Land use type and Thaumetopoea processionea: a
spatial analysis of the oak processionary moth
Pleun Aarts
Supervised by: Olga van de Veer, Emiel van Loon and Peter Roessingh
July 1, 2016
Institute of Biodiversity and Ecosystem Dynamics (IBED), FNWI, University of
Amsterdam. Science Park 904, 1098 XH, Amsterdam, the Netherlands
Phone: +316 47923339; e-mail: pleun.aarts@gmail.com
Abstract
Over the last few decades, the geographic distribution of the Oak Processionary moth, Thaumetopoea
Processionea has shifted northwards. Global warming has long been though to be the greatest driver
of the northern expansion. However, more recent research suggest that habitat change could also be a driver of the recent shift northwards. T. processionea is in particular of interest because of the harmful hairs it releases. Control of the species is labour intensive and thus expensive, and a FAB approach will be tested over the next years where natural enemies will be attracted on sites where the caterpillar is found. With data from the Province of Gelderland, this research investigates whether there is a relationship between land use types and the presence of T. processionea. It is expected that a more complex, i.e. natural landscape has more natural enemies to attack T. processionea and therefore infection of oak trees by the T. processionea is less likely. A distinction between natural and non-natural landscapes could not be made because no significant effects of non-natural landscapes could be found.
Introduction
Due to their short generation times and high reproductive rates, insects respond to seasonal and inter-annual climatic variations quickly. Long-term climatic changes such as the current global warming affect the geographic distribution of insects fundamentally, for example the shift of heat-loving species to the north (Wagenoff & Veit, 2011).
One order of insects of which the geographical range has been modified greatly over the last decades are the Lepidoptera. The impact of climatic change on the geographical range of this order has been well documented. Particularly in North-Western Europe, where different species have been monitored since the mid 1700’s (Groenen & Meurisse, 2012).
One well-documented species is the Thaumetopoea processionea L. (Lep., Notodentidae), the so-called oak processionary moth (T. processionea). T. processionea is an univoltine species which is widely distributed throughout Europe. Although the species originates from warm and sunny sites in Southern and Central Europe, the moth is now found in more Northern regions as well (Figure 1). Highly unpredictable population dynamics are common over the whole range, however recent data has suggested that in the Northern parts of the range, frequency and intensity of T. processionea outbreaks has increased (Meurisse et al 2012). Severe population dynamic fluctuations occur more likely here, at the edges of the range, where conditions unfavorable for the species occur more often (Wagenoff & Veit 2011).
Figure 1. The distribution of T. processionea, grey shaded area (Meurisse et al. 2012).
The larvae of T. processionea feed on different oak species. They have mostly been observed on single oak trees in open landscapes or in oak dominated forest stands. The caterpillars live in large groups, moving in procession to the crown of the tree, on which they feed (Wagenoff & Veit 2011). These dense populations vary in size from 10.000 - 100 000 individuals on a single tree, which they defoliate. Scientists have not yet discovered all species regarded as natural enemies of T.
processionea. However Moraal (2014) and Hellingman & Mulder (2012) have discovered that at least
species of tachinidae, wasps, beetles, bugs, hoverflies, lacewings and ants predate and parasite the moth. During outbreaks, T. processionea is highly threatening to humans and domestic animals through the release of urticating hairs into the environment (Groenen & Meurisse 2012), which penetrate the skin and causes severe allergic-like reactions. The older larvae can carry up to half a
million of these fine hairs, and one of the major issues is that they can be blown over considerable distances by wind. These hairs can be active for more than 10 years and therefore pose a long-term threat to human health (Rahlenbeck & Uthikal 2015).
Considered as an emerging problem in North-Western Europe, multiple governmental bodies use different methods to combat these insects. Until now, these methods consist of mechanical removal, burning, chemical pesticides and more recently, nematode spraying. These solutions are not only costly because of their high labor-intensity but also threaten other, preferable, non-target and protected insects (Fransen et al. 2006).
