Spatio-temporal distribution patterns of Plutella xylostella (Lepidoptera: Plutellidae) in a fine-scale agricultural landscape based on geostatistical analysis

A detailed knowledge on the spatial distribution of pests is crucial for predicting population outbreaks or developing control strategies and sustainable management plans. The diamondback moth, Plutella xylostella, is one of the most destructive pests of cruciferous crops worldwide. Despite the abundant research on the species’s ecology, little is known about the spatio-temporal pattern of P. xylostella in an agricultural landscape. Therefore, in this study, the spatial distribution of P. xylostella was characterized to assess the effect of landscape elements in a fine-scale agricultural landscape by geostatistical analysis. The P. xylostella adults captured by pheromone-baited traps showed a seasonal pattern of population fluctuation from October 2015 to September 2017, with a marked peak in spring, suggesting that mild temperatures, 15–25 °C, are favorable for P. xylostella. Geostatistics (GS) correlograms fitted with spherical and Gaussian models showed an aggregated distribution in 21 of the 47 cases interpolation contour maps. This result highlighted that spatial distribution of P. xylostella was not limited to the Brassica vegetable field, but presence was the highest there. Nevertheless, population aggregations also showed a seasonal variation associated with the growing stage of host plants. GS model analysis showed higher abundances in cruciferous fields than in any other patches of the landscape, indicating a strong host plant dependency. We demonstrate that Brassica vegetables distribution and growth stage, have dominant impacts on the spatial distribution of P. xylostella in a fine-scale landscape. This work clarified the spatio-temporal dynamic and distribution patterns of P. xylostella in an agricultural landscape, and the distribution model developed by geostatistical analysis can provide a scientific basis for precise targeting and localized control of P. xylostella.

Data collection. Forty-six pheromone-lure traps (Enjoy Wing trap, supplied by Zhangzhou Enjoy Agricultural Technology Co., Ltd.) were set up to catch P. xylostella in different patches within the study area, with 15 traps in the 9 patches of cruciferous vegetables, 4 in the vegetable fields inside polytunnels, 6 in the rice field patch, 12 in the patch of pastures, 4 in wastelands, 4 along roads, and 1 in a residential area (Fig. 1b). The chemical compound of the pheromone lure were cis-11-hexadecenyl acetate and trans-11-hexadecenyl acetate absorbed in natural rubber (red) core. The lures hung in the middle of the trap with green roof, 5 cm above a white sticky bottom of the trap. The traps were 30 cm above ground fixed at the top between two bamboo poles inserted into the ground. The distance between the traps ranged from 38 to 930 m. From 30 October 2015 to 17 September 2017, pheromone lures and white sticky plates were replaced fortnightly and the number of captured adult P. xylostella were recorded.
A handheld GPS (Garmin eTrex Legend, Taiwan) was used to set up a grid in the study area and to collate trap position data. The sampling area was divided into landscape elements and the trapping points were visualized using ArcGIS10.2 software (Environmental Systems Research Institute, ESRI 2013).
Meteorological data during the sampling period (from October 2015 to September 2017) were provided by the Fujian Meteorological Service Center 35 . The correlation between adult moth numbers and meteorological data (maximum temperature, minimum temperature, relative humidity and precipitation) was tested using the Pearson's correlation coefficient (P = 0.05).
Geostatistical analysis. Before spatial analysis, moth count data was log-transformed to approximate a normal distribution. The spatial dependence among P. xylostella samples was assessed based on these transformed data using semivariance analysis 28 . The semivariogram analysis was performed with the GS + software (Version 9, Gamma Design Software, Plainwell, MI, USA) for fortnightly counts of P. xylostella, provided that cumulated catches were greater than ten individuals (total 34 cases of 47 cases). Optimal models were fitted with the best fit being measured by the coefficient of determination (R 2 ), residual sum of squares (RSS), range (a) and nugget (C 0 ) 36 . At a certain distance, the semivariance stabilizes at a constant value. This constant semivariance is called the sill (C 0 + C), this distance is the range (a), and the semivariance value at the intercept when the distance is equal to zero is called the nugget effect (C 0 ) 37 . The C 0 /(C 0 + C) ratio (level of spatial dependence, LSD) provides an estimation of the amount of randomness that exists in the data at spaces smaller than the sampling distance 30,38 . The  (Table 1) www.nature.com/scientificreports/ spatial dependence of the semivariogram is considered strong when LSD ≤ 0.25, moderate when 0.25 < LSD ≤ 0.75, and weak when LSD > 0.75 30 . Models obtained from the semivariogram analysis were used to interpolate P. xylostella catches by the means of the inverse distance weight method with the use of the squared values 33 . Spatial analyses were carried out using Surfer Version 14 (Golden software, Golden, CO, USA) with data columns X, Y representing latitude and longitude expressed as Universal Transversal Mercator coordinates, and Z representing the trap counts 28,37 . The obtained interpolation grid was graphically represented using a contour map layered on the base map of the experimental area 31 .
One-way ANOVA by SPSS statistics software was used to test the statistical differences between the yearly catches obtained from different landscape elements. Prior to the analysis, squareroot transformation ( √ x + 1 ) was applied to normalize the distribution. The Tukey-Kramer test (P = 0.05) was used for multiple comparison, upon a significant difference obtained from the ANOVA.

