Discovery of continental-scale travelling waves and lagged synchrony in geometrid moth outbreaks prompt a re-evaluation of mountain birch/geometrid studies

The spatio-temporal dynamics of populations of two 9-10 year cyclic-outbreaking geometrids, Operophtera brumata and Epirrita autumnata in mountain birch forests in northern Fennoscandia, have been studied since the 1970´s by a Swedish-Norwegian research team and, during the last decade, by Norwegian and Finnish research teams. Some of the early results have been challenged by the Norwegian team. To examine the base for disagreements, five of the papers published by the Norwegian team (2004-2011) are reviewed. It is found that conclusions in these papers are questionable or data could not be interpreted fully because two decisive traits in the spatio-temporal behaviour of outbreaks of the two species were not considered.


Introduction
Two recent observations of the spatio-temporal outbreak pattern of some extensively studied geometrid moth species necessitate a re-evaluation of previously published findings. Firstly, it has been demonstrated that defoliating outbreaks on deciduous trees of the winter moth (Operophtera brumata L.) and associated geometrids every 9-10 years travel as a wave across Europe from east to west with the front of the wave stretching from high to low latitudes 1 . Secondly, long-term studies on the temporal behaviour of Epirrita autumnata, an associated geometrid, and O. brumata in the mountain chain of Scandinavia (the Scandes) and northern Finland show that outbreaks of O. brumata have been synchronized with those of E. autumnata with a time lag of about 1-2 years [2][3][4] .
Temporal dynamics of populations of cyclic geometrids have been studied in some parts of central Europe (e.g. Raimondo et al. 2004 5 ; Glavendekić 2002 6 ; Kula 2008 7 ). However, research on spatiotemporal behaviour seems to have been undertaken only in Fennoscandia (Norway, Sweden and Finland), i.e., on O. brumata, E. autumnata and Agriopis aurantiaria on birch (Betula pubescens and B. p. czerepanovii=mountain birch). These studies started with a historical survey of outbreaks on birch of E. autumnata and O. brumata that covered the years 1862-1968 in the geographical area of Fennoscandia 8 . It revealed that 12 outbreak periods had occurred during the surveyed period, with an average interval of 9-10 years. The survey was followed-up for the years 1969-2001 9 , adding three more outbreak periods, this time with quantitative data (including for A. aurantiaria) for the period 1990-2001 3 . In these studies, it was found that outbreaks sometimes occurred contemporaneously along the Scandes and sometimes moved as a wave from north to south along the Scandes or from south to north. Although true within the scope of these studies, it has now been demonstrated that the patterns described are illusions caused by continental-scale outbreak waves passing the Scandes either in parallel or obliquely from north or south 1 . Another study, also restricted to Fennoscandia, showed that waves of E. autumnata outbreaks during the four outbreak periods from 1970-2004 have, on average, travelled across Fennoscandia from about NE/E to SW/W 1 . This aligns with the results in   1 .
In   1 , it is argued that local and even regional population dynamics of a species cannot be properly understood if large-scale waves in cyclic populations occur without them being recognized. The prime examples of such misinterpretations on a regional scale are the descriptions of outbreak waves in the abovementioned papers 8,9 . The purpose of this paper is to expose further examples of similar shortcomings.
Commentary on five papers regarding the local and regional spatio-temporal population behaviour of O. brumata and E. autumnata In 1999, a Norwegian project was launched to cover one cycle of the population ecology of O. brumata and E. autumnata in northern Fennoscandia. The project has hitherto (2013) produced 22 papers (http://www.birchmoth.com/) of which nine focus on local and regional spatio-temporal population behaviour of mainly O. brumata.  15 ). E. autumnata was omitted from the study because of insufficient numbers in this coastal area. In an early study 8 based on historical documentation, it was suggested that climatic forcing (the Moran effect) may have synchronized the 9-10-year cyclic outbreaks of O. brumata (and E. autumnata) occurring in the Scandes and northern Finland. The actual study raised the question as to whether this suggestion withstands closer scrutiny in terms of more detailed quantitative data. Commenting on the short duration of the study, it is said "...although studies of dynamics of cyclic populations for many purposes need to span a longer time period than the time period of a cycle (i.e., 10 years for the focal moth species) the degree of synchronous dynamics can be evaluated from more short termed data (.....)", and, "for cyclic populations, identifying and comparing the phase among sites at any given time would be sufficient to verify the prevalence of synchrony. Here we use the latter approach on our 4-year data set...".