In the Netherlands, T. processionea populations have gone from an extinct status in 1987 to
populations of several billions in 1996. This severe increase has often been related to climate change, although Groene & Meurisse (2012) suggests that the effect of climate change is limited because historical distribution suggests recolonization instead of expansion. The range expansion pattern described could be mostly the result of increasing abundance in particularly suitable habitats in the north of Belgium and south of the Netherlands, where oaks are largely planted in open conditions that are very favourable for the moth (Moraal et al., 2004).
The habitat change factor is related to the hypothesis of ‘natural enemies’. The hypothesis suggests that herbivorous predators and parasitoids are more diverse and abundant in plant rich communities than in monocultures because of the supply of shelter and additional rich food(Castagnerol et al 2014).
The natural enemies theory is often used as basis for the use of Functional Agro Biodiversity (FAB). Hereby beneficial insects are stimulated using flowery fields which subsequently attack pests in crops (aphids, caterpillars), and thus preventing damage to the crops (van Rijn & Wäckers 2007).
In 2017, a project will start in the province of Gelderland, called ‘Bloemrijke wegbermen in
Gelderland’. The method of FAB will be used to attract natural enemies to control the T. processionea outbreaks and to stimulate biodiversity in a monocultural area. Prior to that, the Vlinderstichting and the Province of Gelderland wanted to know whether with existing data the distribution of T.
processionea could already be linked to the expected presence of natural enemies. The aim of this
study is therefore to use existing data to investigate if this relation can be found.
Although specific research on FAB for T. processionea control has not yet been conducted, the effects of landscape complexity on natural enemies and pests have been explored across a range of cropping systems and study regions. Within these studies, the term ‘landscape complexity’ was defined in many different ways. Most commonly, the amount of natural or non-crop habitat in the landscape or the diversity of the habitats around the farm were referred to(Chaplin-Kramer et al., 2011).
Earlier research has found that in many systems, positive relationships between landscape complexity and rates of parasitism or predation exist. However, the literature on actual pest responses to
landscape complexity is much less conclusive (Chaplin-Kramer et al., 2011).
This raises the question if landscape complexity could explain the rapid spreading of T. processionea in the Netherlands. Over the past decades, a great deal of oak trees have been planted in the
Netherlands. In the province of Gelderland, 72% of the 150000 trees on the roadside are oak trees (van de Veer, personal correspondence). Consequently, a monoculture arises. Also, the surrounding habitat with monocultural agriculture or highly cultivated urban areas contribute to less variety.
Huigens et al. (2015) expect that this emerging monocultural landscape is home to a minor amount of natural enemies and thus may offer a suitable habitat for T. processionea.
The aim of the study is therefore to investigate the correlation between landscape complexity and T.
processionea outbreaks along roadsides in the province of Gelderland. The complexity of the
landscape around trees will be determined with and a relationship with the presence of T.
processionea will be tested. In addition, tree characteristics can be used to investigate the causes of T. processionea outbreaks in more detail.
It is expected that T. processionea are more likely to occur in urban and mono-cultural landscapes such as agricultural fields because of the absence of vegetation that is providing energy for the natural enemies of T. processionea.
Materials & Methods
Study area
Study site selection was limited by data available in databases the Province of Gelderland.
Consequently, N-roads were selected as study site, since these roads are property of the province and therefore the amount of information on these sites was most abundant. Also, only roads where T.
processionea could pose a threat to humans were selected since the caterpillar observations were
conducted along these roads.
Caterpillar observations
In the caterpillar stadium (4th/5th) observations were conducted by the Province. When caterpillars were present, location of the infected tree was specified to road, hectometer post and side of the road where tree was situated. Besides, a great part of the locations of the infected trees where specified as a transect in which a certain amount of trees were infected. These observations were not used because it was not clear enough which trees were infected and which were not.
LGN7 data
Alterra (Hazeu et al., 2014) has recorded and visualized the land use types in the Netherlands into a rasterfile with a 25*25m resolution, which shows 39 different Dutch land use types for the year 2012. These land use types combine features of the natural landscape such as vegetation types and
features of the ‘human’ landscape such as different types of crops that are planted, rate of urbanisation and infrastructure.
