The foundation of agroecological-oriented areawide pest management (AWPM) applied to crop protection is the recognition that pest suppression is innate to the agroecosystem. This innate ability to suppress pests is supported by specific pest management tactics that are regionally applied (1). An operational distinction between classical pest management tactics and AWPM is that the classical tactics tend to focus on individual fields (2), whereas AWPM highlights spatial variability of agricultural and environmental conditions that influence pests (1). The underlying premise of AWPM is that many economic pests of agriculture are more effectively managed using pest management tactics applied in a coordinated strategy over a large area than by using a more traditional approach where tactics are applied to individual fields without considering pest management synergies areawide (3).

Natural enemies are an important pillar of the agroecosystem response (4) but their effectiveness in controlling pests may be impacted by agricultural processes and weather conditions. For example, extensive monocultures in agroecological landscape could be expected to negatively impact populations of natural enemies, although the relationship is not straightforward due to complex processes and interactions in the field and indirect effects of other factors, such as timing (5). Habitat manipulation in direct neighborhoods of crop fields is thought to have the potential to counteract these negative processes but in-depth knowledge of the system as a whole and of its components individually is often insufficient (6). Populations of natural enemies and the biological control services they provide are often affected by grain size of the landscape, with fine grained landscapes generally providing greater biological control of crop pests than coarse grained landscapes (7). In addition to effects of grain size, connectivity of the landscape for natural enemies can be increased by the presence of corridors of acceptable habitat and by field edges both of which can serve as habitat and conduits for dispersal of natural enemies across the landscape (7, 8). A robust general rule may not be feasible as various aspects of landscape structure may have different effects on various components of natural pest control (9), and these effects can vary both spatially and temporally (7, 10).

Invasive aphid, Melanaphis sorghi (Theobald) [sorghum aphid; previously published as sugarcane aphid, Melanaphis sacchari (Zehntner) (Hemiptera: Aphididae) (11)] was recognized as a major pest of sorghum [Sorghum bicolor (L.) Moench (Poales: Poaceae)] in the Great Plains of North America in 2013 (12). It has since become a persistent pest of sorghum in nearly all sorghum producing regions of the United States, with invasions recurring annually. The aphid may cause substantial economic loss through negative impact on sorghum plants and through negative impact on harvesting efficiency due to sticky honeydew excretion buildup on leaves (12, 13) but the impact can vary regionally with climate and types of sorghum grown (14). Multiple aspects of the aphid have been studied, including its biology (15, 16), distribution (17), dispersal (18), economic impact (13, 19), and forecasting (20, 21). Areawide pest management has been proposed as a proper strategy to manage the pest (14). A promising, intensely explored, research avenue in the management of sorghum aphid focuses on the natural enemies of sorghum aphid (10, 22–25), consistent with agroecological-oriented AWPM (1).

In the context of the most recent invasions, studies on natural enemies of sorghum aphid have focused on two important functional groups, predators and parasitoids. These species were studied in the context of environmental factors, such as climate, weather and landscape. Faris et al. (23) identified 19 predatory and parasitoid species preying on sorghum aphid and showed that these natural enemies may reside in Johnson grass [Sorghum halepense (L.) Pers.] and in riparian areas during off-season. Additionally, Faris et al. (24) showed that suppression of the aphid by natural enemies is effective on both resistant and susceptible sorghum hybrids. Elkins et al. (10) found that both landscape composition and configuration were associated with the level of sorghum aphid and natural enemy abundance, and higher landscape complexity had negative impact on the numerical response of parasitoids, but not on the numerical response of predators. Despite annual variability of their results, they showed that various aspects of aphid and natural enemy dynamics may need to be considered more holistically. On a larger spatial scale, on which agro-landscape and weather metrics show greater differentiation, the relationships between these metrics and natural enemy dispersal, density, and population growth may vary by region, with landscape composition likely impacting some natural enemy dynamics (11). These studies are important in-depth contributions to the growing body of knowledge on natural pest control, and more broadly, on leveraging ecosystem services in the agroecological context (9, 26–28). Nevertheless, longitudinal data is needed to better understand the dynamics of top-down control of the population of pests (29).

In this study we focus on sorghum aphid and its natural enemies, predatory lady beetles (Coleoptera: Coccinellidae) and parasitoid wasps Aphelinus nigritus Howard (Hymenoptera: Aphelinidae) and Lysiphlebus testaceipes (Cresson) (Hymenoptera: Braconidae) due to their widespread presence and importance to sorghum aphid population dynamics (25). Building on a major data collection effort over 5 years and three US states, we use statistical methods to investigate several spatiotemporal aspects of insect population dynamics. We relate those aspects to environmental characteristics of sorghum fields, the primary habitat of sorghum aphid, the surrounding landscapes, and local weather conditions. The findings will define spatiotemporal attributes with significant implications to sorghum aphid AWPM on the Great Plains.

