OUP UNCORRECTED PROOF FIRST PROOF, 03/13/10, SPi. SECTION 4 Biological consequences of climate change

Transcription

1 SECTION 4 Biological consequences of climate change INDD 87 3/13/2010 7:07:36 PM

2 INDD 88 3/13/2010 7:07:36 PM

3 CHAPTER 9 Changes in migration Esa Lehikoinen and Tim H. Sparks 9.1 Introduction Bird migration has created interest for hundreds of years ( Stillingfleet, 1762 ). Early writers regarded migration as an instinctive, innate process. From historical data we know also that the timing of migration varies, the basic principle being that in early, warm springs migrants arrive early and vice versa. Exact arrival date varies in all migratory species, but less in those species that arrive late when weather is less variable, and which are suggested to be under more strict endogenous control. The birds ability to respond to variations was expected to be sufficient in relation to the interannual variability in spring weather. On departure timing, again the common belief has been that warm autumn weather caused delays to departure. Analysing data for extreme individuals (first to arrive and last to leave) gives hints of changes, but to get a more robust understanding the total migration (arrival, passage, and departure distributions) or some point estimates of distributions need to be studied. Diving into the depths of all kinds of data on migration and the impacts of climate on them has been popular in recent years. It has also revealed clear responses and a wide spectrum of variation in responses. Only 9 years ago Møller ( 2001 ) wrote Whether spring observations of migratory birds also have become earlier as the spring temperature has increased in temperate Europe remains to be determined. Within a few years, the responses of arrival and departure timing were analysed and significant advancement of spring arrival was recorded almost globally. These changes in bird migration were included in many general reviews. The credibility of advancing trends in spring was strongly supported by climatic relationships (temperature, North Atlantic oscillation (NAO), and other measures of weather) and the fact that advancements were observed in areas where climate warming was fastest. Since our previous review (Lehikoinen et al., 2004 ), many papers have confirmed the general results. Researchers have measured the strength of response and background of its variation within and among populations more carefully by studying the whole migration, and suggested likely mechanisms by which earlier arrival is achieved. The study of consequences of changing migration schedules on other life-cycle stages has also intensified. Despite increasing activity, we still have only fragmentary understanding of these processes. Many generalizations tend to rely on only a few studied species or on superficial analyses of timing changes in large groups of species. Since Lehikoinen et al. ( 2004 ), an increasing number of case studies and useful review articles have appeared (e.g. Gordo, 2007 ). A solid platform on which to build new research aimed at solving remaining questions is now established. In this chapter, we briefly discuss how changes may be achieved, look at the ways in which a bird species can change its migration timing, consider sources of data in migration timing and their limitations, and examine the evidence for changes in timing. We critically examine what environmental factors are associated with changed timing and briefly discuss research needs for the future INDD 89 3/13/2010 7:07:36 PM

