## Abstract

Dendrolimus pini L. is a common and widespread moth in Europe, where severe outbreaks can defoliate Pinus sylvestris L. stands. Outbreaks are thought to be climate driven and may become more frequent and widespread with climate warming. The recent discovery of breeding populations of the moth in P. sylvestris plantations in Scotland has highlighted the importance of predicting outbreaks both within the core areas and at the margins of its current range. In this investigation, we used published data of damaging outbreaks plus historical climate data from Germany to build a relationship between climate conditions and outbreaks, and to develop a prediction model. Our analytical approach used principal component analysis and decision-tree data mining. German historical outbreaks showed relationships with climate variables, and provided evidence for a new damaging outbreak prediction model. The model uses the Seljaninov hydrothermal coefficient and decision-tree models on climate observations to predict where and when outbreaks may occur. The model was applied to European observed climate data and two climate projections using a GIS. In Europe, the model predicted future outbreaks in the Baltic States, Scandinavia, Russia and Scotland. In Scotland, more detailed analysis with probabilistic climate change projections showed an increasing risk of outbreaks through the twenty-first century.

## Introduction

Dendrolimus pini L. (Lepidoptera: lasiocampidae) caterpillars feed primarily, but not exclusively, on the needles of Pinus sylvestris L. The primary host is P. sylvestris, and the main other hosts are species of pine. However, D. pini also feeds to a lesser degree on Picea abies (L.) H. Karst, and a few other conifers, and may also represent a risk to Picea sitchensis (Bong.) Carr. in Britain. It is a serious defoliator of pine forests in North Central Germany (Gräber et al., 2012), Poland (Sierpinska, 1998) and other countries throughout its natural range of Europe and Asia. Dendrolimus pini is a large and conspicuous moth (Weckwerth, 1952; Möller et al., 2007): males have a wingspan of 60 mm, and the larger females have a wingspan of 80 mm. Each female lays ∼300 eggs on branches and stems of P. sylvestris in July. In October, the German populations of caterpillars descend the tree to overwinter in the pine litter, usually for one winter (November–February inclusive) and in cooler climates for two winters (Müller, 1992; Altenkirch et al., 2002). Re-emergence from hibernation occurs in March, and development is completed after six and, sometimes, seven instars by late spring. Pupation occurs between May and June, and emergence and mating of adult moths occurs in July.

One D. pini caterpillar can consume between 600 and 1000 pine needles through its development (Björkman et al., 2013). In the north-east German lowlands, D. pini populations undergo cyclical large-scale outbreaks that can cause major damage to P. sylvestris forests. Damaging outbreaks of D. pini have recently been recorded in Poland, Belarus, Ukraine and Germany (Food and Agriculture Organisation [FAO], 2010). Damaging outbreaks seem to show an increasing frequency of occurrence (Sierpinska, 1998). Field-based research on the outbreak ecology of D. pini has described the forest site types and the critical climatic and edaphic conditions required for severe defoliating outbreaks in pine forest regions where the insect is present. Particularly damaging outbreaks of D. pini have occurred in 30- to 60-year-old pure and even-aged P. sylvestris stands on sandy soils after dry winters and hot dry summers (Majunke et al., 1999; Majunke, 2000; Prescott, 2009, e.g. see Supplementary Figure 1).

A review by Sierpinska (1998) reported the occurrence of very severe D. pini outbreaks in Poland, in areas with higher mean annual temperatures, milder winters, lower wind velocities, less precipitation, fewer days with snow cover and lower long-term mean values of the Seljaninov hydrothermal coefficient (Seljaninov coefficient). The Seljaninov coefficient defines the degree of climatic dryness/wetness in months when the mean temperature is equal to or greater than 10°C. Sierpinska (1998) suggested that ‘abiotic factors do not affect the pine moth directly, but rather the pine moth population dynamics indirectly, through the host plant’. Lesniak (1976) defined annual values of the Seljaninov coefficient between 1.0 and 1.5 as characterizing the optimum climatic conditions for outbreaks of D. pini – above 1.5 is too humid and less than 1.0 is too dry for outbreak populations of the moth to develop. The most serious outbreaks occur in areas with a Seljaninov coefficient of 1.2 (Lesniak, 1976). Higher humidity and cooler temperatures have been reported as strongly reducing caterpillar feeding (Schwerdtfeger, 1970) and the rate of instar development.

Severe defoliating outbreaks of D. pini have been frequently observed in mature even-aged single-species P. sylvestris stands (Füldner, 2001) in the north-east German lowlands. Such outbreaks appear to have occurred particularly where the soil is infertile and sandy in texture (Zwölfer, 1962), such as podzolic soils (Mellec et al., 2009) with a thin litter layer of needles (Weckwerth, 1952), sites typical of P. sylvestris forests in Europe. Damaging outbreaks tend to be more severe in forest stands where trees are stressed by drought (Kaulfuß, 2012). When warm and dry summer climatic conditions combine over at least two consecutive years, through a pre-outbreak and outbreak year (Gräber et al., 2012), a very large population of caterpillars can emerge and defoliate P. sylvestris stands. In the north-east German lowland states, the dry climatic conditions combined with P. sylvestris forest on sandy textured soils provide suitable conditions for severe outbreaks.

The course of population increase leading to a severe outbreak is termed a gradation by Schwerdtfeger (1970), who defined the ‘culmination’ phase as the period over which a large population of insects emerge and defoliate forest stands. ‘Culminations’ of D. pini have occurred in southern areas of the north-east German lowlands in 1994–1998, 2003, 2004, 2006, 2007, 2013 and 2014 (NW-FVA, 2010). The culmination phase is described in this article simply as an outbreak, and this is defined as a defoliation loss of at least 50 per cent of the needles (Gräber et al., 2012). Outbreak severity is identified by critical numbers of moths (70–90 moths per pheromone trap; (NW-FVA, 2010) or a critical density of hibernating larvae during winter monitoring, i.e. typically more than 10 per m2) leading to severe defoliation. The first year of a severe defoliation we refer to as the outbreak year. We refer to outbreaks that occurred for more than 1 year as ‘continued-outbreak’ to signify severe defoliation continuing for more than 1 year.

Dendrolimus pini is widely distributed in central European forests, but severe defoliating outbreaks are relatively uncommon in its continental range outside the north-east German lowlands and lowland areas of northern Poland. For example, a recent study (Haynes et al., 2014) reported that five D. pini outbreaks had occurred in Bavarian forests in southern Germany over the last 200 years, but no outbreaks had been reported there since the 1930s. Compared with the southern state of Bavaria, outbreaks of D. pini have frequently occurred across large areas of forest in the north-east German lowlands (Mellec et al., 2011), and this has continued through recent decades (1990s, 2000s and 2010s). Outbreaks occur in synchrony across large areas of forest, apparently in response to climatic conditions. This type of synchrony is typical of populations with a density-dependent structure (Koenig, 2002) that respond in a manner termed a Moran effect (Hudson and Cattadori, 1999). The Moran effect occurs when density-independent factors such as climatic conditions bring the population dynamics into synchrony, probably through increased fecundity, increased larval survival from predators and parasites, and reduced resistance from the host tree (Klemola et al., 2008). Outbreak occurrence might also be related to P. sylvestris forest structural composition and species components; single-species even-aged stands on uniform topographic and edaphic sites may provide less niche diversity for predators and parasites.

Approximately 100 D. pini moths were discovered in five woodlands in north-east Scotland in 2008. The species was not previously recorded in Scotland, and there was uncertainty whether the new records were for an overlooked resident. So D. pini might be native to Scotland, or it could be a relatively recent arrival. The discovery has led to concern that D. pini may now expand its range in Scotland and Britain, possibly leading to future damaging outbreaks. Furthermore, projected changes in the climate in eastern Scotland and eastern England could amplify these outbreaks.

Our methodological rationale for this study was to design an analysis that tested the efficacy of a range of climate variables, including the Seljaninov coefficient, to predict where and when outbreaks of D. pini would occur. Based on an analysis of the historical outbreak records from the north-east German lowlands, we developed for this study (1) a climatic analysis showing the conditions under which damaging outbreaks could occur in Europe; (2) a decision-tree model using climate observations relating to the periods of caterpillar development, and which combine to cause an outbreak; (3) a spatial analysis of how climate change may extend the future range and outbreak potential of D. pini in Europe and (4) an estimation of the changing risk of damaging outbreaks in north-east Scotland using the UKCP09 probabilistic climate data over the century.

## Materials and methods

The Seljaninov coefficient is a combined measure of climatic warmth and wetness. It has been used in eastern and central Europe to describe the wetness/dryness of the warmer summer months when the monthly mean temperature is greater than 10°C (Dronin and Kirilenko, 2008). Previous studies of D. pini (Lesniak, 1976; Sierpinska, 1998) have used the Seljaninov coefficient to describe the optimal forest climatic conditions of places where D. pini outbreaks have occurred. The Seljaninov coefficient has also been used in several studies in central Europe to (1) assess growth in forest stands, (2) measure the occurrence of fire damage (Drobyshev and Niklasson, 2004), (3) demonstrate biogeoclimatic differences between locations (Hubalek and Horakova, 1988) and (4) describe drought years affecting food production in Russia (Dronin and Kirilenko, 2008; Dronin and Kirilenko, 2011).