Correlation for pairs of dependent variables was low (Spearman correlation; r_S < 0.6) and varied annually, with a few exceptions ( Figure 5 ). In particular, correlation through the years was consistently high between nSAmax and nLBmax (but not between nSAmax and nMMmax). Correlation was also consistently high for nSAmax and rLBMMmax, and for dtSAmax and vSA. In general, correlation was low for pairs of dependent and independent variables, with a few year-specific exceptions (e.g., for dtSAmax and sdTMAX, r_S = 0.65 in 2017 and r_S < 0.5 in 2018 and 2019).

Considering the sets of alternative models with ΔAICc < 2, we computed the frequency with which a given independent variable was included in the set of top alternative models for each dependent variable ( Figure 6 ). PLANDs was present in all the alternative models for nMMmax, rLBMMmax, and vSA. The frequency of PLANDs in the case of nSAmax was moderate, but PD was included in eight out of nine nSAmax models. PD was a commonly included variable for nLBmax, rLBMMmax, and vSA. The models for nLBmax and nMMmax had opposite tendencies regarding the inclusion of sorghum related variables PLANDs and PD. Inclusion of PLANDwh and PLANDc seemed also complimentary. PLANDwh was included in all top alternative models for nSAmax, whereas PLANDc was frequently included in models for nLBmax, nMMmax, and vLB. Except for nLBmax, the frequency of SIDI, an overall index of landscape diversity within buffers, was rather low, perhaps due to its low spatiotemporal variation ( Figure 3 ). Among the weather variables, meanPPT was more commonly included than maxTMAX, especially in models for nSAmax, nMMmax, dtSAmax, and vLB. Interestingly, the frequency of meanPPT was high for nMMmax and vLB, but not for nLBmax and vMM, whereas the opposite was true for maxTMAX. The frequencies of sdTMAX and cvPPT were low to moderate. The frequencies of inclusion of independent variables align well with the sets of independent variables selected for optimal models ( Table 5 ). In summary, a PLAND of a specific crop was preferred as an independent variable, and the overall diversity of the landscape (SIDI) was typically less preferred. The landscape configuration metric PD was frequently included in alternative models for most dependent variables. Moreover, the main weather effects, and not their respective variability metrics, were the preferred independent variables among the weather metrics, although their effects seemed to be either redundant or mutually exclusive rather than synergistic.

Multiple factors, and likely their interactions, have impact on the phenomena described by the ten insect metrics. In particular, in the case of metrics that describe activity of parasitoids, the number of top alternative models (ΔAICc < 2), and thus alternative combinations of landscape and weather metrics that fit data best, was relatively high – 19 for nMMmax, 21 for rLBMMmax, and 26 for vMM ( Figure 6 ). Specifically, R² for the optimal nMMmax model equaled 0.252 (adjusted R² = 0.224; Table 5 ). To the contrary, there were only four top alternative nLBmax models with low AICc scores ( Figure 6 ), but far less variability in nLBmax was explained by the optimal model, with a much lower R² of 0.104 and adjusted R² of 0.069 ( Table 5 ). In the case of the far more host-specific sorghum aphid, the R² for the optimal model was 0.151, and the adjusted R² was 0.126 ( Table 5 ). Despite the high correlation between nSAmax and nLBmax (but not for nSAmax and nMMmax), consistent through the years ( Figure 5 ), pairwise overlap in most frequently included independent variables was surprisingly low, and PD was the only common frequently included metric in both nSAmax models and nLBmax models ( Figure 6 ). For the pair nSAmax and nMMmax, meanPPT was a common frequently included metric, and for nLBmax and nMMmax the common metric was PLANDc ( Figure 6 ). Pairwise correlation was also consistently high for the pairs nSAmax and rLBMMmax, and for dtSAmax and vSA. Despite the high correlation, in both cases, the frequencies of inclusion of independent variables were largely different. Judging by the frequency of independent variable inclusion, the speed of increase in sorghum aphid abundance vSA, unlike vLB and vMM, was dependent on both presence of sorghum and configuration of sorghum fields (PLANDs and PD, respectively), and to a lesser degree on meanPPT and SIDI. The speed of increase in lady beetle abundance vLB was more dependent on the presence of cotton (PLANDc) and meanPPT, although both PLANDc and PLANDwh were included in the optimal model. The speed of increase in parasitoid abundance could be explained by 26 top alternative models (ΔAICc < 2), and the frequencies of independent variable inclusion were rather moderate and more evenly distributed. PLANDwh was more commonly included variable than PLANDc in the top alternative models, although the optimal model included PLANDc. For both vLB and vMM, the more common sorghum-related variable was the landscape configuration metric PD rather than the landscape composition metric PLANDs. In summary, for a given insect category, the maximum numbers of insects and their speed of increase in abundance tend to be explained by very different sets of independent variables. In other words, different environmental factors shape different aspects of insect population dynamics in the field.