4 90 EFFECTS OF CLIMATE CHANGE ON BIRDS 9.2 Migration in the annual cycle of birds Migration is a return trip of individuals that have separate breeding and wintering areas. Arrival and departure fit into the annual cycle in such a way that breeding, moult, and wintering phases are optimized, but allowing for many constraints. Research emphasis has been in migration to and from the breeding area, but migration in relation to the wintering area is also important for fitting the annual cycle to the environmental conditions. Annual cycles vary among species and geographical regions, and within species many annual cycle types can occur, from nearly sedentary to longdistance migration. The timing, length, and existence of each phase and life-history stage can be affected by climate change ( Coppack and Both, 2002 ). Directional change of environment, for example climate warming, can alter selection from stabilizing to directional, and the strength of selection may change. The direction of selection can change in either way, and individuals may also become released from strong selection effects due to climate warming. Hypotheses on what will happen to bird behaviour and population dynamics if the predicted warming takes place are based on selection theory, but the difficulty is that selection effects are extremely hard to estimate, especially when behavioural traits, such as timing of migration, are concerned. It is not fully known which aspects of migration are under genetic control. We do know that the direction and initiation of migration in, for example, garden warbler Sylvia borin and blackcap Sylvia atricapilla are under genetic control ( Berthold et al., 2003 ). A large degree of adjustment to changing conditions can take place by behavioural means, for example learning from own experience ( Møller et al., 2009 ) and from other individuals can be important. A specific problem for endogenously strongly controlled life cycles is that they probably respond slowly to changes, particularly in species that have long generation times. This is because evolutionary adjustment takes place by generations, while behavioural adjustments take place on a daily basis if allowed by reaction norms. Reaction norms even buffer the genetic composition of populations against changes. If the warming trend continues as predicted, then limits to plasticity may be exceeded and the fate of the population will be determined by the level of genetic variation available. Many recent papers dealing with long-distance migrants and their response to, and ability to cope with, future changes have regarded endogenous control in these species to be very strict. From this, it is predicted that long-distance migrants will experience greater difficulty from climate warming ( Berthold, 1998 ). However, physiological (neuroendocrine and endocrine) plasticity allows non-photoperiodic cues to modulate timing to enable individuals to cope with climate variability (Dawson, 2008 ; Wingfield, 2008 ). It would be wrong to think that we know which species are, and which are not, strictly controlled. It is important to further remember that plasticity itself is under selection. Furthermore, variability in environmental conditions maintains variation in the trait values among populations, and there will often be genotypes in the population that are a better fit to the changed conditions so long as this is within previously experienced limits. Regrettably, in field conditions we can seldom measure the reaction norms exactly because the response of a given individual is always an additive result of reaction norm and an environmental effect. Therefore, a tendency to contrast plastic response with evolutionary change has emerged. Some authors have concluded that observed changes of migration times already indicate evolutionary change (e.g. Jonzén et al., 2006 ), but this conclusion seems too hasty ( Pulido, 2007 ) and not based on empirical evidence ( Chapter 12 ). Theoretically, it is expected that under directional selection increased plasticity is favoured ( Garland and Kelly, 2006 ). This may mean that observing selection on a trait value (e.g. arrival date) is hampered by concomitant selection on plasticity. In fact, Gienapp et al. (2008 ) and Sheldon ( Chapter 12 ) showed, mainly concerning changes in timing of breeding, that there are hardly any proven cases of evolutionary change. There is, however, no reason why evolutionary change of migration strategies has not already resulted from the moderate levels of climate change that we have so far experienced. Micro-evolutionary changes are inevitable, since plasticity, learning, and maternal effects may not be sufficient to withstand predicted strong and INDD 90 3/13/2010 7:07:36 PM

5 CHANGES IN MIGRATION 91 rapid changes. Experimental evidence for the potential of evolution is strong for insectivorous passerines ( Berthold et al., 2003 ; Newton, 2008 ), especially through studies with Sylvia warblers. An obstacle to definitive results is that proof of evolutionary change requires data on individuals and their progeny as well as fitness measures in the wild ( Pulido, 2007 ). Such data for timing of migration will be sparse until modern technological tools generate adequate samples. 9.3 Ways of changing migration timing Theory suggests that the timing of the full migration cycle is optimized to the benefit of the individual bird. The migrant individual is expected to optimize time, energy, and/or safety ( Newton, 2008 ). These aspects of migration affect each other, so that optimizing one aspect is constrained by the other two. Similarly, migration can be optimized as one stage of the annual cycle only relative to other stages. Timing and duration of spring and autumn migration should not put the bird at excessive risk of death by starvation, predation, or parasitism, and the time spent in wintering and breeding grounds should be fine-tuned to maximize survival and reproduction. Migration has evolved as a strategy to exploit or avoid different environments at different times of the year. To further complicate matters, in some environments a species may be a partial migrant, consisting of both migratory and resident birds. The proportion of a population that is migratory may be influenced annually by local conditions; adverse weather can encourage movement of more birds to milder environments. Where winter resources allow, birds may avoid migration ( Newton, 2008, pp ). Early arrival to breeding areas is theoretically a benefit both for inter- and intra-specific competition for territories ( Newton, 2008 ). However, risk of prebreeding mortality will increase the earlier the bird arrives. Thus, the optimal response cannot be as extreme as the shift in food peak date would require (Jonzén et al., 2007 ). Earlier arrival can increase the duration of the potential breeding season and allow additional clutches for birds capable of raising more than one brood, or replacement clutches should nest failure or predation occur, but this prediction may hold only in higher temperate regions, where too hot and dry summers are not a limiting factor, as in Mediterranean climate zones. Several options to advance arrival date exist ( Coppack and Both, 2002 ): increased migration speed, earlier departure, and shortening of migration distance. Speed can be increased in several ways, and departure of some but not all individuals is an alternative to general change, which we should also consider. Similarly, when looking at migration distances, we need to also carefully look at the distribution of distances and not only point estimates, such as the average distance. Spring and autumn migration may differ in their optimal solutions, and therefore predictions about how they will change are not as straightforward as often presented. Different life-cycle types can also bring about differences between seasons, for example the location of annual complete moult in the annual cycle can affect the response types Departure time from the wintering ground Species that winter in temperate regions are affected by more or less similar large-scale weather systems as in their breeding areas. Therefore, it is more likely that they will respond to climate change by starting spring migration earlier. Arctic geese and waders are examples of the importance of conditions when starting spring migration or continuing it from stopover sites. In capital breeders, sufficient remaining reserves after arrival are also a necessity. In bartailed godwits Limosa lapponica, the highest quality individuals can select the best possible migration strategy, which is not the earliest possible departure, but a later one when they have achieved large reserves. Pink-footed geese Anser brachyrhynchus in their last stopover site in Norway, on the other hand, are currently encountering problems with feeding conditions because of a decrease in per capita food availability ( Drent et al., 2003 ). A continuous chain of suitable stopovers is thus crucial for advancing migration. The situation is different in the tropics. Since photoperiod, the factor controlling the endogenous rhythm in long-distance migrants, follows the same annual pattern, this would suggest that departure INDD 91 3/13/2010 7:07:36 PM