The female O. brumata moth has stunted wings and cannot fly. Physical barriers, such as sea water between mainland and islands at the coast, may therefore prevent the spread of females and isolate island populations. On the understanding that the climate is identical, differing population dynamics on the mainland and nearby islands should imply the disruption of the Moran effect. Six pairs of sites were selected from east to west in the north-western coastal part of Troms, each with one site on the mainland (or a large island) and one on a Changes from Version 1 I thank the reviewers for their comments. I have made the following two corrections to my article: 1) Introduction, second paragraph, last line. Klemola et al. (2006) has been changed to . It was found that population densities were low on all islands, except for one (site 2, Tussøy), and that populations there were maximally out of phase with peaking populations on the mainland, or populations that had recently peaked, except for one (site 1, Sandvik). The authors concluded that "...the distinct asynchrony between adjacent sites (<10 km apart) clearly belonging to the same climatic domain rules out the possibility that climate or, for that matter any other supposedly large-scale phenomenon (.....) could be responsible for the population phase differences".
Thus, the notion of a large-scale spatial synchrony due to a Moran effect was rejected, at least for coastal birch forest.
Later, it was shown 3 that a large-scale E. autumnata/O. brumata outbreak wave arrived at northern Fennoscandia from the east in 1991-1992. The wave travelled westward across the region and reached the eastern part of Troms in 1994-1995, and finally the west coast of Troms in about 2000 where the mainland/island study was performed. It was stated in Tenow et al. (2007) 3 that this wave must be considered when the results of that study are to be interpreted.
Since then, Jepsen et al. (2011) 15 have added seven more years to the data set, which now comprises the years 1999-2009. This offers an opportunity to re-evaluate the 2004 results. The low density island sites, seen in Figure 1 in the paper, may be subdivided into three groups from east to west within the investigation area (cf. enlargement of population curves for 1999-2002 in Jepsen et al. 2011, Figure 3, and comments to that paper below 15 ): one group consisting of the easternmost site and the two northernmost sites that experienced an evident low during the study time (1999-2002) ( Figure 1b: sites 8, 10,12) and one group of the next two island sites to the south-west that underwent the greatest crash phase during the same period (sites 4, 6). Finally, the south-western-most island site (2) exhibited a full peak and a crash. The mainland sites may also be divided into three groups; the easternmost site with an evident low during the study (11), the two northernmost sites with a crash (7, 9; although site 9 shows a deviating population curve), and the three south-western-most sites with a full peak and a crash (1,3,5). This demonstrates that the wave of the 1990s, which arrived to eastern Troms in 1994-1995 and then continued westward, had already passed across the northern part of the investigation area before the study, i.e., about the middle of the 1990s. Then, in 1999-2002, the wave was recorded at the south-western island sites in its crash phase and finally, in 2000-2004, it was caught up lingering at the south-western-most mainland sites in its peak and crash phases. Hence, by first reaching the island sites in the north in the direction E-W, then the island sites along the west coast in the direction N-S, and finally the mainland sites inside the islands, it behaved like a wave "breaking" successively over the mainland part of the investigation area ( Figure 1b). Of the six pairs of sites, only one (9, 10) indicates populations that may have been maximally out of phase-one branch of the wave passed south of the investigation area already in 1994-1995 3 .
In conclusion, there were no distinct, maximally out of phase of population dynamics between sites on mainland and nearby islands. Hence, the Moran effect cannot be discarded. Instead, there was a continuous progression over the investigation area of the large-scale wave of the 1990s, first via the island sites then to the mainland sites. Therefore, at any given time, populations separated in space in the direction of the wave had been in continuous change in larval density. Furthermore, the continuation of the study (cf. Jepsen et al. 2011 15 ) clearly demonstrates that a four-year study in this case was not sufficient to allow a correct interpretation. The paper gives detailed data on spatio-temporal changes in larval abundance of O. brumata and E. autumnata in vertical and horizontal directions on a coastal mountain slope (Figure 1b: site 9, Reinøya) in 2001-2007. It is assumed that the air temperature gradients decrease linearly with increasing altitude whereas the temperature is approximately the same in a horizontal direction at each altitude. It is predicted that the varying sensitivity of different insect life stages to climate can cause a non-linear forcing on population synchrony along altitudinal air temperature gradients on slopes. If there is a nonlinear forcing in this case, a synchronisation caused by a Moran effect could be rejected. It is stated that the `global spline correlogram´, applied for each species, was based on all population time series (i.e., [2001][2002][2003][2004][2005][2006][2007]. In the beginning of the results section, the authors stated: "The 7-year time series of both species exhibit sufficient temporal variation to provide an useful basis for investigating patterns of population synchrony (….). That is, all populations had crashed at the end of the study, enabling us to highlight synchrony of the crash phase of the population cycle".
