A methodological framework integrating habitat suitability and landscape connectivity to identify optimal regions for insecticide application: A case study in Tongzhou, China

2.1 Study area

The study area was the entire region of Tongzhou District (39°36′-40°02′N, 116°32′-116°56′E, area 912.34 km², altitude 8.2–27.6 m) of Beijing, China (Fig. 1). The mean annual temperature record in the area is 11.3 °C, while the average annual precipitation is 620 mm, and the frost-free period is about 190 days per year. The outbreaks of S. cinerearia in 2013 had critically damaged the urban green trees in the study area (Tang et al., 2014). Some insecticides were subsequently adopted for pest control, but based on our fieldwork findings, the population of S. cinerearia persisted with periodic outbreaks in the area (Fig. 1). The adverse impact in this sprawling city caused by the S. cinerearia was a heavy financial burden to the local government. A new city is in the construction and development phase here, as the Chinese capital sub-center (Wanghe et al., 2019). Therefore, it is essential to efficiently control pest outbreaks in the future, which is critical to the new city's development.

Close

Fig. 1

Locations of our study area and the quadrats selected randomly, as well as S. cinerearia occurrence points surveyed.

Export to PPT

2.2 Sampling

We randomly set 24 quadrats (3 km × 3 km) in the urban and rural areas of Tongzhou District (Fig. 1). To achieve random placement of those quadrats, we first divided the study areas into 101 grids. Secondly, 1,000 random points were generated on the grids. These random points were developed by the “Create Random Points” tool in the system toolbox of ArcGIS 10.2.2 software. After repeating the second step 10 times, the top 24 quadrats with the most randomly generated points were selected as the sampling quadrats. Fieldwork was conducted from June to August 2018. We searched and collected the target specimens S. cinerearia using the “Sweep Sampling Method” (Gadagkar et al., 1990) in each quadrat, in which we systematically swept all the ground level vegetation (<2.5 m). The occurrence points were recorded with the help of a Global Position System (GPS) with the World Geodetic System 1984 (WGS 84) coordinate system. Five hundred and seventeen individuals of S. cinerearia were collected in the fieldwork (Fig. 1 and Table S1). The collected voucher specimens were deposited in the Natural Museum of Beijing Forestry University, Beijing, China.

2.3 The scope, assumption, and framework of the modeling

Fig. 2 illustrates the modeling framework of this study. A combine model, integrating MaxEnt and the circuit theory was applied. The scope of this combination was to predict the dispersal ability of the target pest relating to the original habitat suitability without any control measurements and the changes of habitat suitability after some control measures were conducted. This study was based on certain assumptions. First, we assumed that the results of MaxEnt represent the present habitat suitability before pest control. Second, we assumed that the control measures change the current habitat suitability of the pest. By comparing the SDM-circuit values among those different scenarios (explained in Section 2.8), the pest management effectiveness among the scenarios was identified.

Close

Fig. 2

Modeling framework of the study.

Export to PPT

2.4 Environmental variables

Eight environmental variables (Table 1) were selected to model habitat suitability based on current knowledge of the eco-physiological characteristics of targeted species (Jiang et al., 2018; Lozier and Mills, 2011), especially in the urban areas (Godefroid et al., 2020). Additionally, our study area landscape is plain and relatively small, without apparent altitude and climate differences, so we did not select altitude and climate as the environmental variables.

The environmental variables were extracted from the forest resource inventory database and the land-use dataset (Table 1). The forest resource inventory database in 2015 was a polygon shapefile provided by the Beijing Tongzhou Forestry Bureau. The locations of S. japonicum (the primary host of S. cinerearia), the percentage coverages of forests, and shrubs and meadows in a cell were extracted from this database as raster maps at the resolution of 10 m × 10 m. The land-use dataset in 2015 was interpreted from a series of Gaofen-2 satellite images (resolution = 0.80 m) (Wanghe et al., 2020b; Wanghe et al., 2020a; Wanghe et al., 2019). Subsequently, four environmental variables, the distance from a raster to the nearest building, road, cropland, and water, were derived from the land-use dataset using the Euclidean Distance System Toolboxes of ArcGIS 10.2 software (Environmental Systems Research Institute Inc., Redlands, USA).