6 92 EFFECTS OF CLIMATE CHANGE ON BIRDS will occur at the same time each year. Departure under endogenous control is sometimes interpreted unconditionally, which cannot have been the original intention. Even in the tropics, conditions vary and birds must be able to take this into account (e.g. Saino et al., 2007 ). Departure timing is also influenced by habitat quality ( Studds and Marra, 2005 ) in areas with little photoperiodic variation Range shift of the wintering area Range shifts could be studied by winter bird atlases and censuses, but this form of citizen ornithology is not that common and poorly covers important winter ranges of migrants in many flyways. Analyses of winter recovery distributions can be effective in species that are ringed and recovered in high numbers and that have approximately constant recovery probabilities. Winter bird censuses in North America and Northern Europe indicate range shifts northwards: a quantitative analysis by La Sorte and Thompson ( 2007 ) calculated an average shift northwards of km/year in a set of 254 species in North America. Three recent analyses of recovery distributions have been done with climate change impacts in mind ( Siriwardena and Wernham, 2002 ; Fiedler et al., 2004 ; Visser et al., 2009 ). A number of noise factors unavoidable in long-term recovery data were identified in these studies, but, by analysing subsets of selected groups of species and recovery types, use of recovery data is possible. The three studies included 82 species in all. The distribution of species in change classes suggests that at least some migrants have shortened their migration in the last half century ( Table 9.1 ). The results were similar for studies of black-headed gull Larus ridibundus and blackbird Turdus merula of the 13 species included in all three studies, and for a further five species (buzzard Buteo buteo, kestrel Falco tinnunculus, oystercatcher Haematopus ostralegus, common gull Larus canus, and song thrush Turdus philomelos ) two of the three studies showed significant shortening and the third showed no change. In the Dutch database, average arithmetic migration distance (standardized to year 1984) ranged from 12 to 537 km, and rather similar species sets were used in the other two studies, i.e. only intra-continental shortand medium-distance migrants yielded sufficient numbers of recoveries. One-third to one-half of these have shortened their migration in this sample. In none of the studies was direction of migration analysed. The recovery databases for long-distance migrants are generally very scanty. It is not possible to confirm the conclusion of Visser et al. (2009 ) that these analyses explained the difference in advancement rates between short- and long-distance migrants. Despite this, the potential for further analysis is evident. A further complication in interpreting these results is that no attention to the migration direction was made. Range shift rates measured from winter bird censuses in North America ( km/year) and from recovery data in Europe (1.2 km/year) should be positively related to advancement rates (on average 0.2 days/year). The small winter range shifts can explain only a small proportion of the advancement of spring migration, but these have not yet been compared formally. Working with averages may, however, be misleading if ranges of winter distributions are also widening when a proportion of individuals respond strongly. We also need information on possible widening of winter ranges. Birds may also change their sedentary mode of life. The likelihood for a species to become seden- Table 9.1 Changes of migration distance in three European bird ringing recovery studies. Migration distance Siriwardena and Wernham (2002) migrants Siriwardena and Wernham (2002) sedentary Fiedler et al. (2004) Visser et al. (2009) Decreased No change Increased Classifi cation according to reported statistical signifi cance INDD 92 3/13/2010 7:07:36 PM

