for insects depending on weather conditions. The potential winter emergence of D. frontalis would make selecting a biologically meaningful time to begin accumulating degree d for D. frontalis difficult. Another option would be to base the timing of the survey on T. dubius emergence. Their abundance in relation to D. frontalis numbers is an input in the prediction model, and our results suggest less variability in emergence patterns. The utility of this approach requires further study.

As discussed above, using a phenological indicator to time the surveys may be impractical due to variable emergence patterns driven by climate change. It is apparent from our results that survey trappers already are using factors other than dogwood or redbud bloom to deploy their traps (Fig. 1). The inconsistency in survey timing could possibly be explained by time constraints with other management duties such as timber harvests and prescribed burns. Trappers may have used historic trapping data for their region to help determine deployment dates. Local knowledge of climatic conditions and emergence patterns of D. frontalis may serve as the best source of determining when to begin the spring survey in the absence of a phenological cue.

In addition to issues of survey timing, recent problems in predicting D. frontalis outbreaks may be due in part to changes in the lure combination previously described. A recent study has demonstrated that the switch to the α- and β-pinene sleeve as the host component has reduced the trap catch of D. frontalis compared to using turpentine (unpublished data). More recently, an endo-brevicomin lure was included to synergize the attractiveness of the other 2 lures and enhance the survey's ability to detect low population levels of D. frontalis. Differences in the responsiveness of D. frontalis to these new lure components may explain the recent model failures partially. Recalibrating the model for the current lure scheme also could improve the predictive power of the survey.

The inclusion of endo-brevicomin also altered trap placement. Trap locations are now typically in hardwood inclusions within pine stands. Hardwood green leaf volatiles have been shown to significantly decrease D. frontalis trap captures (Dickens et al. 1992; Sullivan et al. 2007), thus the current displacement allows for more non-host species between the trap and preferred D. frontalis habitat. Changing trap locations frequently may affect survey results. Ideally trappers would examine trap catches annually and relocate traps from sites that historically collected very few D. frontalis even when population levels are moderate to high. Establishing and continuously using reliable trap sites would provide consistency in the survey and aid in recalibrating the model to improve the validity of the results. However, maintaining the same sites from yr to yr often is confounded due to turnover in staff and landscape changes from management activities, storm events, etc.

Given the uncertainty in survey timing, revising the prediction model to include climate data could help improve the model accuracy back to previous standards. A forecast system using weather and stand data, previous yr infestation levels, and a hydrological model to predict D. frontalis levels for a county has been developed (McNulty et al. 1998, McNulty 2019). Because an operational version of the model is a recent development, little information has been provided to potential users to date and the short- and long-term accuracy of the results have not been thoroughly evaluated. Perhaps a combination of the 2 prediction methods may serve to improve the overall ability of forest managers to anticipate and prepare for outbreaks.

The spring survey is an integral part of the integrated pest management strategy for D. frontalis. In addition to helping predict seasonal infestation levels, survey results are valuable for preserving a historical record of D. frontalis population trends. The annual trapping also keeps foresters aware of the impacts of D. frontalis and the ecological and economic consequences of outbreaks. A better prediction system would provide additional justification for maintaining the survey. Our results indicate the following could help improve the efficacy of the prediction model: (1) shift trap deployment earlier than traditional dates, perhaps using eastern redbud bloom as a cue; (2) include local knowledge to determine when the trapping period should occur; and (3) incorporate climatic data with trap catch numbers in the model.

Fig. 3.

Southern pine beetle (SPB) and checkered clerid beetle spring trap captures from yr-round baited Lindgren funnel traps. For each site by yr combination, number of beetles captured per trap was analyzed as a smooth function of survey wk using a generalized additive model (GAM) to show peaks in capture data. Black lines are fit GAMs for southern pine beetle, whereas gray lines are the fit GAMs for clerid beetle. Produced using Poisson distributions and log link functions with the mgcv package (Wood 2011) in R statistical software (R Core Team 2020).