Tree database
Recordings of every tree that is in possession of the Province are available through the ArcMap directory of the province. This also includes tree height and the year in which the tree was planted. This map is actualized by the Province every six weeks. Observations are made every year, were approximately ⅓ of the trees is examined.
Distribution Mapping
The tree database is used to create a shapefile of all the oak trees of the province (selection query: soort=eik). This generates a map of approximately 48000 oak trees. The caterpillar observations were imported into ArcGIS. Also, a map ‘kilometrering’ from the province, that contained all the locations of hectometer posts, was added. Next, the locations of infected trees were displayed by route events using the ‘kilometrering’. This way, the observations were spatially displayed by ‘kilometrering’ on the map. However, this generates observations situated at hectometer posts, a few meters from the actual infected tree(Figure 2).Therefore, to obtain information about the more precise location, age and height of the (un)infected trees, a selection of infected trees was made by hand from the ‘eiken’ shapefile. The trees that were closest to the displayed point from the observations were selected. If this was not clear enough, information on the side of the road where the observation was situated was provided within the attribute table of the observations.
Figure 2, mapped caterpillar observations
Figure 3: All selected infected trees in the Province of Gelderland
However, as mentioned a considerable part of the observations was specified to a transect in which multiple trees were infected. Due to the uncertainty of this data, these trees were not selected. Also, a great part of the observations was missing due to an error in the data storage of the province, which caused the area around ‘Planken-Wambuis’ to be missing. 440 ‘infected’ trees remained (Figure 3). Then, this selection was exported from ArcGIS, and by hand the 440 infected trees were labelled as infected into the large tree database, based on the ID-number of the infected trees (Figure 3). Then all the roads that were not observed according to the province, i.e. roads that were not part of the
stocktaking scheme’s, were removed. Subsequently, all trees that occurred in one of the transect-observations were taken out, since it was uncertain whether these trees were infected or not. The remaining oak-trees were labelled as uninfected. As a result, a dataset of all oak trees that were
observed and surely infected or uninfected was produced. With R (R Core Team 2016), 500
uninfected trees were selected randomly from this dataset added to the infected tree’s table. Now this dataset of 970 trees was loaded this into ArcMap again. To obtain information about the surrounding land use types, the location of the trees was needed. Therefore, the dataset was joined with the tree database, based on the ID-number of the trees. Next, a buffer of 200m around the 970 trees was intersected with the LGN7 layer. This way, a table was obtained with information on the tree characteristics and surrounding land use types. Subsequently, table aggregation was done by the Xtools package to sum all the area’s that occurred multiple times within one buffer. Next, aggregation was done in R to aggregate land use types per individual tree, instead of having a separate row for every land-use type per tree. This way, a table with all trees and their surrounding land use types, height and age was produced(Appendix B).
Statistical analysis
This study tests whether the probability of getting infected by T. processionea differs according to proportions of surrounding land use types. A logistic regression model was chosen because it is a special type of non-linear regression developed for a binary response variable, when Y is either 0 or 1, int this case respectively No or Yes. With logistic regression the following equation is fit to the data: Log-odds(absence-presence) = intercept + coefficient(predictor)(Whitlock & Schutler 2009). This means that each one-unit increase in a predictor (e.g. a change from 100 to 101 hectare Forest) will change the log odds by the value of the coefficient. The logistic equation can then be used to estimate the probability of an event for each case. Consequently, these probabilities can be used to assign each case to a certain binary class variable, in this case if the tree is infected or not. Cases will be assigned to the ‘Yes’ class if their probability of being in that class is higher than a to be
determined threshold. The threshold is kept at default, 0.5, for this research because the proportion of ‘Yes’ and ‘No’ in the used dataset is almost 50/50 (Fielding, 2007).