7 CHANGES IN MIGRATION 93 tary when (winter) climate ameliorates is probably inversely related to average migration distance, but constrained by food availability. If rising temperatures also improve food availability, this constraint is relaxed. It is worth observing that in analyses of migration timing, species that are changing from migratory to sedentary mode provide additional difficulties. We should also study these species and the changes they are experiencing, since otherwise we will get a one-sided picture of events. Comparisons between species groups will be biased if arrival dates cannot be determined for some species or years. For long-distance European migrants optimizing latitudinal range changes may be impossible in Africa south of the Sahara. The Sahelian belt has become even more hostile during climate warming. Coppack et al. ( 2008 ) suggested that the pied flycatcher Ficedula hypoleuca could move its wintering distribution by no more than 10 since no good habitats occur further north in sub-saharan Africa. If the jump to winter in southern Europe becomes possible in large numbers, the situation will change. Indeed, first winter records of pied flycatchers around the Mediterranean have been reported ( ). There may be more opportunity for latitudinal shifts in short-distance migrants. We know hardly anything of interaction between genetical control of migration length and its timing. Therefore, both plastic and selective adaptive possibilities of long-distance migrants are difficult to determine Migration speed The progress of migration depends on flight speed and ratio of flight to stopover durations. Flight speed and orientation are a result of the workload the bird does to move in the correct direction and compensate for the effect of wind. Therefore, flight speed can increase, and longer travel distances are attained or the aimed target area reached with less effort with tailwinds of suitable strength. Migrant individuals attempt to select favourable flight conditions: tailwinds, more or less clear sky, and no rain ( Newton, 2008 ). Prevailing weather is the most important factor affecting take-off decisions of small passerines ( Dänhardt and Lindström, 2001 ). A proportion of the advancement of spring arrival may indeed be due to more favourable flight conditions during migration ( Sinelschikova et al., 2007 ). Migration proceeds faster across the USA with increasing temperature ( Marra et al., 2005 ). To be of real benefit, the constraining effect of phenology and food conditions at the next stopover or in the final goal area must also be relaxed. Therefore, not a single weather factor, but rather combinations of more widely favourable weather conditions might be responsible for speeding up migration. It may prove hard to disentangle effects of wind and temperature from one another. Southerly winds promoting northward migration are often associated in the Northern Hemisphere with higher temperatures. The dominance of temperature in bird migration studies is most probably due to the fact that it is more easily used as the explanatory variable than combinations of wind speed and direction. In order to study en route effects on migration speed and progress in detail, we should know the flyways birds are using. This is possible mainly for large day-flying species, for those capable of being satellite tracked, and for those where ringing effort has been so extensive as to generate large numbers of recoveries. In most other cases, flyways are only approximately known. Monitoring migration at migratory bottlenecks provides many opportunities to study both timing and meteorological effects on migratory decisions (e.g. in Israel at the crossroads of the Palaearctic-African flyway ( Shamoun- Baranes et al., 2006 ), or Gibraltar and Messina (Newton, 2008, pp )). The progress of migration can also be enhanced by gaining energy faster (at stopover sites) for migratory flights. Theoretical calculations suggest that small passerines are commonly able to achieve 200 km/day, in some cases even more ( Lindström, 2003 ). There is an upper limit to the rate of storing fat and increased speed of migration, and we do not know how close to these limits birds are currently performing. In larger non-passerines, which use power flight and extreme storage of fat, the whole migration can occur in a single or only a few flights. When the bird s physical condition and environmental conditions are suitable, flight can be rapid. For example, satellite-tracked bar-tailed godwits INDD 93 3/13/2010 7:07:36 PM