2.5 Species distribution model (SDM)

2.6 Connectivity model

The connectivity model was built using Circuitscape version 4.0.5 (McRae and Shah, 2014), a software designed to model habitat connectivity using the electrical circuit theory (McRae et al., 2008). The input was a conductance map from the habitat suitability map estimated by the MaxEnt model. The standard conductance of each cell was defined by rescaling the habitat suitability map between 0 (minimum suitability) to 1 (maximum suitability). Circuitscape ran in advanced mode to allow for the use of arbitrary current sources and grounds. The cells at the host location were set as current sources of one ampere. The cells at the perimeter of the study areas represented the ground of one ohm, following Falaschi et al., (2018). Thus, the model could predict the possible spread of S. cinerearia, considering all possible directions in one fell swoop (Cowley et al., 2015). The current densities of the cells estimated by this connectivity model were defined as the SDM-circuit values. The higher SDM-circuit indicated cells with more necessary net passage probabilities for random movement, moving from one raster to another (McRae et al., 2008). The movement boundaries and the “pinch points” among populations were identified by Kernel Density analyses (Silverman, 2018) based on the SDM-circuit values, using the Kernel Density Tool (Silverman, 2018) of ArcMap 10.2 software (ESRI, 2011).

2.7 Scenario developments

The SDM and connectivity model identified the regions of suitable habitat and corridors of S. cinerearia. However, widely spraying insecticide at all those regions was costly and would further impose some unfavorable impacts on the ecosystem, such as damaging the pests’ natural enemies Therefore, a cost-effective solution was to detect the significant and economical regions/areas for chemical control. To determine that cost-effective solution, we developed five scenarios and compared the pest management effectiveness of those scenarios.

Firstly, we defined that part A was the host’s location and part B was the location of other tree species except for the host. Secondly, part B was divided into five sub-parts based on the SDM-circuit values ranked by Jenks natural breaks method (Chen et al., 2013): top 20% (sub-part B20), top 40% (sub-part B40), top 60% (sub-part B60), top 80% (sub-part B80), and top 100% (sub-part B100). Those natural break classifications were in terms of natural groupings inherent to their SDM-circuit values, whose class breaks were identified that the best group was similarly valued and that maximizes the differences between classes (ESRI, 2011). Thirdly, the regions for insecticides of scenario Bn were defined as the union of part A and sub-part Bn (n = 20, 40, 60, 80, and 100, respectively) (Fig. S2). Based on the second and third assumptions described in section 2.3, the habitat suitability of the regions for insecticides was modified to a value approaching zero. The habitat suitability of the regions without insecticides remained unchanged. Finally, the five scenarios were independently inputted to the connectivity model.

2.8 Scenario comparisons

To examine pest management performance/effectiveness, those five scenarios were compared with the connectivity model's original result (without the assumptive scenario). The “fuzzy numerical” algorithm (Jiang and Chen, 2003) (Eqs. (1) and (2)) was implemented, allowing a cell-to-cell comparison between two raster maps. In Eqs. (1) and (2), map A was the initial result of the connectivity model, and map B was one of the scenarios. The raster maps A and B were in the same location with the same rows and columns. The a_i and b_i were one cell on the maps A and B in the same location, and j was the total number of cells. Consequently, S_(ai,bi) depicted a new map highlighting the divergences between maps A and B. S_(A,B) referred to the average differentiation between the entire maps of A and B, ranging from zero to one: one meaning two completely different maps, and zero representing an identical pair. Therefore, a higher S_(A,B) indicated a better pest management performance of a sub-scenario than the original result (Visser and De Nijs, 2006). Those calculations were programmed by Python 2.2 using the ArcPy site package (ESRI, 2011).

Ultimately, a regression model between the two variables, S _{(A, B)} and the control area of sub-part Bn, was conducted by the basic Trendline R package. The computer device executing all analyses was a Mac Pro (2019) graphic workstation with 3.5 GHz Intel Xeon W CPU, 32 GB RAM, and Windows 10 Professional operating system.