First, the response + predictors are fitted into a stepwise logistic regression where the best fitted logistic regression model is found using the Akaike Information Criterion (AIC). The output model consists of all the predictors that are used for the prediction of odds to be infected. To test for accuracy of this model the area under the ROC curve (AUC) of this model is calculated and the decline in deviance is evaluated. It is expected that as the predictors are added, deviance will decline. Significance of deviance decline is tested using a Chi-square test. If p<0.05 there is strong evidence that adding the predictors has significantly reduced the unexplained variation (Fielding, 2007). The receiver operating characteristic (ROC) curve is constructed with the package ‘ROCR’ in R (Sing et al. 2005). The plot is a threshold-independent measure for the performance of a model that assigns cases into binary classes. ROC curves have long been used in the analysis of medical systems and more recently in testing the accuracy of ecological presence/absence models. ROC curves show the trade-off between the ability to identify correctly true positives and the classification of negative cases as positive. In addition, the area under the ROC curve, the AUC, is regularly used as an performance measurement. If this value is 0.5 the scores for two groups do not differ, while a score of 1.0 indicates that the group scores do not overlap. In other words, a value of 0.5 indicates that the model is
selecting randomly, just as a coin toss (Fielding 2007).
There are several methods evaluating the significance of the AUC value. However, the best method has not been found yet and due to time restrictions this could not be evaluated for this research. One way is to split the data into a 75% training set and a 25% testing set and see if these values are somewhat similar. Then again, it is not clear how to evaluate the difference between these values.
Results
Model output
Appendix A summarizes the final model and the effects of different predictors on infection. In the model, there are six significant predictors; namely natural grasslands, agricultural grass’, forest in primarily developed area, grass in primarily/secondarily developed area and ‘Other crops’. As seen in Figure 4, the probability of infection appears to decline as ‘Forest in primarily developed area’ increases. The coefficients for the other significant predictors are positive, suggesting that the probability that a tree is infected is greater if these areas are more present.
Figure 4. Predictions of infection over different land use types
Accuracy of the model
The deviance for the null model is 1331,7, which reduces to 1197,3 when all of the predictors are added. The reduction in the deviance (134,4) is analysed with a Chi-square test, which gave a significant p-value (p=2.2e-16, α=0.05), suggesting that the predictors significantly reduce the
a fair amount of wrong-predicted cases is discovered. However, prediction performance is further measured by the AUC, the area under the ROC-curve (0.706, Figure 5). This value suggests that the model predicts infection of trees better than random, which would in that case have a value of 0.5. To test for accuracy, AUC values were 0.717 for the training set and 0.681 for the testing set, which suggests that the testing set could be explained by the training set, but some difference remains. The reduction of deviance is significant for both the training set and the testing set as well.
Observation
Prediction Absent Present
Absent 389 192
Present 152 237
Table 1. Confusion table for regression model
Conclusion & Discussion
A model was produced using stepwise regression, to test whether the probability of getting infected by T. processionea differs according to proportions of surrounding land use types. According to this model, trees surrounded by grassy areas are more likely to get infected with T. processionea, while trees surrounded by forest areas are less likely to get infected by T. processionea. More detailed, the land use types that increase the odds of infection are ‘Grass in primarily/secondarily built area’ which includes all grasslands in urban areas (primarily) and semi-urban areas (secondarily) such as
residential, retail, business and sporting environments (Hazeu et al., 2014). The presence of T.
processionea in more Grassy/urban areas could be explained by the fact that open landscapes in cities
and roadside plantings are usually warmer than forests (Moraal, 2014).
There is also a positive relation between T. processionea and agricultural grasslands or ‘other crops’, which includes respectively ‘Agricultural land with grass used for agricultural production’ and ‘agricultural land with crops that do not fall within the individually classified crops such as potatoes and beetroot’ this class includes ‘Horticultural crops, cabbage crops, hemp, rapeseed, etc.’ (Hazeu et al., 2014). Typical for these land use types is the monocultural, non-natural environment. The change to a monocultural landscape together with high agrochemical input in crop fields is believed to be the primary cause for the rapid decrease of biodiversity in these landscapes (Bianchi et al., 2006).
Another land use type which significantly influenced the presence of T. processionea was the land use type ‘Natural grasslands’. Hazeu et al., (2014) specify these areas as ‘natural grasslands that are extensively managed’. The ‘natural’ aspect of the class suggests that this habitat is more suitable for natural enemies and thus the preference rejects the null hypothesis. However, since ‘Natural
grasslands’ are extensively managed in this case the ‘Natural’ aspect and abundance of suitable vegetation should be reconsidered.