Calculation of the Seljaninov coefficient involves, for a given year, accumulating the mean monthly temperature if it was greater than or equal to 10°C, and accumulating the total rainfall (in those months when T ≥ 10°C). The total rainfall for this period is divided by one-tenth of the accumulated temperature over the same period (equation 1). For months when the mean monthly temperature was below 10°C, no accumulation of rainfall or temperature was made in equation 1. Consequently, the number of months used in the calculation of the Seljaninov coefficient could vary from year to year.

(1)
$SHC=P∑T≥10T/10.$

Alternative climatic indexes and coefficients to the Seljaninov coefficient have been reviewed by Stadler (2005), and include the Köppen Index, Thornthwaite Index and the Meigs Index. The Köppen aridity index and Meigs aridity index are applied through an aridity classification system that gives lower resolution categories of aridity than the continuous coefficient of Seljaninov. The Thornthwaite aridity index requires the intermediate calculation of potential evapotranspiration, and this introduces error if accurate net radiation data are not available (Allen et al., 1998). Therefore, we decided to proceed with the Seljaninov coefficient, as it provides continuous combined values of climatic warmth and wetness, and there are no intermediate steps that introduce error. Our aim is to use the Seljaninov coefficient to build on previous work of D. pini outbreak ecology (Lesniak, 1976; Sierpinska, 1998).

We decided to analyse D. pini outbreak records with historical climate observations from the north-east German lowlands to assess which climatic drivers influence the occurrence of an outbreak. Our thinking was that if climatic conditions could be shown to be driving D. pini outbreaks, we might apply these to projected future climates, and assess the possibility of future outbreaks across Europe. In addition, we could test the likelihood of climatic drivers causing outbreaks using UK probabilistic climate projections to assess the risk of future outbreaks in parts of Scotland where a recent colonization of the insect has occurred.

We used a principal component analysis (PCA) to separate clusters of points representing Seljaninov coefficient values in pre-outbreak, outbreak and non-outbreak years, and we were able to statistically test clusters of these data for differences. We also tried a data mining technique (decision tree) to better understand climatic variable ranges relating to important phases of the life history of D. pini.

Finally, having satisfied that the Seljaninov coefficient is robust in predicting conditions suitable for an outbreak, we calculated the coefficient spatially across Europe using projected climate data from published global climate models (ECHAM5 and HADCM3), and also for parts of Scotland using probabilistic regional climate model projections from the United Kingdom Climate Programme (UKCP09), published in 2009.

### Dendrolimus pini outbreak records

Outbreak records were obtained from Gräber et al. (2012) (see Supplementary Figure 2) and derived from the annual reports of the Federal State Department of Forest Protection and Monitoring. Records extend back to the early 1920s and outbreaks were determined when: listed as an outbreak; extensive feeding appeared regionally (>50 per cent defoliation) or extensive insecticide treatment had been implemented. We focussed on those outbreak records from 1950 onwards to ensure the consistency of a complete climatic dataset from each of 12 climate stations with historical data in the region. Outbreaks were partitioned into four classes: (1) year(s) prior to outbreak – ‘pre-outbreak’, (2) year of ‘outbreak’, (3) ongoing or ‘continued-outbreak’ and (4) ‘non-outbreak’.

### Climate data

We assembled three climatic datasets. The first comprised observed climate data from the north-east German lowlands. Mean monthly temperature and total monthly rainfall values were obtained from the German Meteorological Service (Deutscher Wetterdienst – DWD) (http://www.dwd.de/EN/Home/home_node.html (accessed 7 January 2016)), for 12 climate stations across four states of the north-east German lowlands: Lower Saxony, Mecklenburg-Vorpommern (northerly) states with a more oceanic climate and Saxony-Anhalt and Brandenburg (southerly) states with a more continental climate. Pinus sylvestris stands in these states have experienced periodic outbreaks of D. pini (Supplementary Figure 3). Data were available for extensive historic periods, from the early twentieth century for some of the climate stations, but often with missing periods. We decided to focus on the period from 1950 to 2011, to maintain a complete data record. Data were obtained for the climate stations: Angermuende, Bremen, Brocken, Cottbus, Hamburg, Hannover, Magdeburg, Marnitz, Neuruppin, Potsdam, Schwerin and Tegel (Berlin) (see Supplementary Figure 4).

We also wished to assess how climate change projections may cause the Seljaninov coefficient range optimal for D. pini outbreaks to shift in Europe. This was assessed using a second dataset of high-resolution (0.5 arcsec) mean monthly temperature and total monthly rainfall, obtained from WorldClim (Hijmans et al., 2005), for the 1961–1990 climatic baseline period. This is global mean data for the climate baseline period providing (1) high spatial resolution and (2) a spatial baseline with which to compare future climate projections. For climate projections at the same spatial resolution as WorldClim, we obtained mean monthly temperature and totally monthly rainfall from the CCAFS data portal (http://www.ccafs-climate.org/data/ (accessed 7 January 2016)), for two Global Climate Models (GCM) – HadCM3 and ECHAM5 (IPCC, 2007a, b) for the 30-year climate projection periods centred on 2050 and 2080. These data were all prepared for the land area of Europe.

The third source of climate data was the UKCP09 (http://ukclimateprojections-ui.metoffice.gov.uk/ui/admin/login.php (accessed 7 January 2016)) online weather generator providing probabilistic climate data (Murphy et al., 2009) from simulations of daily temperature and precipitation in the UK. The weather generator provides climatic variables from 2010 to 2099 for the IPCC SRES A1FI, A1B and B1 scenarios, for seven 30-year climatic periods (represented by decades). We used only the A1FI and B1 (IPCC, 2007a, b) emission scenarios in the analysis to examine differences between projections for the upper and lower emission scenarios. For the baseline climate (1961–1990), and for each of the seven decades (2020s–2080s) of the B1 and the A1FI emission scenarios, we obtained climate data with probabilities for each of the 5 km sample squares that covered the six P. sylvestris locations assessed in Scotland. The outputs from the weather generator included daily precipitation, maximum and minimum temperature, and these data were used to calculate monthly mean temperature, total precipitation and annual values of the Seljaninov coefficient.

### Site data

The incorporation of soil data was an important factor in the analysis. Dendrolimus pini outbreaks in the north-east German lowland forests are particularly associated with P. sylvestris forest stands on sandy textured mineral soils. Soil maps (NLFB, 1999; LAGB, 2012) showed that the soil types of Lower Saxony and Saxony-Anhalt in forest areas subject to outbreaks were largely dominated by acidic and freely draining, podzol and podzolic brown earth soils derived from fluvioglacial sands and gravels and windblown glacial sand. The soil characteristics suitable for both P. sylvestris forest and the occurrence of D. pini across Europe were obtained from the European soil raster (http://eusoils.jrc.ec.europa.eu/ESDB_Archive/ESDB_data_1k_raster_intro/ESDB_1k_raster_data_intro.html (accessed 7 January 2016)) mapped at a scale of 1-km resolution (Liedekerke et al., 2006; Panagos et al., 2012). The five soil types – acrisols, cambisols, leptosols, podsols and umbrisols – were selected to reflect the areas edaphically suited to P. sylvestris forest.

A second soil database of the Soil Survey of Scotland (obtained from Lilly et al., 2010) was used to select forests likely to experience a D. pini outbreak. Using a GIS, we overlaid forest areas with the 1 : 250 000 soil map, selecting forests where soils had the following characteristics: freely draining sandy textured soils derived from the Moine schist and from Old Red Sandstone lithologies to form humus-iron podzols, podzols and podzolic brown earths. Three of the forests selected are plantations (Beauly, Black Isle and Culbin), and three are semi-natural ‘Ancient Caledonian Pinewood Reserves’ (Glen Affric, Great Glen and Glenmore). These sites were chosen for their proximity to the current breeding area of D. pini in Scotland.

### Analysis

The analysis consisted of three stages: firstly, validating the relationship between the Seljaninov coefficient and annual and seasonal climatic variables on outbreak years, pre-outbreak years and non-outbreak years, using statistical models on data from the north-east German lowlands; secondly, a GIS-based spatial analysis to predict the potential shifts in climate space within sites suited to D. pini in Europe and thirdly, a detailed analysis of probabilistic climate data on the six P. sylvestris forest sites in Scotland to establish the likelihood of future outbreaks.

Firstly, we wanted to test whether the Seljaninov coefficient is a good indicator of areas where D. pini outbreaks had occurred. However, since it is a less common coefficient, we also used familiar climatic variables, mean values for temperature – from monthly mean temperature (T), and total values for rainfall and precipitation (separately) – monthly total rainfall (P), all from the German historical climate records between 1950 and 2011. For annual periods, using these data, we calculated mean values of temperature and total values of precipitation, and calculated the Seljaninov coefficient each year.