3.1 S. cinerearia habitat suitability

The AUC value (0.895 ± 0.018, mean ± standard deviation) indicated that the results of SDM had a receivable result (Fig. S1). The high AUC value (compared to the possible maximum 1.0) indicated that the quality of habitat suitability map derived by the MaxEnt models was high and the sampling quadrats were sufficient (Phillips et al., 2006). The relative importance of environmental variables was assessed by the Jack-Knife test (Fig. 3), which implied that forests and hosts appeared to have the most useful statistical information. The contributions of environmental variables are in Table S3. The habitat suitability map of S. cinerearia (Fig. 4) illustrated that the average suitability in the entire study was 0.1231 (SD = 0.1791). Comparably, the mean suitability in the location with S. japonicum was 0.5315 (SD = 0.1468), while the average suitability in the regions without S. japonicum was 0.1101 (SD = 0.1640). Our results indicated that the places of S. japonicum were ordinarily more suitable for S. cinerearia than other land-use types.

Close

Fig. 3

Relative importance of environmental variables (training gains) in MaxEnt models by the Jack-knife test. The names and descriptions of those environmental variables see Table 1.

Export to PPT

Close

Fig. 4

Habitat suitability map for S. cinerearia.

Export to PPT

We compared the habitat suitability among the top ten widely planted tree species in Tongzhou District (Fig. 5). S. japonicum, the host of S. cinerearia, provided the most suitable habitat for this pest insect. In other words, S. japonicum was the most vulnerable species to S. cinerearia, compared with the other nine tree species, which was also widely planted in the study area. The genus Populus (P. tomentosa and P. canadensis) presented the least suitable habitat to S. cinerearia compared to the other eight species.

Close

Fig. 5

Comparison of habitat suitability providing to S. cinerearia among the top ten widely planned urban greening tree species.

Export to PPT

3.2 S. cinerearia habitat connectivity

The map of current densities/SDM-circuit values (Fig. 6) indicated the passage probabilities of the target pest. A cell with a high current density value implied that it would provide great landscape connectivity to S. cinerearia (McRae et al., 2008). Our results showed that the mean current density of the host tree species (S. japonicum) was 39.268 (SD = 30.456), while the average current density of other tree species was 25.973 (SD = 25.892).

Close

Fig. 6

Map of the current densities/SDM-circuit values of S. cinerearia. Higher current densities indicate cells with higher net passage probabilities for random walkers moving from one patch to the other. The map highlights “pinch points”, which are in some linear narrow pathways but providing critical habitat connections between or within two populations.

Export to PPT

Six populations with high current densities were distinguished (Fig. 6). Our results recognized some “pinch points” (Fig. 6). The “pinch points” were narrow linear pathways of S. cinerearia between or within those populations.

3.3 Scenarios comparisons

The area of part A is 27.03 km², and the areas of sub-parts B20, B40, B60, B80, and B100 (the red patches in Fig. S2) are 3.22, 20.55, 65.47, 137.90, and 253.28 km², respectively. Fig. 7 describes the differences in management performances among these scenarios. While the effectiveness of pest management on the host was similar from scenarios B20 to B100, by increasing the area of chemical control, the passage probabilities of S. cinerearia in the other tree species were decreasing.

Close

Fig. 7

The management performances among the scenarios. The y-axis is the value of S(A,B) described in Equation (2). In the x-axis, “A, Bn” means the scenario Bn vs. the original connectivity model.

Export to PPT

Fig. 8 displays the regression model results between two variables, S _{(A, B)} and A (the area of the sub-part Bn). The most suitable model was the polynomial regression, whose R² = 0.9937. This result indicated that management effectiveness increased while enlarging the regions of chemical control when A < 200 km². The marginal benefit of scenario B20 was the maximum among all scenarios. Therefore, scenario B20 was the most efficient one to conduct chemical prevention, in which the areas of chemical application were increased only by 3.22 km² (11.9%). At the same time, the connectivity of the other trees species was decreased by 27.7%.

Close

Fig. 8

Result of the regression model. The y-axis is the value of S(A,B) described in Equation (2). The x-axis means the area of sub-part Bn (km²). The grey shaded areas are 95% confidence intervals.