It is shown that E. autumnata went through all stages of a regular cycle (increase, peak, and crash during 5 years, followed by a low) during the 7 years. The changes of the O. brumata population showed more variation but the overall trend was less steep and often lagged behind E. autumnata. There is no mention of cycle phases other than the crash phase (O. brumata), nor mention of successive population cycles or of outbreak waves (cf. Tenow 1972 Figure 4 in the paper that the spatiotemporal population behaviour during the same phases was similar for the two species. These circumstances cast doubt on the claim that there was an "anisotropic patterned population synchrony in climatic gradients" for O. brumata. The "target" concept is mentioned in the paper; implying that the target, the Moran effect, will be aimed at with precision and accuracy. However, the precision of advanced statistical methods is in vain when accuracy is poor, as can be suspected in this case. The study did not account for the cycles in two successive outbreak waves that passed the site. This makes it open to questions. Does this omission imply that the analysis of the O. brumata data became biased and that the result therefore is unreliable? This study monitored the defoliations of mountain birch forests across northern Fennoscandia (Figure 1a) in 2000-2008. This was achieved by satellite imagery and the different contributions from the two geometrids E. autumnata and O. brumata were not specified. One result identified was that the defoliation had spread north-eastward along the mountain chain. Thereby, the conclusion in Tenow et al. (2007) 3 is challenged namely that outbreaks, like that during the preceding period 1990-1999, obey a consistent pattern by moving in a wavelike fashion broadly from east to west. Instead, it is stated that "Even without a formal analysis of the spatial dynamics of defoliation, our data indicate that this is not so. Future analysis of the spatio-temporal dynamics of defoliation patterns will substantiate this".
In any case, here the existence of unidirectional outbreak movements in northern Fennoscandia seems to be accepted (cf. below).  13 . The satellite mapping did not discriminate between E. autumnata and O. brumata defoliations because, again, their different contributions could not be separated. The authors applied two analytical approaches to the spatio-temporal defoliation data. The first approach was focused on population synchrony, namely to assess how synchrony depends on distance. Synchrony can take the form of phase-dependent synchrony, meaning that the degree of synchrony may differ between the increase, peak and crash phases (here evidently implying that phases for the two species coincided). The second approach focused on the rate of spread of defoliation. It was found that defoliation could be divided in three phases. Thus, the analyses comprised estimation of the area of defoliated forest during each of the incipient (2000-2002), epidemic (2003)(2004)(2005) and crash (2006-2008) phases ( Figure 2 in the paper). The analyses revealed that the incipient phase was characterized by high regional synchrony and long defoliation spread distances per year, whereas the epidemic and crash phases were characterized by much lower regional synchrony and a much shorter spread of defoliation. This suggests two independent spreadsteps, one long-range process and one very short-range diffusionlike process. As for defoliation, the incipient phase of the outbreak showed a higher regional synchrony in the onset of plant growth (budburst) than the later phases of the outbreak.
There is no discussion about potential outbreak waves crossing the region (cf. Klemola. 2006 10 ;Tenow et al. 2007 3 ). It is assumed (in the Introduction) that the spatio-temporal synchrony of large-scale outbreaks may be explained by a putative regionalized Moran effect, e.g. a match/mismatch between moth and host phenologies "Such a phenological mismatch-driven Moran effect could also be responsible for the more complex dynamics recently reported for birch forest moths (Klemola et al. 2006;Tenow et al. 2007)".
We now know that geometrid outbreak wave travelled from southeastern Europe to the Atlantic coast in the 2000s and in the 1990s (as well as in earlier decades 1 ). It is evident from comments to papers above that both waves crossed the region. During the investiga- A time lag between the defoliations is not discussed despite being acknowledged in previous papers 12,17 . Instead, it is stated that E. autumnata and O. brumata ".... exhibit largely synchronous dynamics, with winter moth dominating at termination of the outbreaks...".
The lag between species was distinct for the Varangerfjord area to the east (high populations 4 ). Also, in the western part (Figure 1b), a time lag occurred between the two species (low populations) at site 11 (Lyngenfjord) 18 and was repeated one year later at site 9 (Reinøya) to the northwest 16 .
Because of the wave and the lag, the phase-dependent outbreak dynamics depicted was in reality two separate dynamics, one for each outbreak species. The incipient phase (2000)(2001)(2002) Figure 3 15 ) during the crash phase. In this area, the satellite did not record any extensive defoliation despite severe outbreaks. The topography is steep and the forest is fragmented. Because of this, the area of O. brumata defoliation is most likely to have been underestimated in the western coastal region compared with the area in the Varanger region with its smoother topography and more continuous forests 13 . This is supported by reports on large O. brumata outbreaks across northern Norway in 2006 17 , i.e., in areas that were closer to the coast than the E. autumnata outbreaks (cf. Jepsen et al. 2008 19 ). When summarising the time scales, it is obvious that E. autumnata dominated defoliation that moved progressively from east to west during the incipient and epidemic phases (2002)(2003)(2004)(2005) and O. brumata, with some overlap, in a similar way during the crash phase (2005)(2006)(2007)(2008), all according to the time lag between the species. Thus, neither dynamic fitted the applied presumed periods of the phase-dependent synchrony.