We also wanted to predict years in which outbreaks would occur from the climatic data associated with the important growth stages of D. pini. The Seljaninov coefficient and the mean temperature were calculated each year for fixed time periods corresponding to the most likely start and end months of D. pini growth and developmental stages. We settled on the following growth stage periods: the first to third/fourth instar period (August–October), the third/fourth to seventh instar period (March–June), the flight period (July) and the month marking the end of the winter hibernation (March) when caterpillars emerge from the leaf litter and move back in to the foliage. In addition, we accumulated total precipitation for the winter period October–March inclusive (rain and snow), as well as the accumulated winter rainfall when mean daily temperature was over 0°C (rain), to assess the wetness of winter conditions. Finally, the total rainfall in the first to third instar period (August–October), and the third to seventh instar period (March–June), was also calculated.

Data from the two northern states and the two southern states of the north-east German lowlands indicated, respectively, 3 and 14 outbreaks between 1950 and 2011. These data were analysed using the Canoco (http://www.canoco5.com (accessed 7 January 2016)) (ter-Braak and Smilauer, 1998) PCA to interpret the annual variations in climate across the north-east German lowlands. Resulting principal components from 12 climatic variables, including the Seljaninov coefficient, were used to identify inter-relationships between climatic variables and outbreaks of D. pini.

To test the number of years when differences in the climatic conditions led to an outbreak, we set up a statistical analysis. This was designed to test for differences in the value of Seljaninov coefficients from 1 to 5 years prior to either an outbreak occurring or no outbreak occurring. For the analysis, we created two sets of Seljaninov coefficient data from the climate station records: (1) a set of data covering Seljaninov coefficients from 1 to 5 years prior to a year when an outbreak was recorded only in the last year and (2) a similar set of data from 1 to 5 years prior to no outbreak having been recorded in any of the years. We were constrained in the number of antecedent years that could be analysed, as there were insufficient data to assess more than 5-year antecedent climatic data.

The analysis involved comparing differences in the estimated PCA distance between each Seljaninov coefficient shown on the bi-plot from the PCA centroid, for each of the 1–5 years prior to an outbreak, and for each of the 1–5 years prior to a year when no outbreak occurred. The statistical analysis was performed in a bespoke Fortran programme (GNU Fortran compiler (https://gcc.gnu.org/fortran/ (accessed 7 January 2016))) to estimate confidence thresholds from a 1000 iteration bootstrap procedure from the observed outbreak (and no outbreak) records. These data allowed us to test the degree of clustering of Seljaninov coefficients about the PCA centroid, for years leading to an outbreak (or no outbreak). The statistical test was made by comparing the mean distance from the PCA centroid of all the data with the annual Seljaninov coefficient distance from the centroid over 1 year, 1–2 years, 1–3 years, 1–4 years and over the period 1–5 years before an outbreak or no outbreak occurred. Differences in mean distance were assessed against the P-value confidence thresholds estimated from the bootstrap procedure.

In a second analysis to model when D. pini outbreaks might occur, we used the SAS Enterprise Miner 7.1 software to build a decision-tree model composed of rules among explanatory climatic variables. Decision weights were allocated using inverse prior probabilities. For example, because 19 per cent of the dataset comprised outbreak years and 81 per cent non-outbreak years, the weighting for estimating an outbreak correctly was 5.3 units (1/0.19) and the weighting for correctly identifying a non-outbreak correctly was 1.2 units (1/0.81). Only three outbreaks occurred in the two northerly states, and these data were omitted from the data mining model. As the southern states dataset contained only 74 records, all of the data were used to train the decision tree, leaving no validation dataset. The Pearson χ2 statistic was used as the splitting method for branches and nodes.

### Future changes in climate space indicating the potential outbreaks of D. pini in Europe

We calculated an annual Seljaninov coefficient for Europe for three periods: the generally accepted historic climatic baseline period (1961–1990) from WorldClim and two future 30-year projected climate periods centred on the 2050s and 2080s. We used both the HadCM3 and the ECHAM5 SRES A1B emission scenarios, using the same method as described for the north-east German lowlands historic data. These data were processed in a GIS raster calculator, with monthly mean temperature and total monthly rainfall to calculate the Seljaninov coefficient. Maps of the spatial variation in the Seljaninov coefficient were intersected with a European raster of five soil types suited to P. sylvestris forest and likely to coincide with D. pini habitat in Europe.

### Future probabilistic climate projections in north-east Scotland

The probabilistic data, consisting of 100 model runs for each of the seven decades (2020s–2080s), were randomly selected across the probability density function from the 10 000 available probabilistic model variants available in the UKCP09 weather generator. Using these data, we calculated monthly mean temperature and the Seljaninov coefficient for each forest location in north-east Scotland. The output consisted of probabilities for Seljaninov coefficient ranges.

## Results

### Synchrony of historic climate data across the north-east German lowlands

Summer season Seljaninov coefficients for the north-east German lowlands indicated that the climate observations from 11 of the 12 climate stations exhibited a high degree of spatial synchrony. The 12th station in Hamburg showed cooler more oceanic climate observations, resulting in a greater Seljaninov coefficient (0.4 units) in most years. The remaining 11 climate stations were more than 50 km from the North Sea Coast and distributed across the north-east German lowlands, resulting in very similar monthly mean temperatures and rainfall in the summer months. This resulted in the Seljaninov coefficient varying by no more than 0.4 units across distances of 200 km (Figure 1) in any single year.

Figure 1

Average annual difference in the Seljaninov coefficient among 12 climate stations in the north-east German lowlands of Lower Saxony, Brandenburg, Saxony-Anhalt and Mecklenburg-Vorpommern.

Figure 1

Average annual difference in the Seljaninov coefficient among 12 climate stations in the north-east German lowlands of Lower Saxony, Brandenburg, Saxony-Anhalt and Mecklenburg-Vorpommern.

### Relationship between D. pini outbreaks and the Seljaninov coefficient

There were large differences in D. pini outbreak frequency between northern states and southern states. This is demonstrated by 3 outbreaks occurring in 61 years in the northern states compared with 14 outbreaks in the southern states. We wanted to investigate whether there were climatic differences that might account for the variation in outbreak frequency. Consequently, data were organized into two separate PCAs: the oceanic northern states (Figure 2a) and the more continental southern states (Figure 2b).

Figure 2

(a and b) PCA bi-plots showing the distribution of climate in Dendrolimus pini years of ‘outbreak’ (•), ‘pre-outbreak’ (○), ‘continued-outbreak’ (◊) and ‘non-outbreak’ (x) years for climate stations in the northern oceanic states (Lower Saxony and Mecklenburg-Vorpommern) and the southern continental states (Brandenburg and Saxony-Anhalt). Climatic vectors of precipitation, temperature and Seljaninov coefficient for key periods in the development of the caterpillars of D. pini are shown. Note: joining lines (convex hull lines) include 90% of the data points. SHC = Seljaninov coefficient.

Figure 2

(a and b) PCA bi-plots showing the distribution of climate in Dendrolimus pini years of ‘outbreak’ (•), ‘pre-outbreak’ (○), ‘continued-outbreak’ (◊) and ‘non-outbreak’ (x) years for climate stations in the northern oceanic states (Lower Saxony and Mecklenburg-Vorpommern) and the southern continental states (Brandenburg and Saxony-Anhalt). Climatic vectors of precipitation, temperature and Seljaninov coefficient for key periods in the development of the caterpillars of D. pini are shown. Note: joining lines (convex hull lines) include 90% of the data points. SHC = Seljaninov coefficient.

For each PCA, the classified outbreak data were combined with the climatic data. The centre of each bi-plot (Figure 2) represents the mean position of the climatic variables transformed by the PCA axes, and outbreak classes are identified by a symbol delineating to which of the four classes it was ascribed. The PCA showed that for both sets of data (Figure 2a,b), ∼50 per cent of the variation in the 11 dimensional PCA space was described by PCA axes 1 and 2. The annual value of the Seljaninov coefficient aligned parallel with PCA axis 1, showing a high degree of certainty (vector of 0.9 of the axis length) in both northern and southern states. Winter rainfall aligned parallel to axis 2, also with a high degree of certainty, and other variables were explained by a combination of PCA axes 1 and 2 with a reasonably high degree of certainty.

For the northern oceanic states, the two classes of pre-outbreak and outbreak years clustered to the left of the PCA centroid (Figure 2a). This demonstrated a trend of lower Seljaninov coefficient values associated with pre-outbreak and outbreak years, where the average value of the Seljaninov coefficient was located at the PCA centroid. Additionally, Figure 2a shows lower than average rainfall and warmer conditions during the March–June period corresponding to the third/fourth to seventh instar period in both pre-outbreak and outbreak years. The bi-plot for the southern continental states (Figure 2b) showed the 14 outbreak years distributed more centrally about the centroid of the bi-plot, showing outbreaks occurred in years with average climatic conditions. However, for an outbreak to occur in the following year, the pre-outbreak year tended to have a lower Seljaninov coefficient, and to be slightly drier and warmer than years preceding non-outbreak. Continued-outbreaks were also more common in the southerly states. The climatic conditions for continued-outbreaks were similar to average conditions, but without the wider extreme cooler and wetter conditions, and typified by lower Seljaninov coefficients between August and October, associated with a reduced late-summer/early-autumn rainfall.