Export to PPT

4.1 The implications for pest management

The list of forests being damaged by pests is growing worldwide (Aukema et al., 2010; Loo, 2008; Ramsfield et al., 2016; Santini et al., 2013). Compared with natural forests, urban trees are more vulnerable to pest incursions (Paap et al., 2017). The limited tree diversity in the urban ecosystem further predisposes urban forests to potentially devastating pest outbreaks (Laćan and McBride, 2008; Paap et al., 2017). For example, ash is a widely planted city street tree in the United States, consisting primarily of a limited number of cultivars of white and green ash (Fraxinus americana and Fraxinus pennsylvanica) (MacFarlane and Meyer, 2005). This situation resulted in low genetic diversity and enhanced the risk of pest incursions to urban plants. As a result, extensive damage occurred in urban greening of the northeastern United States with the accidental incursions of the emerald ash borer (Agrilus planipennis) (Kovacs et al., 2010). In the current study, the urban forests were composed of dozens of different tree species (Fig. 5). However, the periodic outbreaks of S. cinerearia have been causing significant disruption to the urban forests, although some chemical control focused on the host. Similarly, Simulium chutteri broke out periodically in South Africa, even though some integrated control measures have been adopted (Rivers-Moore et al., 2008). The issues remained as neither increasing the diversity of urban forest nor host-specific pest management measures may completely solve periodic pest outbreaks, which could be explained by the theory of island biogeography and landscape ecology (Simberloff, 1974; Zimmerman and Bierregaard, 1986). Our study focused on an area of the urban ecosystem at the local scale, where the urban trees were always isolated by some artificial surfaces and infrastructures, such as road networks and buildings (Wanghe et al., 2019). Consequently, each isolated land unit of urban greenery (e.g., an urban park) could be regarded as an analogous island (Zimmerman and Bierregaard, 1986). Because of the homogenous environment at the local scale, these widely planted urban greening trees (Fig. 5) in each island likely occupied similar niches (Shushan et al., 2016), whose community structure would adapt itself to a dynamic equilibrium among the parasitoid and host, as well as their habitat (Zimmerman and Bierregaard, 1986).

Our results highlighted that the targeted insect's connectivity and migration should be considered a significant component within pest management. In many cases (Dodds and Orwig, 2011; Haack et al., 2010; Paap et al., 2017; Santini et al., 2013; Smith et al., 2007), some clear evidence proposed that after the arrival of these pests into urban areas, their subsequent spread into natural or planted forest landscapes was through multiple means and migratory pathways. For example, the invasive pathways of Asian long-horned beetles contained more than 15 families of trees, which remained in the urban settings until the 2008 outbreak in the natural forest in Massachusetts, the United States (Dodds and Orwig, 2011; Haack et al., 2010). In the current study, although S. japonicum was the primary host and provided the most suitable habitat to S. cinerearia, other tree species might still provide multiple niches to the moth (Zhang et al., 2013). The current study identified 42.9% of the occurrence points near but not located in S. japonicum. This result indicated that the imago of S. cinerearia could migrate and select multiple niches (tree species) other than its primary host (Fig. 5). In addition, the result of the connectivity model (Fig. 6) identified that not only the host tree species of S. cinerearia helped the pest to migrate, but other tree species also served the same purpose. We built and tested different scenario comparisons, suggesting that there would be multiple migratory pathways of S. cinerearia if the most suitable habitat (i.e., the primary host) had been removed. Those representations possibly supported the hypothesis of insects’ multiple niches because of their migration and adaptation abilities.

We presumed that decreasing the pests’ connectivity at some momentous pathways (e.g., the pinch points in Fig. 6) might sufficiently alleviate the probability of periodic outbreaks and consequential spread. S. japonicum was the most vulnerable to the incursion of S. cinerearia, compared with the other widely planted nine tree species found in the area. A study suggested that S. japonicum was not the only food source of S. cinerearia (Tang et al., 2014). The pest prevalence in other plant species indicated that it could survive at different habitats in an urban ecosystem. Therefore, spraying pesticides only in its favorable habitat (S. japonicum) would not eliminate the outbreak, as some survivors with resistance genes could move to the other niches through some high connectivity corridors (Feng et al., 2019; Tang et al., 2019). As a result, there would always be a risk of a secondary pest outbreak. Those representations might explain the reasons for periodic pest outbreaks in our study area, even if some chemical control had been conducted on the most vulnerable tree (i.e., S. japonicum). The other habitat providing critical landscape connectivity, such as sub-part B20 (Fig. S2a), was also the potential and efficient pest management region (Fig. 9). Our results showed that the genus Populus (P. tomentosa and P. canadensis) presented the most miniature habitat suitability. In other words, it was more resistant to S. cinerearia than the other eight species. Therefore, planting P. tomentosa and P. canadensis at “pinch points” would probably halt the outbreak and migration of S. cinerearia. Moreover, some anthropogenic barriers should be built in the “pinch points” to decrease the connectivity. For example, although the Sixth Ring Road isolated the Populations B and C (Fig. 6), some “pinch points” between the two populations still acted as corridors for S. cinerearia. Thus, cutting off those “pinch points” could minimize the probability of migration between Populations B and C.