8 94 EFFECTS OF CLIMATE CHANGE ON BIRDS (body mass 630 g with maximally stored fat) made single non-stop journeys of ,000 km, covering at least 1000 km a day ( Gill et al., 2009 ). In such species, timing of migration is a direct function of pre-migratory fattening in wintering/breeding areas. In the majority of species, migration proceeds in many flights with events en route modifying the progress and affecting the timing of arrival to the goal area. Many studies show that there is much flexibility in fattening strategies ( Jenni and Schaub, 2003 ) and suggest that some level of adjustment to changing conditions is achievable. Stopover durations depend on phenological conditions and food availability, but may depend also on individual condition at arrival to stopover. There is some information suggesting that changes in stopover time are possible (Drent et al., 2003 ; Bairlein and Hüppop, 2004 ). We know less about what happens during spring migration. It seems to be generally true that spring migration is performed faster than autumn migration ( Newton, 2008 ). The reasons for this are not yet fully known, but there are several theoretical arguments for faster spring migration ( Chapter 13 ) Constraints to change of migration timing The optimal timing of migration in relation to both large-scale seasonal weather systems and emergence of food resources is of utmost importance, since a large proportion of annual mortality takes place during migration ( Newton, 2008 ). Any change of resource availability, related to climate change or not, will have consequences for timing the arrival to breeding quarters optimally. A welldocumented case is for red knots Calidris canutus, whose feeding conditions have been quickly deteriorating in some critical stopover sites. This is expected to have negative consequences for breeding and population maintenance. The crossing of ecological barriers such as the Mediterranean Sahara and Gulf of Mexico needs, in addition to attaining high resource levels, to be timed to avoid the likelihood of bad weather ( Jenni and Kéry, 2003 ). This may constraint some of the timing changes otherwise expected from climate warming and lengthening of the growth period alone, especially for autumn migration. 9.4 Is migration timing changing? Response variables Because we have only recently had the chance to map migratory pathways of individuals of mediumand large-sized bird species, we have to rely on other types of data to answer questions on changing timing of arrival and departure in longer time frames and in all small species. The best option is to use point estimates of arrival, departure, or passage of species at particular sites, where migration dates have been recorded for a sufficiently long period. These data can only reveal little on how advancements and delays are accumulating. It is clear that in this type of data, variation is present for a number of reasons, which fall into three categories: methodological, biological, and climatic. Some of the variance is noise and a part of this may be due to other trends that parallel climate warming and affect point estimates. It is important that these problems are taken into account, but it is as important that they do not prohibit us from analysing the patterns emerging. The arrival variables that have been used in investigations are many and diverse, and each may tell a different story. The message gained from multiple response variables can be more diverse than simple questioning of whether or not arrival has advanced. Types of phenological variables are summarized in Table 9.2. Lehikoinen et al. ( 2004 ) created some order in response variables. Yet many papers are not clearly expressing which of the many types of response variables they are using and what can we really conclude from the results ( Table 9.2 ). If our main goal is only to answer the basic question Are there changes in timing of migration?, the diversity of response variables used does little harm, although it makes meta-analyses difficult. If we want to go further and understand what the mechanisms behind the observed changes are, it is important to explicitly identify the possibilities and limits of response variables used. It is clear that different life-cycle phases (rows in Table 9.2 ) are affected by different factors, but the same also seems to apply to time points within each life-history stage. Timing of migration is observed at fixed localities, which are start, goal, or interval ( en route ) locations. This increases variability between both species INDD 94 3/13/2010 7:07:36 PM

9 CHANGES IN MIGRATION 95 Table 9.2 Phenological events in the annual cycle used as response variables to analyse changes in timing of spring arrival to the breeding grounds. Event in the annual cycle Code Resp 1 Resp 2 Resp 3 Resp 4 Resp 5 Arrival to breeding area ABA FAD MAD QADxx Departure from breeding area DBA LDD MDD QDDxx Arrival to wintering area AWA FAD MAD QADxx Departure from wintering area DWA LDD MDD QDDxx Observations en route OER FD MD QDxx PEAK LAD PEAK FDD PEAK LAD PEAK FDD PEAK LD An interpretation for classifying variants of response types used (Resp #). In different papers various codes have been used and in most analyses arrivals (and departures) to breeding or wintering areas and passage measurements are not separated. This diversity is one source of variability. ABA, arrival to the breeding area; DBA, departure from breeding area; AWA, arrival timing from the wintering area; DWA, departure timing from the wintering area; OER, observations en route; AD, arrival date; DD, departure date; F, fi rst; M, mean (or median); L, last; Q xx, quantile (at xxth percentile). Beyond this typology also average fi rst arrival (of n sites/observers) and the n th bird record, e.g. FAD 5 and FAD 10, have been applied to avoid the expected larger variance in directly observed FAD. and localities. In most published analyses, arrival dates are not identified exactly in relation to variable types in Table 9.2. In fact, we have very little real arrival to the breeding area (ABA) data, most being observations en route (OER) type of data. These individuals still have a chance to speed up or slow down their migration as a response to prevailing conditions. This may be a significant possibility if a considerable proportion of migration distance is still ahead. This difference in apparently equivalent data on arrival may be the reason why a few model species dictate the prevailing knowledge of responses to climate change, as the pied flycatcher is currently doing. We do not attempt to differentiate between ABA and OER in this article. The possibility of using different approaches, however, is advantageous, and we should not ignore citizen science data and rely only on model species, which can be studied in great detail. The fact that most variation in climate response is at the species level ( Rubolini et al., 2007 ) is, however, a strong argument for doing large-scale comparative analysis with less intensively collected data even if the sampling design is sub-optimal. Because all individuals have their own migratory calendar, a distribution of individual dates is formed and different parts of the distribution are observable with different accuracies. In goal areas the beginning (and also with less precision, the end) is easily observed, but the distribution of arriving and departing individuals is more difficult to obtain. En route whole distributions can be measured, but the populations at each location (observatory) are mixed, which raises other difficulties for analyses. If environmental changes in the goal area of each population differ, they can be reflected in timing observations at bird observatories along the migratory route. In addition, sex and age groups may depart, migrate, and arrive at different times ( Newton, 2008 ). These effects have been taken into account when the groups can be identified and distributions studied separately. The spread and form of the distribution of timing also need consideration. Symmetrical unimodal distributions are not the norm and various levels of skewness and kurtosis exist, and possible year-to-year variation and longterm trends need more investigation ( Sparks et al., 2005 ; Møller, 2008 ). The observation that early and late parts of the migration have changed differently may have more than one explanation, including recorder effort, bird population differences, age structure, and increasing skewness towards early arrival. Quantile approaches are useful for studying effects on the total distribution Sources of data Data can derive from a variety of sources, with very different duration of the time series, and very different real or perceived data quality (see Gordo ( 2007 ) for a thoughtful treatment of data quality variation) INDD 95 3/13/2010 7:07:36 PM