We tested the degree of clustering of Seljaninov coefficients about the bi-plot centroid in outbreak, pre-outbreak and non-outbreak years, to determine the type and length of antecedent weather conditions that may drive an outbreak of D. pini. The mean distance of 119 years of records about the bi-plot centroid was 1.60 units. By bootstrapping these data 1000 times, we calculated a one-sided lower 95 per cent confidence threshold at a distance of 1.31 units from the bi-plot centroid, and a 99 per cent confidence threshold at a distance of 1.23 units from the bi-plot centroid. In Table 1, a comparison of outbreak and non-outbreak antecedent weather conditions shows a greater clustering about the centroid in the year of an outbreak, with a mean distance from the centroid of 1.15 units (P < 0.01) compared with non-outbreak years (distance from centroid 1.82 units, n.s.), and highly significant clustering in the pre-outbreak year ‘year-1’ (distance from centroid 1.20 units, P < 0.01) compared with years prior to non-outbreak (distance from centroid 1.80 units, n.s.). Although the year-2 (2 years prior to an outbreak) is also significantly clustered, we have to take account of the influence that 66 per cent of these data are from the year of outbreak and the pre-outbreak year. The result nevertheless suggests that either one or two climatically optimal pre-outbreak years may be needed for an outbreak to occur. The observed data for year-3 and year-4 showed no significant clustering.

Table 1

Mean distance from the PCA centroid to Seljaninov coefficient locations based on climate data from the continental north-east German lowland states of Saxony-Anhalt and Brandenburg from 1950 to 2010

No. of years prior to an outbreak Mean distance from PCA bi-plot centroid (NNo. of years prior to non- outbreak Mean distance from PCA bi-plot centroid (N
Outbreak year 0 1.15 (14) ** No-outbreak year 0 1.82 (47)
+ year-1 (non-outbreak) 1.20 (14) ** + year-1 (non-outbreak) 1.80 (47)
+ year-2 (non-outbreak) 1.30 (9) * + year-2 (non-outbreak) 1.91 (34)
+ year-3 (non-outbreak) 1.45 (7) + year-3 (non-outbreak) 1.94 (28)
+ year-4 (non-outbreak) 1.40 (6) + year-4 (non-outbreak) 1.81 (28)
No. of years prior to an outbreak Mean distance from PCA bi-plot centroid (NNo. of years prior to non- outbreak Mean distance from PCA bi-plot centroid (N
Outbreak year 0 1.15 (14) ** No-outbreak year 0 1.82 (47)
+ year-1 (non-outbreak) 1.20 (14) ** + year-1 (non-outbreak) 1.80 (47)
+ year-2 (non-outbreak) 1.30 (9) * + year-2 (non-outbreak) 1.91 (34)
+ year-3 (non-outbreak) 1.45 (7) + year-3 (non-outbreak) 1.94 (28)
+ year-4 (non-outbreak) 1.40 (6) + year-4 (non-outbreak) 1.81 (28)

Data classified by pre-outbreak years (year-1, year-2, … , year-4), outbreak year (year 0) and control data with four non-outbreak years preceding a non-outbreak (year 0), showing the level of significance of clustering of Seljaninov coefficients.

Level of significance of PCA distance values for outbreak and non-outbreak years (and the 4 years prior to these), confidence thresholds estimated by a bootstrap procedure to calculate P-values: 1.23 units (**P < 0.01), 1.31 units (*P < 0.05), the mean distance of Seljaninov coefficients for all years from the bi-plot centroid was 1.6 units.

The analysis showed clearly that for an outbreak to occur in a given year, the weather conditions in that year and 1 year (or possibly 2 years) prior to an outbreak must have Seljaninov coefficient values close to the bi-plot centroid. The analysis also showed that non-outbreak years need not be preceded by extreme weather, and that one of the two or three pre-outbreak years with a more extreme Seljaninov coefficient (drier, wetter, hotter or cooler) would prevent an outbreak occurring. The main finding is that two (or possibly three) consecutive years very close to the average Seljaninov coefficient (1.2) for the southern more continental states will normally lead to an outbreak.

### Climatic data corresponding to D. pini caterpillar life cycle stages leading to outbreaks

There were a number of climatic differences between D. pini outbreak and non-outbreak years (Table 2). The average temperature of the July flight period during an outbreak year was warmer (P < 0.05) than a non-outbreak year. Temperatures were higher in pre-outbreak years for the period of the first to third instar (August–October, P < 0.05) and in outbreak years for the third to seventh instar (March–June, P < 0.001). In the year prior to an outbreak, the period of the third to seventh instar was warmer (P < 0.01) compared with years of non-outbreak. Outbreak years had warmer early spring temperatures in March (P = 0.06) at the time when caterpillars emerge from leaf litter on the forest floor and ascend into the forest canopy. As well as increased temperatures during the third to seventh instar period of outbreak years, the total rainfall was significantly less (P < 0.05) than in years of non-outbreak. In contrast to non-outbreak years, D. pini outbreak years were more likely to be associated with two (or possibly three – Table 1) consecutive years of annual Seljaninov coefficients in the range 1.0 < SHC < 1.5 (P < 0.05) (Table 2).

Table 2

Mean climatic values (SD) for periods relating to Dendrolimus pini caterpillar development from climate stations in the north-east German Lowlands

Climate variable by month – instar/flight period relevance Outbreak year No outbreak year Units
Temp July – flight period 18.4 (1.5)* 17.5 (1.6) °C
Temp August–October – first–third instar period 13.9 (0.6)* 13.4 (0.9) °C
Temp March – emergence from soil to canopy 4.6 (1.2)* 3.9 (2.2) °C
Temp March–June – third–seventh instar period 11.1 (0.8)*** 10.1 (1.1) °C
Temp March–June – third–seventh instar period in year prior to outbreak 10.8 (1.2)** 10.1 (1.0) °C
Rain August–October – first–third instar period 172 (53) 159 (50) mm
Rain October–March – winter hibernation period 130 (69)* 146 (62) mm
Rain March–June – third–seventh instar period 170 (43)* 201 (59) mm
SC for a year August–July – instar development and flight period 1.18 (0.27)* 1.27 (0.36) –
SC August–October – first–third instar period 1.21 (0.51) 1.21 (0.49) –
SC March–June – third–seventh instar period 1.20 (0.53) 1.35 (0.55) –
Climate variable by month – instar/flight period relevance Outbreak year No outbreak year Units
Temp July – flight period 18.4 (1.5)* 17.5 (1.6) °C
Temp August–October – first–third instar period 13.9 (0.6)* 13.4 (0.9) °C
Temp March – emergence from soil to canopy 4.6 (1.2)* 3.9 (2.2) °C
Temp March–June – third–seventh instar period 11.1 (0.8)*** 10.1 (1.1) °C
Temp March–June – third–seventh instar period in year prior to outbreak 10.8 (1.2)** 10.1 (1.0) °C
Rain August–October – first–third instar period 172 (53) 159 (50) mm
Rain October–March – winter hibernation period 130 (69)* 146 (62) mm
Rain March–June – third–seventh instar period 170 (43)* 201 (59) mm
SC for a year August–July – instar development and flight period 1.18 (0.27)* 1.27 (0.36) –
SC August–October – first–third instar period 1.21 (0.51) 1.21 (0.49) –
SC March–June – third–seventh instar period 1.20 (0.53) 1.35 (0.55) –

SC, Seljaninov coefficient. Level of significance between values in outbreak years and non-outbreak years within rows: *P < 0.05; **P < 0.01; ***P < 0.001.

### Testing climate variables to predict when an outbreak may occur

We conducted a data mining analysis, to develop a decision-tree model of climatic conditions likely to lead to an outbreak, to test when outbreaks were likely to occur. There were insufficient data on outbreaks from the northern states (three outbreaks in 61 years) and these were removed from further analysis. Results for southerly states (Figure 3) show a set of chronologically organized rules relating climatic values to relevant insect life stages to predict the likelihood of an outbreak in July. For an outbreak to occur, all of the following rules must be met: (1) the mean temperature of the previous August–October period is greater than or equal to 13.15°C, (2) the total precipitation through the previous August–October period is greater than or equal to 122.5 mm, (3) the temperature in March of the year of outbreak is greater than or equal to 1.75°C and less than 6.45°C and (4) the temperature of the spring and early summer (March–July) in the year of outbreak is greater than or equal to 9.4°C. A single set of decision-tree rules were able to predict every one of the 14 outbreak events, but the rules over-predicted five outbreaks where non-outbreak years were recorded (false positives).

Figure 3

Classification-tree model to predict when a Dendrolimus pini outbreak is about to occur. The model uses temperature and rainfall observations collated for time periods relevant to specific insect life cycle development stages.

Figure 3

Classification-tree model to predict when a Dendrolimus pini outbreak is about to occur. The model uses temperature and rainfall observations collated for time periods relevant to specific insect life cycle development stages.