Results
Temporal dynamics. During the sampling period 3543 P. xylostella males were collected in all traps set up in the study area. Based on the 2-year data of trap-captured specimens, we observed an early spring (March to April) peak of P. xylostella population, while summer and winter were less favorable, with a numerical decline of the population in these two seasons. Plutella xylostella population were significantly associated with the average daily minimum temperature (P = 0.002, r = − 0.450) and average daily maximum temperature (P = 0.011, r = − 0.370). There was no significant relationship between P. xylostella population size and relative humidity and average daily precipitation (P = 0.144, r = 0.216 and P = 0.781, r = − 0.042, respectively). In

Spatial distribution.
A total of the 47 semivariograms were calculated from the 2 years of sampling, and 34 mathematical models were successfully developed ( Table 2). Of these, 19 cases were presented as Spherical models, and two cases (on 20 January 2016 and 10 February 2016) were shown as Gaussian models. These 21 population samples displayed strong spatial autocorrelation and aggregation. The remaining semivariograms did not result in an asymptotic model, indicating a random distribution.
The spherical and Gauss models showed small nugget values (0.008-0.168), large sill values (0.1498-1.766), strong spatial heterogeneity, and spatial variance ratios (C 0 /(C 0 + C)) ranging from 0.01 to 0.26 (Fig. 3), which provided evidence for strong spatial autocorrelation. The spatial pattern of P. xylostella populations showed an aggregated distribution, with an estimated range from 82.00 to 317.31 m ( Table 2). The 21 samples in the clumped pattern were collected during growth period (during which theleaf and harvestable vegetative plant parts develop in the cruciferous) vegetables, and comparatively, the samples collected during the crucifers' maturity stage or when no crucifers were cultivated were randomly distributed. Table 1. Seasonal schedule for growing cruciferous vegetables in different patches of the landscape within the study area, from October 2015 to October 2017. "-" represents non-cruciferous vegetables grown at the corresponding growing dates in the patch. www.nature.com/scientificreports/ Contour maps of spatial distribution. Two years contour maps created by the inverse distance squared weighted procedures exhibited a distribution pattern of P. xylostella in agricultural landscape (Figs. 4,5). In January 2016, the P. xylostella population was relatively low, mostly inhabited in the patch of A7. Then P. xylostella population continued to increase from February to March, and spread widely on farms, with a marked peak in April and the main hotspot located in patch A9. After May, the population decreased, with a small hotspot being observed in patch A9, and then the population declined sharply with very low number of individuals trapped (Fig. 4).
In January 2017, P. xylostella adults were first found in the northeast of the experimental area, and then some small hotspots A1, A3, A4 and A7 were observed with the increase in the fields of cruciferous vegetables. In March, large hotspots began to appear in B. oleracea fields (A4, A5, A8 and A9), and this pattern of distribution continued to May. In late May, with the harvest of cruciferous vegetables, only a small number of P. xylostella could be attracted and no hotspots were observed in June (Fig. 5).
Effect of landscape elements on the P. xylostella distribution. Significant differences were found in P. xylostella numbers between the landscape elements in both studied years (in the first year: F 5, 39 = 7.402, P < 0.01; in the second year: F 5, 39 = 9.776, P < 0.01). In the first year, traps positioned in cruciferous vegetable fields captured more P. xylostella adults, when compared with traps in polytunnels, pastures, wastelands or roads. However, the difference was not statistically significant compared to rice fields (Fig. 6A). In the second year, the number of P. xylostella adults caught in cruciferous vegetables was also significantly higher than those in other landscape elements (Fig. 6B). Traps in residential areas did not catch any adults of P. xylostella.