Close

Fig. 9

Flow chart for the combinational modeling between habitat suitability and landscape connectivity to identify the potential regions for pest management.

Export to PPT

4.2 The availability and limitation of this combinational model for pest management

Many studies (Fischbein et al., 2019; Jiang et al., 2018; Müller et al., 2019) recognized potential habitat aimed at pest management by SDM without analyzing landscape connectivity. Thus, some significant pathways of the target species could not be identified (Cowley et al., 2015). For example, the potential dispersal of the Asian long-horned beetle was predicted using (but not combining) SDM and the circuit theory (Elmes et al., 2019), but the relationship between habitat suitability and landscape connectivity was ignored. However, few studies used this associated method on pest insect management (Elmes et al., 2019). Nonetheless, combining these two methods could identify a more practical pest management strategy without increasing the fieldwork labor and data input. In the current study, we illustrated a straightforward working procedure of this combinational modeling (Figs. 2 and 9). This integrated method has also been empirically applied to predict the expansion of species invasion (Elmes et al., 2019; Falaschi et al., 2018), distribution range shift of endangered mammals (Puddu and Maiorano, 2016; Zhao et al., 2019), the process of gene flow of a disease vector (Medley et al., 2015), and phylogeographic analysis (Dupas et al., 2014). These case studies identified that this combination model was available to different ecological processes.

Geographic information system (GIS) has grown in popularity in recent years, and it has realized the integration of various types of data and algorithms, providing sophisticated tools for analyzing, editing, mapping, and performing spatial queries on data (Günen, 2021a). Meanwhile, some integrated GIS-based spatial models have improved the decision support processes in many research fields, such as energy planning (Günen, 2021b), urban greening planning (Pelorosso et al., 2014), endangered animal conservation (Ahmad et al., 2020b; Ahmad et al., 2020a), and even helping to clarify the extent and impact of the COVID-19 pandemic (Franch-Pardo et al., 2020). In this paper, we proposed an integrated GIS-based modeling framework to improve research and applications focused on pest management. Three advantages were concluded as follows. Firstly, the researcher easily acquired the input data, only including the occurrence points (Fig. 1) and eco-physiological environment layers (Table 1) of the targeted pest. Secondly, the results explicitly determined the relationship between the pest and the environment of its habitat (Table S3), as well as the potential regions for pest management. These results mapped the pests' vulnerable regions for forestry managers (e.g., the suitable habitat in Fig. 4 and the pinch points and corridors in Fig. 6). Thirdly, the management effectiveness (Figs. 7 and 8) of different scenarios were reasonably quantified. Thus, customized solutions could be proposed in various pest situations (scenario B20 in Fig. 9).

We concluded that our model has great potential in designing pest management policies and strategies in other similar situations around the world. Nonetheless, this approach had some limitations, which could be eliminated in further studies. For example, the spatial autocorrelation and pseudo-replicates may decrease the results’ confidence (e.g., AUC value). Because our occurrence points were only obtained by our fieldwork at the local scale, whose spatial autocorrelation and pseudo-replicates were likely ignorable (Guisan et al., 2017), we did not examine the spatial autocorrelation in model residuals. Instead, if the occurrence points were from different sources at a large scale, the spatial autocorrelation should be detected by the Moran’s I test on the generalized linear models (Guisan et al., 2017). Additionally, our results lacked validation. In a future study, we would experimentally implement scenario B20 to validate this model.