### Projections of D. pini outbreaks in current and future European climates

Combining soil data with the baseline climate and GCM projections, we conducted a spatial European continental scale assessment of projected climate change effects on the Seljaninov coefficient critical value range for outbreak. Critical values between 1.0 and 1.5 are shown in Figure 4a for the baseline climate data. This is a period through which some of the outbreaks have occurred in the north-east German lowlands. Figure 4b–e shows the projected future distribution of Seljaninov coefficients centred on the projected 30-year climate periods of 2050 and 2080 for the A1B emission scenario. European distribution maps for the baseline period, the 2050s and the 2080s future projections for both GCM models showed considerable agreement in the spatial and temporal changes of the Seljaninov coefficient associated with outbreaks of D. pini. Indeed, the climate projections suggested climatic conditions becoming more suited to D. pini outbreaks in northern Germany, Poland and southern Sweden and Finland. More southerly parts of Germany, the Netherlands, west-central, southern and eastern France, the Czech Republic, Slovakia, the Balkans, southern England, Wales and eastern Scotland could also be potentially affected in the future.

Figure 4

(a–e) Comparison of the Seljaninov coefficient below, between and above the range 1.0–1.5 in regions of freely draining, infertile, soil types (Liedekerke et al., 2006) associated with Pinus sylvestris forest in Europe that could be suitable for damaging outbreaks of Dendrolimus pini. (a) Baseline climate 1961–1990, (b) projected Seljaninov coefficient from HadCM3 A1B in 2050, (c) projected Seljaninov coefficient from HadCM3 A1B in 2080, (d) projected Seljaninov coefficient from ECHAM5 A1B in 2050 and (e) projected Seljaninov coefficient from ECHAM5 A1B in 2080.

Figure 4

(a–e) Comparison of the Seljaninov coefficient below, between and above the range 1.0–1.5 in regions of freely draining, infertile, soil types (Liedekerke et al., 2006) associated with Pinus sylvestris forest in Europe that could be suitable for damaging outbreaks of Dendrolimus pini. (a) Baseline climate 1961–1990, (b) projected Seljaninov coefficient from HadCM3 A1B in 2050, (c) projected Seljaninov coefficient from HadCM3 A1B in 2080, (d) projected Seljaninov coefficient from ECHAM5 A1B in 2050 and (e) projected Seljaninov coefficient from ECHAM5 A1B in 2080.

### Projections of future D. pini outbreaks in Scotland

Results for Culbin Forest – consisting of even-aged stands of P. sylvestris (Figure 5a,c) – and Glen Affric Forest – one of the largest ‘Caledonian Pine Forest Reserves’ in Scotland (Figure 5b,d) – show the cumulative probability distributions of the Seljaninov coefficient from baseline (1961–1990) through each of seven successive 30-year climatic periods at decadal intervals. For Culbin Forest, a projected increase in the probability of the Seljaninov coefficient value of 1.2 (optimal for D. pini damage) occurred in the A1FI (Figure 5a) and the B1 scenario (Figure 5c). For Glen Affric Forest (Figure 5b,d), only small changes in the projected Seljaninov coefficient occurred, affecting higher values (wetter climate) of the Seljaninov coefficient greater than 1.5.

Figure 5

Cumulative probability (%) distributions of Seljaninov coefficient for the baseline climate (1961–1990) and for future simulated climatic periods at seven decadal time steps through the twenty-first century. (a) Culbin Forest A1FI emission scenario, (b) Glen Affric Forest A1FI emission scenario, (c) Culbin Forest A1B emission scenario and (d) Glen Affric Forest A1B emission scenario, Scotland.

Figure 5

Cumulative probability (%) distributions of Seljaninov coefficient for the baseline climate (1961–1990) and for future simulated climatic periods at seven decadal time steps through the twenty-first century. (a) Culbin Forest A1FI emission scenario, (b) Glen Affric Forest A1FI emission scenario, (c) Culbin Forest A1B emission scenario and (d) Glen Affric Forest A1B emission scenario, Scotland.

From the summary of the six forest areas in Table 3, Beauly Forest, Culbin Forest and the Black Isle Forest show an increased risk of future critical coefficient values in the range 1.0–1.5. Between the baseline climate and the B1 projection for 2080, for Beauly Forest, the risk of Seljaninov coefficients in the range 1.0–1.2 rises slightly from 9.5 to 13 per cent, with a bigger increase from 30.5 to 50 per cent in the range 1.2–1.5. For the A1FI projection at Beauly Forest, there is a rise in the probability of Seljaninov coefficients in the ranges of 1.0–1.2 and 1.2–1.5, from 9.5 to 20 per cent and from 30.5 to 40 per cent, respectively. Increased probabilities also occur for the two emission scenarios at Culbin Forest and the Black Isle Forest. For the P. sylvestris plantation forests, the increase in probability is greater in the wetter Seljaninov coefficient range 1.2–1.5 for the B1 emission scenario, and greater in the drier 1.0–1.2 range for the A1FI emission scenario.

Table 3

Probability (%) of Seljaninov coefficient values in the critical range (1.0–1.2 and 1.2–1.5) associated with damaging outbreaks of Dendrolimus pini, calculated from the UKCP09 weather generator for six sample squares associated with pinewoods in proximity to the breeding population in north-eastern Scotland

SRES emission scenario Seljaninov index range Time period Probability of Seljaninov index (%) in critical range for forests in north-east Scotland

Glenmore Glen Affric Great Glen Beauly Black Isle Culbin
B1 1–1.2 1961–1990 0.5 1.5 5.5 9.5 7.5 14.5
2040–2050 11 14 14
2070–2080 13 17 16
1.2–1.5 1961–1990 21.5 30.5 19.5 25.5
2040–2050 32 43 30 33
2070–2080 35 50 32 33
A1F1 1–1.2 1961–1990 0.5 1.5 5.5 9.5 7.5 14.5
2040–2050 10 26 11 26
2070–2080 11 20 16 27
1.2–1.5 1961–1990 21.5 30.5 19.5 25.5
2040–2050 10 32 34 29 32
2070–2080 12 12 33 40 30 31
SRES emission scenario Seljaninov index range Time period Probability of Seljaninov index (%) in critical range for forests in north-east Scotland

Glenmore Glen Affric Great Glen Beauly Black Isle Culbin
B1 1–1.2 1961–1990 0.5 1.5 5.5 9.5 7.5 14.5
2040–2050 11 14 14
2070–2080 13 17 16
1.2–1.5 1961–1990 21.5 30.5 19.5 25.5
2040–2050 32 43 30 33
2070–2080 35 50 32 33
A1F1 1–1.2 1961–1990 0.5 1.5 5.5 9.5 7.5 14.5
2040–2050 10 26 11 26
2070–2080 11 20 16 27
1.2–1.5 1961–1990 21.5 30.5 19.5 25.5
2040–2050 10 32 34 29 32
2070–2080 12 12 33 40 30 31

Two of the Caledonian Pine Forest Reserves (Glenmore and Glen Affric) showed only small increases in the probability of Seljaninov coefficients in the critical range, rising from less than 5 per cent in the baseline climate period to 9 and 12 per cent for the B1 and A1FI scenarios in 2070–2080 climate projections. However, for the Great Glen Caledonian Pine Forest Reserve, there was an increase from 21.5 to 33–35 per cent probability of Seljaninov coefficient in the range 1.2–1.5 by the 2070–2080 period, in both the B1 and A1FI scenarios.

## Discussion

### Outbreaks, population density dependence and climate drivers

Previous work (Lesniak, 1976; Sierpinska, 1998) has stated that the Seljaninov coefficient values between 1.0 and 1.5 can be used to indicate where damaging outbreaks of D. pini might be expected in Poland. Our analysis indicated a mean Seljaninov coefficient of 1.2 for the southern states of the north-east German lowlands. This is consistent with the previous studies (Lesniak, 1976; Sierpinska, 1998) that a value of 1.2 indicates optimal climatic conditions for outbreaks. Our study shows a greater clustering of Seljaninov coefficient values close to the centroid mean value of the PCA in outbreak years, and also for two consecutive pre-outbreak years, suggesting that two or perhaps three consecutive years with Seljaninov coefficients close to 1.2 will provide the climatic conditions necessary for an outbreak. The stochastic nature of the climate with a regular occurrence of more extreme climatic variables from year to year, represented by a shift of the Seljaninov coefficient out of the critical range 1.0–1.5, will prevent an outbreak from occurring. Therefore, for the German southern states, although the extrinsic climatic driver for D. pini outbreaks occurs in an average year, outbreaks do not occur every year because the climate conditions vary about the mean, and outside the critical Seljaninov range 1.0–1.5.

In large species of Lepidoptera, studies have shown a strong temperature-dependent relationship with realized fecundity (Berger et al., 2008). Warm day and night time temperatures during larval development and pupation produce large females capable of maturing and laying a large number of eggs. For D. pini outbreaks to occur, the two (or three) successive years with Seljaninov coefficients in the range 1.0–1.5 may provide ideal climatic conditions for the development of bigger females with increased fecundity and higher fitness (Boggs, 1986; Berger et al. 2008).