According to the definition of a wave, defoliations that were widely separated in space can hardly have been contemporaneous during the incipient, epidemic and crash phases. In addition, these phases are split in one and the same area due to the time lag between species. Therefore, it cannot be claimed that the spread of defoliation was determined by a stationary, step-wise process, one incipient long-range step and one later short-range.
The onset of plant growth (budburst) had a phase-dependent pattern of spatial synchrony during the incipient phase that was similar to that in the occurrence of defoliation (above; although with a much higher level of spatial synchrony for the former). This suggested to the authors that spring phenology plays a decisive role in the synchrony of moth outbreaks, i.e., an indirect Moran effect. However, contrary to defoliation, spring and budburst do not arrive to the region as a wave moving broadly from east to west. In line with a much higher level of spatial synchrony for budburst than for defoliation (above), budburst may occur over large areas at the same time and, probably, does not have much to do with the degree of defoliation (cf. Bylund 1999 20 ;Nilssen et al. 2007 9 ). Defoliation events in northern Fennoscandia (as well as in Fennoscandia as a whole 8 ), are cyclic and tightly connected to the outbreak waves that pass broadly from east to west approximately every 9-10 years 1,3,10 . Why these waves travel as they do and if/how they force cyclicity upon local populations is not known. What can be said is that there is no known direct or indirect effect of climate forcing on any match/mismatch of plant/insect synchrony that, firstly, can cause a 9-10-year cyclicity locally and, secondly, travels across Europe from east to west every 9-10 years 1 . In addition, in northern Fennoscandia such an effect should act twice, first on E. autumnata and then on O. brumata, with a lag of 1-2 years. Should the match/ mismatch apply only to E. autumnata or to both species?
In conclusion, the division of outbreak dynamics in phases with different synchrony and spread speed must be questioned. Instead, it is the wave and the time lag that are important. Furthermore, contrary to what is suggested, there can be no climate-mediated Moran effect that drives the 9-10 year passages of continental travelling waves over the region and/or drives the local 9-10-year population cycles. Taken together, comments on the papers above reveal that waves of O. brumata population peaks have travelled broadly from east to west in northern Fennoscandia. This is further substantiated by the population curves depicted in Figure 3 in the paper. There is only one comment on the representation of O. brumata in these curves "Curiously, O. brumata displayed a second, much smaller, peak in abundance during the years and sites where A. aurantiaria was most abundant (2005)(2006)(2007)(2008)(2009))".
In fact, Figure 3 in the paper catches two successive travelling waves, the waves of the 1990s and the 2000s, which together shed light on the "curious" peak. The O. brumata population curves for the eastern-most and western-most sites reveal this most clearly. At the eastern-most inland pair of sites (11, 12) and the north-easternmost island sites (8, 10), the wave of the 1990s had already passed at the time of the study (cf. Ims et al.  Figure 3 in the paper). Hence, during the study period, the two waves crossed the area in continuous movements from east to west, first the wave of the 1990s, then that of the 2000s. The interval between the peaks was short, about 7-9 years compared with the overall average of 9-10-years; however, short intervals have not been uncommon in northern Fennoscandia 8,9 .
In conclusion, as demonstrated with the "curious" O. brumata peak as an example, it is not possible to interpret long-term data fully if the continental-scale outbreak waves of O. brumata (and associated geometrids) that pass northern Fennoscandia are neglected or not known (however, see Hagen et al. 2010 18  The study of forest insect population dynamics has a long and storied history, and many forest insect species have served as model systems for studying conceptual processes of population biology and ecology. This is particularly the case for two geometrid species undergoing cyclical populations in Fennoscandia. A recent paper by  , recommended in F1000Prime ) revealed the et al. presence of regular travelling waves for the winter moth ( ) across the European Operophtera brumata continent, providing evidence of the importance of continental processes in understanding regional and local dynamics of cyclical populations.
In this commentary paper, Tenow contends that due to the presence of these continental-scale travelling waves, other studies within this system that do not account for these patterns could be in fact missing an important driver of population dynamics. Tenow critiques five papers to challenge their respective conclusions.
This commentary should stimulate a thoughtful dialogue by all involved in these studies, which undoubtedly would a benefit to others involved in the study of insect spatial and temporal dynamics.