To examine changes in insect counts through time, the numbers of sorghum aphids, lady beetles and mummies were plotted as a function of time ( Figures 7 – 9 ). The temporal field sampling resolution allowed us to capture the emergence of patterns in insect population dynamics. The patterns in three individual examples from 2017 ( Figure 7 ; IDs 81, 149 and 150) and all individual examples from 2018 and 2019 ( Figures 8 , 9 ) resemble the classic predator-prey population dynamics, with spikes in numbers of predatory lady beetles and the parasitoid wasps following in time the spikes in numbers of sorghum aphids. Evident departure from this pattern can be observed in one of the examples in 2017 ( Figure 7 ; ID 130), where the rapid increase of the counts of mummies seems to precede population growth of aphids. Population declines of prey and delayed declines in population count of predators and parasitoids can also be observed in some cases (e.g., year 2017, ID130; Figure 7 ), although the fate of the insects is unknown. The data does not prove causative relationships, but the patterns are consistent through the years. In general, in N GP the response of parasitoids was stronger than lady beetles during all three years, but in S GP the responses varied annually. In 2018 the response of the predators appeared stronger overall, whereas in 2019 the response of parasitoids was stronger. The responses were relatively weaker in 2017. Since each individual example reflects a series of measurements from one field, the measurements, and the resulting fluctuations, may be influenced by the stochastic nature of data sampling.

In this work we investigated important aspects of population dynamics of sorghum aphid and its natural enemies in the context of variable landscape and weather conditions. Our work highlights the value of longitudinal data that represent both the breadth of environmental conditions that typically change through space and their temporal variability within and between seasons. In the case of wind-borne insects, the spatial aspects of data collection should be perceived through the lens of variability in the local environmental conditions rather than the physical distances of sampling sites. Physical distance may not directly correlate with the effects of environmental conditions upon the studied system and may not be the main factor that limits distribution (44). The choice of sampling locations should account for these aspects.

Spatiotemporal effects are particularly important in the case of AWPM, in which timely and synchronized activities across distant locations are a key consideration. Spatial effects are often considered from the perspective of temporal variability (annual or seasonal variability of a given factor at a given location), but temporal effects are considered arguably less often in the context of space, both physical (differences in temporal variability of a given factor at different physical locations) and environmental (differences in temporal variability of a given factor under different environmental conditions). Clinal changes in environmental characteristics can be expected across extensive landscapes such as the Great Plains of North America, but our study shows that various measures of species responses to changing environment may show different patterns. For example, our data show that in the case of sorghum aphid the time to maximum aphid count per leaf, dtSAmax, is longer in N GP than in S and S GP, but differences in nSAmax and vSA between S, S GP and N GP are not consistent between seasons. It appears that changing local effects (and likely their interactions) have varying impact on the three sorghum aphid metrics ( Figure 2 ). Similarly, maximum numbers of lady beetles and parasitoids (nLBmax and nMMmax, respectively) and their presumed response times (dtRespLB and dtRespMM, respectively) also did not show consistent spatiotemporal trends ( Figure 2 ). Patterns that emerge from changes in observed insect numbers through time support these observations ( Figures 7 - 9 ). As the environmental conditions change in space and time, observing one of the population metrics in isolation may lead to biased conclusions regarding species dynamics.

Considering landscape features, crop-related metrics and SIDI were the most frequently selected metrics in our regression analysis. The configuration metric PD was preferred over the related composition metric PLANDs in models for some dependent variables (e.g., for nLBmax), PLANDs was preferred over PD for some others (e.g., for nMMmax), and in some cases both were frequently used (e.g., vSA). Therefore, both descriptors of landscape should be considered in future studies. Weather effects were represented in top alternative models predominantly by maximum values of maximum temperature (maxTMAX) and by mean precipitation (meanPPT) ( Figure 6 ), and the related variability metrics were selected less frequently. Analysis of longer-term data with alternative spatial resolutions may help better understand the impact of weather variability metrics, especially for those dependent variables for which these metrics were not considered ( Table 4 ). Nevertheless, the importance of weather variability metrics may depend on the metric itself and on specific study design.