In our findings, there were differences in the climatic drivers during outbreak and pre-outbreak years between the northern oceanic states of Lower Saxony and Mecklenburg-Vorpommern, and the southern continental states of Saxony-Anhalt and Brandenburg. In particular, the mean temperature in March, when caterpillars emerge from the forest leaf litter, was warmer in southern states during outbreak years. In the northern states, a lower rainfall total in the late instar period (March–June) was important for an outbreak to occur but less important during the early instar period (August–October). In the northern states, a below average Seljaninov coefficient was required in two to three successive years to trigger an outbreak, whereas in the southern states, two to three years with average Seljaninov coefficient values could trigger an outbreak. This demonstrates that currently the climate of the northern states is ‘marginal’ for regular D. pini outbreaks, whereas the climate in the southern states supports a core area for D. pini outbreaks to occur more frequently.

The correlation between climate and outbreaks in the southern states of the north-east German lowlands indicates synchronous D. pini population dynamics among the P. sylvestris forest stands to climatic drivers, a response termed the Moran effect (Hudson and Cattadori, 1999). In the southern states, the insect population responds to ‘average’ climatic conditions for the region. The large population increase leading to an outbreak and capable of causing extensive defoliation is likely to require two or three consecutive years of ideal climatic conditions. This has been observed for Bupalus piniaria (Straw, 1996), and modelled for Panolis flammea (Watt and Leather, 1987). Dendrolimus pini does not disperse widely, because the females are large heavy insects with limited flying ability (Varley, 1949; Majunke, 2000). Thus, insect dispersal is unlikely to be the cause of widespread synchronous outbreaks. We found that the Seljaninov coefficient varied spatially by only 0.4 units within a year for climate stations up to 200 km apart, and this small variation demonstrates very similar climatic conditions across the region. The outbreaks reported in the north-east German lowlands, therefore, demonstrate synchrony to the ‘extrinsic’ annual climatic driving mechanism (Varley and Gradwell, 1960).

In the absence of factors such as predators and parasites, there must be a widespread synchronous and substantial increase in the population of D. pini present in P. sylvestris plantations across the region, and the population dynamic response is rapidly triggered by very suitable climatic conditions to cause an outbreak. Climatic variability may also have a subtle effect on P. sylvestris plantation ecosystem processes. In studies of forest pests, many have shown complex interactions between: prey and predators and parasites (e.g. for D. pini, see Glowacka-Pilot, 1974), plant and herbivore interactions (Jactel and Brockerhoff, 2007) and feedbacks and population cyclicity linked to climatic drivers (Haynes et al., 2014).

Majunke et al. (1999) reported observations made by Schwerdtfeger (1949) that especially high summer temperatures during the caterpillar period in spring and autumn exacerbated the onset of damaging outbreaks. Higher sugar levels occur in the needles of P. sylvestris stands that have been subject to drought stress in dry summers (Schwenke, 1978) have been implicated in the cause of outbreaks of D. pini. Indeed, Sierpinska (1998) concluded from her review that climatic factors do not affect D. pini directly, but indirectly through changes in the host plant. The summer heat and dryness of average climatic conditions may cause physiological stress to P. sylvestris (Schwenke, 1978; Sierpinska, 1998). When growing in pure even-aged stands, the population density-dependent site and stand factors may operate with climatic factors to cause an increase in the population of D. pini over two (or three) successive years, leading to an outbreak.

Our results strongly suggest that one pre-outbreak year with optimal conditions is likely to drive the population dynamics of the insect to cause an outbreak. However, it is very likely that two or more generations with increased fecundity, fitness and survival would be required to cause a large population increase, capable of an outbreak, and large-scale defoliation, and our PCA results suggest this could be the case. It is also likely that even-aged single-species forests of P. sylvestris in the north-east German lowlands have a naturally high population of D. pini, and it is the climatic extreme years that are either wetter and colder (giving Seljaninov values more than 1.5) or drier and hotter (giving Seljaninov values less than 1.0) that actually prevent more regular outbreaks.

It is also interesting that outbreaks have occurred more widely in Germany, further north in the north-east lowlands (Majunke, 2000), and in the past further south in Bavaria (Zwölfer, 1962; Haynes et al., 2014). However, Haynes et al. (2014) found a lower frequency of outbreaks in Bavaria in years and decades featuring unusually warmer temperatures, and their analysis suggested that D. pini outbreaks followed drought. Their finding seems a little odd since warmer years, and particularly those with unusually warmer temperatures, tend to be dry droughty periods. The literature suggests that in wetter and cooler summers, the impact of pathogens and parasites maintains a lower population density with fewer outbreaks (Schwerdtfeger, 1970; Majunke et al., 1999). We speculate that this process may be more prevalent in the more oceanic northern states, and it probably maintains a less frequent occurrence of outbreaks in the southern states than would otherwise be the case if there was less variability about average summer climatic conditions.

### Success of outbreak predictions

The decision-tree analysis provides a new method to assess when a D. pini outbreak may happen. The decision tree indicated several threshold climatic values, during pre-outbreak and outbreak years, in order for an outbreak to occur. The observed outbreak record indicated that if any one of these conditions was not met, then an outbreak did not occur. The climatic rules within the decision-tree model, therefore, provide the detail of the climatic conditions that cause the Seljaninov coefficient to be within or exceed the critical range for an outbreak to occur.

By monitoring standard rainfall and air temperature measurements, and using these data to calculate the climatic variables through the D. pini caterpillar development period, the model predicted outbreaks with an accuracy of 64 per cent at the end of March when caterpillars normally emerge from winter hibernation. The accuracy increased to 74 per cent by mid-summer if spring and early summer mean temperature was greater than 9.4°C (although outbreak damage would already be occurring). Importantly, the decision-tree model predicted all the actual outbreak events, plus a small number (five) of false-positive events. This suggests that if the method was combined with winter hibernating caterpillar sampling, better predictions of damaging outbreaks could be made, providing forest managers with some advance warning of likely damage, and some early evidence to intervene to stop the outbreak, thus helping to reduce the risk of total defoliation in stands of P. sylvestris.

The decision-tree data mining procedure was limited by two factors: a small dataset prevented the partition into training and validation data, and the narrow spatial extent of outbreak area data was limited to the continental states of the north-east lowlands. With a longer record and broader spatial area of data, the decision-tree approach might be improved to fit rules relating to critical climate periods in the years prior to outbreak, and it may be possible to explore in greater detail the effect of two pre-outbreak years with good climatic conditions for an outbreak.

### Predictions and implications

Assuming that the Seljaninov coefficient range 1.0–1.5 is a reasonable indicator of where D. pini outbreaks may occur, prediction of the changing climate space where this is likely to occur across Europe is useful. However, uncertainties associated with climate projections exist due to limited knowledge about the climate system. To better acknowledge the degree of uncertainty, we used two Global Climate Model projections: HadCM3 and ECHAM5. Climate change projections suggest warmer and drier summers in Europe (IPCC, 2007a, b). These projections are expected to cause a northward migration of invertebrates (Sparks et al., 2005) and forest pests (Ruckstuhl et al., 2008). Therefore, more frequent dry and warm summers could be advantageous for D. pini if not too warm and dry.

The projections of climate change with more suitable climate conditions for D. pini outbreaks to occur in Europe suggest that outbreaks could occur in the Baltic States and southern Scandinavia, as in the northern states of north-east German lowlands. Indeed, the first serious outbreak of D. pini occurred in Lithuania in 1993 (Gedminas and Ziogas, 2008), with a 2-year pre-outbreak period and an outbreak covering 28 000 ha forest prior to pesticide control. In Sweden, a ‘rare’ outbreak was recorded (Björkman et al., 2013) on the Stockholm archipelago in 2012. There is an implication here of differences in the outbreak ecology of D. pini between ‘core areas’ (e.g. currently the southern states of the north-east German lowlands) and the ‘range margins’ (e.g. currently the northern states of the north-east German lowlands) of the pest, where the climatic drivers are slightly suboptimal for more frequent outbreaks to occur.

### Future implications of D. pini in Scotland

Our analysis of probabilistic climate data showed that for two of the three more westerly and inland Caledonian Pine Forest Reserve sites, the current and the projected summer climate – as reflected in the probability distribution of the Seljaninov coefficient – would remain wet and cool through the twenty-first century. For the Great Glen Caledonian Pine Forest Reserve, the climate will change sufficiently for a substantial increase in likelihood of years in the Seljaninov coefficient range between 1.0 and 1.5, and particularly between 1.2 and 1.5 in an A1FI emission scenario.

Values of the coefficient in the more easterly, lowland plantation forests of P. sylvestris are expected to become drier and warmer, with higher probabilities of Seljaninov coefficient values associated with D. pini outbreaks. In particular, for Beauly Forest, the probability of values between 1.2 and 1.5 is projected to rise by 20 per cent from baseline to 2080 climates for the A1B emission scenario, and from 9.5 to 20 per cent in the 1.0–1.2 Seljaninov range in the A1FI emission scenario. This is less extreme than the southern states of the north-east German lowlands where there is a more regular occurrence of very warm dry summers (Werner and Gerstengarbe, 2007). Surveys in Scotland suggest that the D. pini population is semi-voltine requiring 2 years for caterpillars to mature, and evidence from the north-east German lowlands suggests that this may continue since the current and projected climate in north-east Scotland seems too variable to provide environmental conditions for more rapid univoltine development.