10 96 EFFECTS OF CLIMATE CHANGE ON BIRDS Phenological networks have operated in several countries in connection with meteorological services. The oldest of them were founded in mid-19th century, and Spanish and Estonian networks, for example, are still operating today. First arrival and last departure dates (LDDs) of easily identified species are the typical observations made. Some of these schemes obey strict rules, others are less standardized. County records are records, typically only of first arrival and last departures, collected by a number of observers in a geographical area. They are usually made by skilled amateur ornithologists during field trips. Whilst the area is usually fixed (there are incidences of boundary changes and amalgamations), the numbers of observers may vary from year to year. Where the resident human population is large and ornithology is popular, there has been an increase in recorder effort. Individuals records, typically of first arrival only, are collected by a single person, usually in a restricted area. Records may be inherited and continued by relatives. Durations of series may typically vary from 10 to 70 years. In contrast to data collected for interest only, there may also be a few schemes in which records are collected over a scientist s lifetime as part of an intensive study. Bird observatory data are collected at a single location or in a small adjoining area. Records may be made by paid employees or a collective of amateurs. Recorder effort may or may not be known and may or may not be variable. Records may be observations or records collected during ringing activities, or both. Where the locality of an observatory does not have a breeding population the information on all aspects of the passage distribution can be obtained, otherwise it may generate the full distribution of the arrival distribution (possibly mixed with passage birds) (usually OER). Nature reserve data may be of equal or better quality than observatory data, if they are based on standardized routines. They may even surpass observatory data because the response is more often of a known population (real ABA) Limitations of data First arrival data The detection of the first individual may be influenced by recorder effort (number of observers, number of hours, and size of recording area) and the behaviour of the bird species itself. For an obvious species with known habitat associations, for example sand martin Riparia riparia, recording should be reasonably reliable. For rare species, recording may be more problematic. Species undergoing population declines may become increasingly difficult to observe and therefore likely to be gradually recorded later ( Tryjanowski et al., 2005 ). Increases in recorder effort are likely to result in earlier observations, all other things being equal ( Figure 9.1 ). Some authors suggest that lack of (or reduced) phenological change in a species implies inability to cope with climate warming and hence population decline. If the phenological change is estimated from first arrival date (FAD) in a declining population, this may be a circular argument, but mean arrival date (MAD) should be unaffected by observational bias ( Møller et al., 2008 ) Total distribution The mean (median) passage or arrival time should be independent of recorder effort and bird population size, providing the recorded birds are a representative sample of the population. Other quantiles and summary statistics of the arrival distribution should also be useful in spite of multi-modality in many species. However, although passage/arrival distributions are often multi-modal because of sex, age, and population differences, a more comprehensive understanding of timing changes can be attained using them. One example of the relationship between first, median, and 5% and 95% quantile arrival dates is given in Figure 9.2. Such data are not entirely without problems even if the multi-modality is ignored. For example, where birds are resident in the vicinity of the bird observatory, there may be repeat observations of individuals or captures of birds that have been resident for some days. It is not always easy to separate spring and autumn passages, as the distributions may overlap. In such cases, there will be a need to separate distributions on an arbitrary date or by statistical methods. In some locations recording periods may be narrower than the arrival period, leading to truncation of records at either or both of the beginning and the end of the season. Jonzén et al. (2006 ) used fitted normal distributions to overcome truncation problems INDD 96 3/13/2010 7:07:36 PM