Our analysis highlights the complexity of ecological interactions in the field and indicates the need for a broad consideration of landscape and weather features in AWPM studies. Alternative inclusion of landscape composition and configuration metrics for various species-specific regression models indicates complementary nature of the two, but also indicates that these metrics should not be associated with a species or with a specific aspect of insect activity. Instead, the category of landscape metrics that best explains a given phenomenon likely depends on a broader environmental context. For example, although cotton and wheat habitats were more relevant for our analysis than grassland, pasture, and herbaceous and woodland habitats, PLANDwh was included in all top alternative models for maximum number of sorghum aphids nSAmax and only rarely for nLBmax and nMMmax, and PLANDc showed the opposite trends. Analysis focused on sorghum aphid alone would likely miss the effects of PLANDc on the predator-parasitoid-prey system. Moreover, the two groups of natural enemies responded differently to landscape composition and configuration features as well as to the weather conditions. Therefore, there seems to be no single metric or a set of metrics that can holistically describe the activity of sorghum aphid and of its specific natural enemy in the field. Instead, various aspects of the insect population dynamics need to be explained in light of different environmental factors. Moreover, it may be important to retain the hierarchy between the metrics as landscape components (landscape composition) may need to be considered first before landscape features (landscape configuration) can be analyzed or their effects interpreted.

The complexity of intra- and inter-specific insect dynamics in the field is also evident in the large portion of variation not explained by regression models in our study. Impact of landscape and climate and weather factors not considered here should be investigated further in future studies. For example, landscape configuration metrics for land cover categories other than sorghum may provide more insight into the effects of those categories. Deeper understanding of causative relationships between the investigated metrics is also essential, and it can only be done with thorough investigation of the studied system, and complexity of such exploration in practice may be prohibitive. Moreover, complex ecological relationships among species in the local agroecosystem may result in intricate and implicit effects. All these phenomena may extend beyond the spatial buffer in focus, a boundary that facilitates analysis but narrows analytical perspective.

Our study complements previous analytical efforts to describe sorghum aphid and its natural enemies. In a study that spanned over a similar spatial range, Brewer et al. (11) noted that the correlations between insect interactions (sorghum aphid and natural enemies) and environmental characteristics (landscape features and weather conditions) varied among the S, S GP and N GP regions. In the S region, the relationships between insects were strong but apparently independent from the environmental factors. Influences of environmental factors became apparent in S GP, where parasitoids negatively correlated with percentage of buffer occupied by cropland and lady beetle adults negatively correlated with percentages of buffer occupied by shrubland and by sorghum. In the N GP the significant correlations between insects were limited to predators but activity of parasitoids could be linked with percentages of buffer occupied by sorghum and by shrubland. Elliott et al. (26) found that percentage of buffer occupied by wooded land was significantly correlated with parasitism by L. testaceipes on aphids in winter wheat in Oklahoma, and that both multiple landscape composition and multiple landscape configuration metrics were significantly correlated with parasitism by L. testaceipes. They also found that parasitisism increased with increasing landscape diversity, indicating that L. testaceipes utilizes multiple habitats throughout the year. To the contrary, Elkins et al. (10) found that parasitism decreased with increase of landscape complexity. Compared to the southerly Texas Gulf Coast, the more challenging winter conditions in Oklahoma make availability of suitable winter habitats more challenging, a likely reason for the differences. Differences between these studies and our study likely stem from differences in timing and sampling locations, but also from differences in spatiotemporal scale and resolution.

Timing of field activities, such as data collection, is of high importance. Simulation modeling is one approach that could help mitigate such challenges by aiding management decisions, potentially in near-real-time (20, 21). The regional aphid infestation may be perceived as a northward movement of the “aphid invasion front,” which is limited primarily by sorghum planting dates (45). However, winters are mild in the south and outbreaks may occur locally from remnant aphid populations overwintering on rudimentary plants. Nevertheless, dates of first infestation within season may be surprisingly close across large distances, they can be highly variable within a region, and they may vary significantly annually ( Figure 10 ). Data collections that start too early may result in futile efforts. Late data collections may fail to capture the first infestations. To describe some of the processes, we relied on data series with initial observations of no insects (e.g., vSA), and only those data series could be used for such estimates. Proper timing is essential to in-depth data analysis, but it may also be the most challenging factor in scheduling field activities. Computational modeling can aid field research (14) but ultimately decisions remain with practitioners.

A blend of laboratory studies, common field studies, and field data collection is needed to increase our understanding of causative effects of environmental factors on the dynamics of pest and natural enemy populations. Further investigations of pest and natural enemy interactions and impact on their respective populations need to account for temporal scale (29). A thorough understanding of the dynamics of pests and their natural enemies is critical for AWPM programs and thus for sustainable agriculture. Broader understanding of these dynamics requires holistic approaches to data collection, insect metric selection, and inclusive analysis of environmental factors in future studies. The practical application of the research results to AWPM include adjusting the timing of monitoring for pests geographically based on knowledge of historical infestation patterns, assessing the likelihood of economic infestations based on spatiotemporal features of landscapes and broader scale regions where knowledge of infestation history is applicable to predicting future infestations and their severity. In addition, finer scale attributes of landscape composition and configuration can be used to predict severity of infestations at the scale of individual fields.