### Pests, parasites and pathogens

Since the climate of the north-east German lowland states is more variable with cooler and more humid years, the impact of pathogens (and parasites) on caterpillar development may be greater. The wetter winter and spring soil conditions, and cooler spring weather have an influence by increasing parasites and pathogens of D. pini, and partly from the abiotic effects on mortality from cooler conditions (Füldner, 2001) (as well as extremely hot conditions). The effect of entomogenous bacteria and fungi in regulating D. pini populations in Poland have been studied (Glowacka-Pilot, 1974) and found to have a significant impact in reducing the population size.

### Adapting forest management to reduce impacts

The adaptation of forests to reduce the impacts of climate change is topical in European forest policy and practice (Lindner et al., 2010). Studies over 60 years ago by Weckwerth (1952) suggested that a practical method of reducing the impact of D. pini outbreaks could be achieved by mixing other tree species into P. sylvestris plantations. He reported that the caterpillars of D. pini will climb all the trees in a stand but prefer only to feed in the P. sylvestris canopies. The caterpillars waste energy and are exposed to predators and pathogens by this activity. Mixed stands and an increased structural diversity (Bernadzki et al., 1986) may help to reduce the extent of damage of D. pini outbreaks, although this presents a more challenging form of silvicultural management on infertile sites (Zerbe, 2002). Haynes et al. (2014) reported no outbreaks in Bavaria since the 1930s, and they concluded that the Bavarian summer climate was probably too warm for outbreaks of D. pini. Our results suggest a similar finding with a range of climatic tolerance for outbreaks, not too hot and dry, and not too cold and wet, where the Seljaninov coefficient range is 1.0–1.5. However, in Bavaria, there is evidence (Borchert, 2007) of widespread transformation from even-aged monocultures to mixed-species, mixed-age structure stands. This will have reduced the population density-dependent factors by increasing the range of tree species and niche types, and broadened the range and number of competitors, predators, pathogens and parasites of D. pini (Jactel and Brockerhoff, 2007). This may limit population growth, keeping numbers at lower levels, and thereby reducing D. pini below a damaging outbreak density. The introduction of birch (Betula pendula), rowan (Sorbus aucuparia) and other species suited to infertile sites may help improve the fertility of P. sylvestris plantations and provide other benefits, such as biodiversity, recreation and landscape aesthetics. Bernadzki et al. (1986) strongly advocated this type of multipurpose forestry to reduce pest and pathogen impacts, and also to provide an increased range of goods and services from woodlands. Scotland's Ancient Caledonian Pine Forest Reserves include a mix of species with P. sylvestris, and this may further reduce the risk of future defoliation by D. pini in ancient woodlands. Single-species P. sylvestris plantations or those with lower tree species diversity at a lower elevation in the east of Scotland (and other areas of Europe) will be gradually exposed to a higher risk of damage through climate change over the coming decades. The level of risk may be reduced by the introduction of other tree species into P. sylvestris stands on infertile and freely draining mineral soils.

## Conclusions

Values of the Seljaninov coefficient in the range between 1.0 and 1.5, associated with outbreaks of D. pini in P. sylvestris plantations on sandy textured soils of the north-east lowlands of Germany, are very similar to the critical values used to define areas of severe D. pini outbreaks in Poland.

Even-aged plantation monocultures of P. sylvestris on freely draining low fertility sites exert a very uniform site and niche structure, habitat that is very well suited to D. pini populations. Across the P. sylvestris forests of the southern states of the north-east German lowlands, summer continental climatic conditions are consistently warm and dry, typified by a mean Seljaninov coefficient value of 1.2. It seems likely that these factors frequently combine in at least two (and sometimes three) successive years to cause widespread D. pini outbreaks.

In the southern states of the German lowlands, our decision-tree model using easily accessible climatic observations predicted with good accuracy when a damaging outbreak event was about to happen. This could be developed, parameterized and tested further with a longer time series of data and with outbreak data from other parts of Europe.

Downscaled high-resolution climate projections showed an enlarged area in Europe where Seljaninov coefficients in the critical range 1.0–1.5 may occur in the future. Affected countries include northern Germany, Poland, the Baltic States, southern Finland and Eastern Sweden. This may herald a future increase in the frequency of outbreaks of D. pini in the P. sylvestris forests of Scandinavia and around the Baltic Sea.

In north-east Scotland, the small recently discovered population of D. pini may extend its range across eastern lowland P. sylvestris plantations. Dendrolimus pini could spread to some Ancient Caledonian Pine Forest Reserves on warm and dry sites. However, most semi-natural pinewoods are less likely to be damaged as they are composed of a mix of tree species, which have been shown to be less vulnerable to D. pini outbreaks.

By 2080, climatic conditions in the coastal lowland margins of north-east Scotland will change to experience a higher probability of years (from ∼1 in 10 to 4–5 years in 10) with a Seljaninov coefficient in the range 1.0–1.5. This climatic pattern is more similar to the current northern, more oceanic, climate of the north-east German lowlands. Dendrolimus pini outbreaks may occur in the future in Scotland, but these are likely to be rare as is currently the case in Lower Saxony and Mecklenburg-Vorpommern.

There is an urgent need to adapt woodlands to the combined impacts of pests and pathogens and climate change, including the recently discovered population of D. pini in Scotland. For D. pini, this could be achieved through managing mixed-species stands.

None declared.

## Funding

The work was funded by the Forestry Commission (GB) Climate Change Programme.

## Acknowledgements

We wish to acknowledge the work of D. pini entomologists in Germany and Poland, and particularly the data figures and photographs provided by J. Gräber, K. Möller and R. Kätzel. We also thank Chris Quine and Eckehard Brockerhoff and Kevin Chase and two anonymous reviewers for very constructive comments.

## References

Allen
R.G.
,
Pereira
L.S.
,
Raes
D.
,
Smith
D.
1998
Crop evapotranspiration guidelines for computing crop water requirements
.
FAO irrigation and drainage paper no. 56
.
FAO
.
Altenkirch
W.
,
Majunke
C.
,
Ohnesorge
B.
2002
Waldschutz auf Ökologischer Grundlage
.
Eugen Ulmer
.
Berger
D.
,
Walters
R.
,
Gotthard
K.
2008
What limits insect fecundity? Body size- and temperature-dependent egg maturation and oviposition in a butterfly
.
Funct. Ecol.

22
,
523
529
.
E.
,
Kowalski
M.
,
Szujecki
A.
1986
Problems of Utilisation of Lower Deciduous Layers in the Protection of Pine Stands
.
Warsaw Agricultural Univerisity Press
.
Björkman
C.
,
Lindelöw
Ä.
,
Eklund
K.
,
Kyrk
S.
,
Klapwijk
M.J.
,
Fedderwitz
F.
et al
.
2013
A rare event – an isolated outbreak of the pine-tree lappet moth (Dendrolimus pini) in the Stockholm archipelago
.
Entomol. Tidskr.

134
,
1
9
.
Boggs
C.L.
1986
Reproductive strategies of female butterflies – variation in and constraints on fecundity
.
Ecol. Entomol.

11
,
7
15
.
Borchert
H.
2007
Veränderung des Waldes in Bayern in Den letzen 100 Jahren
.
Bayerisches Staatsministeriumfür Ernährung
.
Drobyshev
I.
,
Niklasson
M.
2004
Linking tree rings, summer aridity, and regional fire data: an example from the boreal forests of the Komi Republic, East European Russia
.
Can. J. For. Res.

34
,
2327
2339
.
Dronin
N.
,
Kirilenko
A.
2008
Climate change and food stress in Russia: what if the market transforms as it did during the past century?
Climatic Change

86
,
123
150
.
Dronin
N.
,
Kirilenko
A.
2011
Climate change, food stress, and security in Russia
.
Reg. Environ. Change

11
,
167
178
.
Food and Agriculture Organisation (FAO)
.
2010
Global Forest Resources Assessment 2010
.
World Forestry and Agriculture Organisation (FAO)
.
Füldner
K.
2001
Entwicklungserfolg von Kiefernspinner (Dendrolimus pini Linnaeus, 1758: Lepidoptera, Lymantriidae) An Douglasie (Pseudotsuga menziesii), Fichte (Picea abies) und Kiefer (Pinus sylvestris) unter Laborbedingungen. Allgemeine Forst und Jagdzeitung (Allg. Forst JZtg)
.
174
,
84
88
.
Gedminas
A.
,
Ziogas
A.
2008
The influence of Dendrolimus pini L. outbreak on the surrounding stands and forest litter entomofauna
.
Acta Biologica Universitatis Daugavpiliensis

8
,
287
296
.
Glowacka-Pilot
B.
1974
Entomogenous bacteria and fungi occurring in the caterpillars of the pine moth (Dendrolimus pini)
.
Prace IBL

427
,
3
60
.
Gräber
J.
,
Ziesche
T.
,
Möller
K.
,
Kätzel
R.
2012
.
AFZ-Der Wald

9
,
35
38
.
Haynes
K.J.
,
A.J.
,
Klimetzek
D.
2014
Forest defoliator outbreaks under climate change: effects on the frequency and severity of outbreaks of five pine insect pests
.
Global Change Biol.