11 CHANGES IN MIGRATION a b 10 Migrants arriving (%) a' b' d Day Figure 9.1 The effect of detectability change on fi rst arrival date (FAD) and mean arrival date (MAD). The bell curves show a unimodal distribution comprising all individuals, e.g. passage through a bird observatory. Horizontal lines are detectability horizons before (solid line cf. solid curve) and after (dotted line cf. dotted curve) a 2-day shift in timing and an improvement in detectability. They describe a situation where the whole migration distribution shifted, but its form was unaltered. In this example, a true advancement is observed by MAD without error (b to a) when detectability improves (d), i.e. a higher proportion of birds is observed, but in FAD a proportion of the advancement estimate is arbitrary due to the change from b to a. In this example, FAD advances 4 days while MAD advances 2 days. The same principles hold for departure distributions Julian day (from 1 January) FAD QD 5 Median QD Year Figure 9.2 First arrival date (FAD), fi fth percentile (QD 5 ), median and 95th percentile (QD 95 ) of arrival of willow warbler Phylloscopus trochilus at Jurmo, Finland, Coeffi cients of variation indicate greater noise and larger variability in the early end-point estimates of the total passage distribution: FAD 4.0%, QD 5 3.3%, median 2.1%, and QD %, respectively INDD 97 3/13/2010 7:07:36 PM

12 98 EFFECTS OF CLIMATE CHANGE ON BIRDS Knudsen et al. (2007 ) have described mathematical tools to overcome distribution problems, but clearly more work has to be done in this area Last day of departure The small number of articles analysing LDD data is most probably caused by knowledge that many causes, other than just climate, can delay the delayed start of migration. In spring, the tail of the arrival distribution is often formed by non-breeders and birds in poor condition, and in autumn similar causes produce additional variability. To some degree, these may also affect the later quantile points. For the moment, we suggest that LDD observations should be treated with caution in climate impact studies The problem of series duration and period studied Bird migration phenology, and FADs in particular, can exhibit a lot of noise in time series (see Figure 9.3 for an example). Short time series have low statistical power to detect trends, and estimated trends may be heavily influenced by single-year values and/or short-term phenological reversals. In comparing series, even a 1- or 2-year difference in the end year can be very important. Ideally series should be complete, of the same duration and the same time period. This has not been the case even within a single study, to say nothing of differences between studies. To illustrate this point, we have taken a 57-year series of house martin Delichon urbicum FADs from Suffolk, UK ( Figure 9.3a ). This time series of FAD shows much variation from year to year and some exceptionally early observations. Calculating trends from all possible 10-year periods (i.e , , up to ), and similarly for 20-, 30-, and 40-year periods are shown in Figure 9.3b. Depending on which time slice is taken, the 10-year time series shows huge variations. A 10-year period to the mid-1960s would indicate a delay in arrival (trend is positive), as indeed would the most recent period ( ). The longer time series show much less fluctuation, and indeed the 20-, 30-, and 40-year series ending in 2006 produce similar trend estimates. What is clear is that a more reliable impression of phenological change can only be obtained from longer series (the big picture). Short series can produce unreliable estimates of trends. Sufficiently long series (20+ years) are even more important for bird migration phenology that appears more variable than other phenological series. Comparing trends from series of different periods should be undertaken with caution. We emphasized earlier that non-linearity is also an issue that should be taken into account ( Lehikoinen et al., 2004 ). In broad comparisons, however, analyses using shorter and identical periods with rather monotonic trends are easier to work on in comparative studies. There are other confounding effects, which have been dealt with variously in papers on phenological change. Tryjanowski et al. ( 2005 ) tried to control for effects of detectability and change in population size to better measure the effects of migration distance on FAD and analyse how well FAD is explained by MAD. For other confounding factors we refer to Lehikoinen et al. (2004, pp ) Changes in arrival timings to the breeding area The overwhelming balance of evidence in published studies suggests that migration is advancing. Yet there is great variation between studies and even between species within a single study. It was not possible to separate between ABA and OER time series. We have compiled a large but not comprehensive database derived from both published studies and some preliminary, as yet unpublished, data sets to which we had access. This has generated information on 3827 data series from 19 countries covering 455 species. There are limitations on the current state of the database. Median (10 90%) duration is 31 (21 53) years for FAD and 34 (18 46) years for MAD, with almost no missing values in different time series. Geographical distribution is heavily biased towards the western Palaearctic. Overall, the majority of trends were towards earlier arrival (82% of FAD trends and 76% of mean/ median trends were negative). Where statistical significance was reported, 47% of FAD series became significantly earlier and 5% significantly later. The INDD 98 3/13/2010 7:07:37 PM