Discussion
This study presents new information on the seasonal fluctuations of P. xylostella in a fine-scale agricultural landscape with different cropping and non-cropping areas, providing valuable contribution to the phenology of this destructive pest. More than 20 generations of P. xylostella can develop per year in south China, and chemical control is the main management strategy against them on cruciferous crops 38 . The characterization of the temporal dynamics and the spatial distribution of P. xylostella in the agricultural landscape provide important information for monitoring P. xylostella and assisting to develop effective pest management strategies targeting this pest. Although for management the presence of females is more important than that of males, and our baited traps mainly caught males, the general population patterns most likely can also be extrapolated to females.
The seasonal population dynamics and population peaks were apparent in the studied 2 years. The spring and autumn population peaks of P. xylostella in our study (Fig. 1) were consistent with previous reports 2, 39,40 . In the southern regions of China (including Fuzhou), low temperatures in January and high temperatures in July and August are not favorable for P. xylostella 38,41,42 , thus number of captured individuals remained low in these months. In fact, the peaks of pheromone trap catches in November and March-April each year (Fig. 2) were well aligned with the largest presence of food crops (Table 1). Unlike other studies 43-45 , we did not find significant relationships between P. xylostella population size and relative humidity or the average daily precipitation. This may be due to the long (fortnightly) sampling interval without heavy rain or long duration rainfall, resulting in no differences in population size. Another reason may be that the boat-type trap used in the experiment may have a rain-shielding effect. www.nature.com/scientificreports/ www.nature.com/scientificreports/ Geostatistical analysis and semivariogram models exhibited spatial dependence in 21 of 47 samples in the agricultural landscape (i.e., spatial aggregation). Overall, the dispersion patterns of P. xylostella were aggregated during the growth periods of their hosts (from March to April 2016 and from January to April 2017) and random during the mature stage of the host plants (from May to July 2016 and from May to June 2017).
Contour maps indicated an aggregation of P. xylostella in the agricultural landscape, mainly synchronized with the availability of food plants in the area (patches of A7, A8 and A9; Figs. 4,5), where cruciferous vegetables were grown. Individuals in some months were also located with low numbers, outside cabbage fields, most likely www.nature.com/scientificreports/ because moths were caught during their host searching flight. This varying response, reported also by other authors, was strongly related to the presence of cabbage 46,47 and the dispersal pattern of P. xylostella population dynamics is associated with the shortage of favorable food 48 . When the crops were in their growth period (from March to April in 2016 and 2017), the ecological environment gradually stabilized and became more suitable for P. xylostella. The populations developed rapidly and stabilized in this period. While, at the mature stage of cruciferous vegetables (from May to July in 2016 and 2017), the deteriorated quality and the harvest of the crops is likely to be resulted in an unfavorable environment for the survival of P. xylostella 49 . Although the numbers www.nature.com/scientificreports/ of adult caught varied among patches, the trend of the population's spatial distribution was highly similar in the two studied years, and the maps indicated that highest densities of P. xylostella were located in the areas of cruciferous vegetables. Therefore, the hot spots seemed to be linked not only to the species of host plants, but also to the growing stage of the plants influences the spatial distribution of P. xylostella population.
In recent years, significant attention has been paid to the impact of agricultural landscape on integrated pest management 50,51 . The observed pattern of P. xylostella distribution in our study increased towards the area of cruciferous vegetable growing, especially the cultivation of Brassica crops. Plutella xylostella captures were highly influenced by cropping systems at the regional level and the spatial trend of dispersion was consistent with the cabbage field 46 . In farmland ecosystems, P. xylostella showes a distinctive spatial distribution pattern among patches, and the layout of host plant patches is one of the drivers that affect this distribution pattern 5 .
Our results and similar studies of temporal dynamics and spatial patterns, as well as those use geostatistical analysis, can provide important information to develop a control measure in agricultural landscape. Cruciferous vegetables planting area is the main occurrence area of P. xylostella. Thus, one of the possible implications of this study for the management of P. xylostella is that a reasonable number of traps can be placed in and around cruciferous vegetable fields at the early growing stages of cruciferous vegetables to catch males to reduce mating and thus decrease population. Traps can be placed early in crucifers' growth season to prevent further damage and this way, the use of pesticides can be minimized.
This study characterizes the temporal dynamics and the spatial distribution of P. xylostella in an agricultural landscape, and demonstrates that host distribution and growth stage may have a great impact on the spatial distribution of P. xylostella population. The results advance our understanding of temporal and spatial distribution of the P. xylostella population on a diversified farm in subtropical region, and provide knowledge of using pheromone baited traps and geostatistical analysis method as tools for monitoring and forecasting of the population dynamics and implementing the program of integrated pest management 52,53 . www.nature.com/scientificreports/