The only land use type that was negatively correlated with the presence of T. processionea was ‘Forest in primarily developed area’. This class is defined as ‘Forests in primarily built area’. No distinction was made regarding species. The negative relationship could be explained by the results of Jactel & Brockenoff (2007), who have tested the relationship between tree species diversity and forest insect herbivory of the pine processionary moth. They found that trees in mixed stands consistently experienced less herbivorous pressure when the proportion of non-host trees increased. This could be the case for the T. processionea as well. Also, Moraal(2014) found that forests are usually colder and this could explain the decrease in likelihood to be infected around forests.
Overall, the hypothesis could not be confirmed or rejected because there was not a clear division between more plant rich areas and more cultivated ‘monocultural’ areas, since all significant land use types were cultivated and/or monocultural areas. Within these classes no real difference could be distinguished regarding richness of the plant community. It could be that there is a methodological flaw or that there is no effect of landscape complexity on the presence of OPM at all.
Methodological problems
It might be that the LGN7 is not precise enough to figure as a proxy for vegetation and thus natural enemies. For example, it is not clear what type of vegetation is present in ‘Natural grasslands’ and to what extent the area is mowed. Also, the LGN7 has an accuracy of 85% (Hazeu et al., 2014) which contributes to uncertainty when drawing conclusions.
The buffer zone of 200m could be reconsidered, as the exact natural enemies and their flying distance is not exactly known yet. Natural enemies may differ in their ability to disperse and search for food, which impacts their response to landscape complexity and thus might affect pest control at the landscape level (Bianchi et al 2006). More research is needed here.
Also, due to disorganization regarding the caterpillar observations, a fair amount of data was lost or could not be used. In the future, it might be helpful if the observations are conducted in a more consequent fashion.
Moreover, it remains unclear to what extent the model could be used as a method to predict infection. Significance of the AUC-value should be tested to confirm whether the model is predicting infection significantly better than a random model, and if not, conclusions based on the model should be reconsidered.
No effect
It could be that the effects of natural enemies are masked by other, more important factors such as temperature, precipitation, wind patterns, or soil compaction (Chaplin-Kramer 2011).
Also, the composition of the landscape could be an important factor. The extent to which a habitat functions as a source or sink for natural enemies depends also on its quality and size compared to the surrounding habitats (Bianchi et al. 2006).
For example, the isolation of a habitat patch may also be associated with the abundance and diversity of insect species. Natural habitat surrounded by human-dominated areas are less likely to be re-colonized by the native biota and natural enemies might be unable to track prey in patches isolated by unsuitable habitat. (Rickman & Connor 2003).
If this is the case with natural enemies of OPM, not only the sum of non-urban areas should be considered, but also the connectivity to other natural habitats.
Altogether, this research has not confirmed that a functional agrobiodiversity approach is a useful method to control T. processionea outbreaks. However, more research is highly recommended, as attracting natural enemies can be effective in many crop systems and therefore could not only save money for management but also increase biodiversity and contribute to a, in my opinion, more appealing landscape.
Acknowledgements
I would like to thank the Vlinderstiching, in particular Ties Huigens and Jurriën van Deijk, for involving me into the project. Also, I would like to thank Olga van de Veer for her guidance through the databases of the Province and helping me with ArcGIS. Furthermore, I would like to thank Peter Roessingh and Emiel van Loon for helping me with my many statistical problems.
Literature references
Bianchi, F. J. J. A., Booij, C. J. H., & Tscharntke, T. (2006). Sustainable pest regulation in agricultural landscapes: a review on landscape composition, biodiversity and natural pest
control. Proceedings of the Royal Society of London B: Biological Sciences, 273(1595), 1715-1727.
Castagneyrol, B., Jactel, H., Charbonnier, Y., Barbaro, L., & Dulaurent‐Mercadal, A. M. (2014). Egg mortality in the pine processionary moth: habitat diversity, microclimate and predation effects. Agricultural and Forest Entomology, 16(3), 284-292.
Chaplin-Kramer, R., O’Rourke, M. E., Blitzer, E. J., & Kremen, C. (2011). A meta-analysis of crop pest and natural enemy response to landscape complexity. Ecology letters, 14(9), 922-932.