20
,
2004
2018
.
Hijmans
R.J.
,
Cameron
S.E.
,
Parra
J.L.
,
Jones
P.G.
,
Jarvis
A.
2005
Very high resolution interpolated climate surfaces for global land areas
.
Int. J. Climatol.

25
,
1965
1978
.
Hubalek
Z.
,
Horakova
M.
1988
Evaluation of climatic similarity between areas and biogeography
.
J. Biogeogr.

15
,
409
418
.
Hudson
P.J.
,
I.M.
1999
The Moran effect: a cause of population synchrony
.
Tree

14
,
1
2
.
IPCC
.
2007a
Summary for policymakers
.
In Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Fourth assessment Report of the Intergovernmental Panel on Climate Change
.
Parry
M.L.
,
Canziani
O.F.
,
Palutikof
J.P.
,
Linden
P.J.v.d.
,
Hanson
C.E.
(eds).
Cambridge University Press
, pp.
7
22
.
IPCC
.
2007b
Summary for policymakers
.
In Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change
.
Solomon
S.
,
Qin
D.
,
Manning
M.
,
Chen
Z.
,
Marquis
M.
,
Averyt
K.B.
et al
. (eds).
Cambridge University Press
.
Jactel
H.
,
Brockerhoff
E.G.
2007
Tree diversity reduces herbivory by forest insects
.
Ecol. Lett.

10
,
835
848
.
Kaulfuß
S.
2012
The monitoring and prognosis of insect pests: early recognition, monitoring, analysis and action. Wald Wissen.Net, Informationen für die Fortspraxis -Forest Pest Handbook
.
http://www.waldwissen.net (accessed on 24 May, 2012)
.
Klemola
T.
,
T.
,
Ruohomäki
K.
2008
Fecundity of the autumnal moth depends on pooled geometrid abundance without a time lag: implications for cyclic population dynamics
.
J. Anim. Ecol.

77
,
597
604
.
Koenig
W.D.
2002
Global patterns of environmental synchrony and the Moran effect
.
Ecography

25
,
283
288
.
LAGB
.
2012
Übersichtskarte der Böden (BÜK400d)
.
Landesamt für Geologie Bergwesen Sachsen-Anhalt
.
Lesniak
A.
1976
Climatic and meteorological conditions of the pine moth (Dendrolimus pini L.) outbreaks
.
Ekol. Pol.

24
,
515
547
.
Liedekerke
M.V.
,
Jones
A.
,
Panagos
P.
2006
ESDBv2 Raster Library – a set of rasters derived from the European Soil Database distribution v2.0 European Commission and the European Soil Bureau Network, CD-ROM, EUR 19945 EN
. .
Lilly
A.
,
Bell
J.S.
,
Hudson
G.
,
Nolan
A.J.
,
Towers
W.
2010
National Soil Inventory of Scotland: site location, sampling and profile description protocols (1978–1988), NSIS1
.
Technical Bulletin
.
Macaulay Institute
.
Lindner
M.
,
Maroschek
M.
,
Netherer
S.
,
Kremer
A.
,
Barbati
A.
,
Garcia-Gonzalo
J.
et al
.
2010
Climate change impacts, adaptive capacity, and vulnerability of European forest ecosystems
.
Forest Ecol. Manage.

259
,
698
709
.
Majunke
C.
2000
Die Massenvermehrung des Kiefernspinners (Dendrolimus pini L.) in Brandenburg- Analyse Der Witterung in der Progradation
.
Mitt. Dtsch. Ges. Allg. Angew. Entomol.

12
,
75
78
.
Majunke
V.C.
,
Möller
K.
,
Walter
C.
1999
Massenvermehrung des Kiefernspinners im Land Brandenburg
.
AFZ-Der Wald

7
,
364
367
.
Mellec
A.L.
,
Habermann
M.
,
Michalzik
B.
2009
Canopy herbivory altering C to N ratios and soil input patterns of different organic matter fractions in a scots pine forest
.
Plant Soil

325
,
255
262
.
Mellec
A.L.
,
Karg
J.
,
Bernacki
Z.
,
Slowik
J.
,
Korczynski
I.
,
Krummel
T.
et al
.
2011
Effects of insect mass outbreaks on throughfall composition in even aged European pine stands – implications for the C and N cycling
.
Earth Sci. Climatic Change

1
,
6
.
Möller
K.
,
Engelmann
A.
,
Eberswalde
L.
2007
Die aktuelle Massenvermehrung des Kiefernspinners, Dendrolimus pini (Lep., Lasiocampidae) in Brandenburg
.
Mitt. Dtsch. Ges. Allg. Angew. Entomol.

16
,
243
246
.
Müller
H.J.
1992
Dormanz bei Arthropoden
.
Elsevier
,
289 pp
.
Murphy
J.
,
Sexton
D.
,
Jenkins
G.
,
Boorman
P.
,
Booth
B.
,
Brown
C.
et al
.
2009
UK Climate Projections Science Report: Climate Change Projections
.
.
NLFB
.
1999
Bodenkundliche Übersichtskarte von Niedersachsen und Bremen 1: 1:50,000 (BUEK50)
.
Niedersachsen Landesamt für Bodenforschung
.
NW-FVA
.
2010
Waldschutz Information 2010: KieferngroBschädlinge und Nonne
.
Abteilung Waldschutz
,
4
pp
.
Panagos
P.
,
Liedekerke
M.V.
,
Jones
A.
,
Montanarella
L.
2012
European Soil Data Centre: response to European policy support and public data requirements
.
Land Use Policy

29
,
329
338
.
Prescott
T.
2009
Pine-tree lappet Dendrolimus pini – friend or Foe
.
Atropos
,
37
,
3
10
.
Ruckstuhl
K.E.
,
Johnson
E.A.
,
Miyanishi
K.
2008
Introduction. The boreal Forest and global change
.
Phil. Trans. R. Soc. B

363
,
2245
2249
.
Schwenke
W.
1978
Die Forstschädlinge Europas
.
Paul Parey
.
Schwerdtfeger
F.
1949
Kampf dem Kiefernspinner; EinfüHrung in die Lebensweise und Bekämpfung des Kiefernspinners (Dendrolimus pini L.)
.
Neumann
.
Schwerdtfeger
F.
1970
Die Waldkrankheiten
.
Paul Parey
.
Sierpinska
A.
1998
Towards an integrated management of Dendrolimus pini L
. In
Proceedings: Population Dynamics, Impacts and Integrated Management of Forest Defoliating Insects
.
McManus
M.L.
,
Liebhold
A.M.
(eds.).
USDA Forest Service
.
Sparks
T.
,
Roy
D.
,
Dennis
R.
2005
The influence of temperature on migration of Lepidoptera into Britain
.
Global Change Biol.

11
,
507
514
.
S.J.
2005
Aridity indexes
. In
Encyclopedia of World climatology
.
Oliver
J.E.
(ed.).
Springer
.
Straw
N.A.
1996
The impact of pine Looper moth, Bupalus piniaria L. (Lepidotera: Geometridae) on the growth of Scots pine in Tentsmuir Forest, Scotland
.
Forest Ecol. Manage
.
87
,
209
232
.
ter-Braak
C.J.F.
,
Smilauer
P.
1998
CANOCO Reference Manual and User's Guide to Canoco for Windows: Software for Canonical Community Ordination (Version 4)
.
Micocomputer Power
.
Varley
G.C.
1949
Population changes in German forest pests
.
J. Anim. Ecol.

18
,
117
122
.
Varley
G.C.
,
G.R.
1960
Key factors in population studies
.
J. Anim. Ecol.

29
,
399
401
.
Watt
A.D.
,
Leather
S.R.
1987
Pine beauty moth population dynamics: synthesis, simulation and prediction
. In
Population Biology and Control of the Pine Beauty Moth
.
Forestry Commission Bulletin 67
.
Leather
S.R.
,
Stoakley
J.T.
,
Evans
H.F.
(eds).
HMSO
.
Weckwerth
W.
1952
Der Kiefernspinner und seine Feinde
.
Geest & Portig
,
40 pp
.
Werner
P.C.
,
Gerstengarbe
F.-W.
2007
Welche Klimaänderungen sind in Deutschland zu erwarten?
In
Der Klimawandel – Einblicke, Rückblicke Und Ausblicke
.
Endlicher
W.
,
Gerstengarbe
F.-W.
(eds).
Potsdam-Institut für Klimafolgenforschung und Humboldt-Universität zu Berlin
, pp.
56
59
.
Zerbe
S.
2002
Restoration of natural broad-leaved woodland in Central Europe on sites with coniferous forest plantations
.
Forest Ecol. Manage.
,
167
,
27
42
.
Zwölfer
W.
1962
Über Abwehreinrichtungen unserer Waldbäume gegen Insektenschäden
.
J. Appl. Entomol.

51
,
364
370
.