Dussault, C., Ouellet, J. P., Courtois, R., Huot, J., Breton, L., & Jolicoeur, H. (2005). Linking moose habitat selection to limiting factors. Ecography, 28(5), 619-628.
Fransen, J.J., Stigter, H., Drijver, C.A., Koomen, I., Moraal, L.G., Spijker, J.H. (2006).
Literatuurstudie en voorstudie veldonderzoek Eikenprocessierups: Monitoring, bestrijdingsstrategieën en neveneffecten. Plantenziektekundige Dienst, Ministerie LNV.
Groenen, F., & Meurisse, N. (2012). Historical distribution of the oak processionary moth
Thaumetopoea processionea in Europe suggests recolonization instead of expansion. Agricultural and Forest Entomology, 14(2), 147-155.
Hazeu G.W., Schuiling, C., Dorland, G.J., Roerink, G.J., Naeff H.S.D., Smidt, R.A., (2014) Landelijk Grondgebruiksbestand Nederland versie 7 (LGN7): Vervaardiging, nauwkeurigheid en gebruik. Alterra-rapport 2548 ISSN 1566-7197.
http://content.alterra.wur.nl/Webdocs/PDFFiles/Alterrarapporten/AlterraRapport2548.pdf
Hellingman, S. en S. Mulder, 2012. Opnieuw natuurlijke vijand eikenprocessierups
ontdekt. Bericht uitgegeven door Biocontrole Onderzoek en Advies en De Natuurkalender op 18 mei 2012.
Huigens, personal correspondance 6-04-2016
Huigens, T., van Alebeek, F., Hellingman, S., Koopman, G., (2015). Bloemrijke wegbermen in Gelderland: Tussenrapportage. De Vlinderstichting. P-2015.034
Jactel, H., & Brockerhoff, E. G. (2007). Tree diversity reduces herbivory by forest insects. Ecology letters, 10(9), 835-848.
Leusink, E.H.P. en J.M. Reitsma, 2014. Monitoring natuurwaarden wegbermen Provincie
Gelderland: wegendistricten Noord en Zuid, 2014. Bureau Waardenburg B.V.,Rapportnummer 14-250, 110 pag.
Vakblad Natuur Bos Landschap 109: 3 - 5.
Meurisse, N., Hoch, G., Schopf, A., Battisti, A., & Grégoire, J. C. (2012). Low temperature tolerance and starvation ability of the oak processionary moth: implications in a context of increasing
epidemics. Agricultural and forest entomology, 14(3), 239-250.
R Core Team (2016). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/. [Assessed in May-Jun e 2016]
Rahlenbeck, S., & Utikal, J. (2015). The oak processionary moth: a new health hazard?. Br J Gen Pract, 65(637), 435-436.
Rickman, J., & F Connor, E. (2003). The effect of urbanization on the quality of remnant habitats for leaf‐mining Lepidoptera on Quercus agrifolia.Ecography, 26(6), 777-787.
Van Rijn, P.C.J.,Wäckers F.L., (2007). Bloemrijke akkerranden voeden natuurlijke vijanden. Entomologische Berichten 67(6): 226-230.
Sing T, Sander O, Beerenwinkel N and Lengauer T (2005). “ROCR: visualizing classifier
performance in R.” Bioinformatics, 21(20), pp. 7881. http://rocr.bioinf.mpi-sb.mpg.de. [Assessed in June 2016]
Wagenhoff, E., & Veit, H. (2011). Five years of continuous Thaumetopoea processionea monitoring: Tracing population dynamics in an arable landscape of south-western Germany. Gesunde Pflanzen, 63(2), 51-61
Van Alebeek, personal correspondence, 19-04-2016
Appendix A
Model output from stepwise regression model
Legend:
‘Natuurgraslanden’ = natural grasslands ‘Agrarisch gras’ = agricultural grass
‘Bipbb/Bos in primair bebouwd gebied’ = Forest in primarily developped area ‘Gipbb and Gispbb’ =grass in primarily/secondarily developed area
‘Overige gewassen’ = Other crops