13 CHANGES IN MIGRATION (a) Day of year Year (b) 2 10 years 20 years 30 years 40 years Trend (days/year) End year of series 2010 Figure 9.3 (a) First arrival dates (FAD) for house martin Delichon urbicum recorded in Suffolk, UK. (b) Trend estimates derived from linear regression for all possible periods of 10, 20, 30, and 40 years plotted against end year. respective figures for mean/median series were 40% significantly earlier and 2% significantly later. A summary of averages for trends in FAD is given in Table 9.3 for species with 20 or more time series available and separately for four geographical areas. Every overall mean value is negative, indicating earlier arrival and 18 out of 29 are statistically significant at P < 0.05 (Bonferroni corrected). Comparisons between countries should be done with caution; locations, species mixes, duration of series, and end years may all vary. Regional sub-division in Table 9.3 does, however, suggest more or less similar trend patterns, but different levels of trends in Europe, which is confirmed by a linear model on the subset INDD 99 3/13/2010 7:07:37 PM

14 Table 9.3 Average trends of FAD in species with at least 20 time series in our database (as of August 2009). Species N FAD mean SE UK n Germany n Finland n SE Baltic+ Russia n Delichon urbicum Acrocephalus scirpaceus Sylvia atricapilla Columba palumbus Fringilla coelebs Phylloscopus collybita Turdus philomelos Riparia riparia Alauda arvensis Motacilla alba Ficedula hypoleuca Locustella naevia Sylvia borin Sylvia curruca Sylvia communis Phoenicurus phoenicurus Apus apus Anthus trivialis Phylloscopus trochilus Hirundo rustica Phylloscopus sibilatrix Saxicola rubetra Muscicapa striata Sterna hirundo Oenanthe oenanthe Cuculus canorus Acrocephalus schoenobaenus Actitis hypoleucos Motacilla fl ava In order to illustrate the geographical distribution of the database, and its effect on overall trend estimates, regional estimates are also presented for four areas where data permitted. Species ranked according to overall advancement rate and Bonferroni-corrected signifi cance of FAD (both ABA and OER-FD data included) are shown in bold INDD 100 3/13/2010 7:07:37 PM

15 Frequency (%) Frequency (%) OUP UNCORRECTED PROOF FIRST PROOF, 03/13/10, SPi CHANGES IN MIGRATION 101 of UK, German, and Finnish data on 13 common species. The main effects of region and species were significant, but the interaction was not (country P = , species P = , interaction P = 0.32, d f country = 2, d f species = 12, d f interaction = 22, d f error = 359). It is clear that the overwhelming trend is towards earlier arrival, averaging 2.8 days/decade (SD = 5.9) earlier for first arrivals and 1.8 days/decade (SD = 3.0) earlier for the smaller number of results of trends in mean/median arrival in the whole database ( Figure 9.4 ). The trend rates differ between FAD and MAD. This is due to the larger standard deviation of FAD. The FAD trend rate very probably includes some as yet unknown proportion of methodical bias leading to larger variance and larger estimates. FAD and MAD similarity can be more effectively investigated in a subset of trend estimates where both exist for the same sites and periods. Samples from Jurmo, Finland (Lehikoinen and Rainio (unpublished data), all species and (a) > Change in arrival date (days/year) (b) > Change in arrival date (days/year) Figure 9.4 Distributions of trend rates according to (a) median/mean dates (MAD) and (b) fi rst arrival dates (FAD) over all species in our database. Black bars suggest advancing and white ones delaying trends of arrival/passage in arrival to the breeding area (MAD, FAD) /observations en route (OER) (MD, mean date; FD, fi rst date) types of data. Averages were calculated for those species for which more than one data point existed ( N for FAD 440 species, 3201 series and N for MAD 214 species, 440 series) INDD 101 3/13/2010 7:07:37 PM

16 102 EFFECTS OF CLIMATE CHANGE ON BIRDS passerines, which had complete data for ), and two data sets from Manomet and Mt Auburn, North America ( Miller-Rushing et al., 2008a, b ) showed that first and median arrival dates do correlate significantly over species (Jurmo, all species, r = 0.25, N = 111 species; Jurmo, passerines, r = 0.52, N = 51; Manomet, r = 0.29, N = 32; and Mt Auburn, r = 0.66, N = 17). FAD and MAD should also correlate within species over years if they measure the same response to some degree. Because responses of different populations and the two ends of the passage distribution can differ for a number of reasons, correlations need not be particularly strong. This was the case in three long-distance migrants (willow warbler, chiffchaff Phylloscopus collybita, and pied flycatcher) from five bird observatories (range , mean 0.38, 7 out of 14 statistically significant) ( Sparks et al., 2005 ) Changes in departure from breeding area We collected 683 series, of which 374 are LDDs and 150 are median departure dates (MDDs). The departure series represent 246 species from seven countries. The reservations and cautions expressed in the section on arrivals also apply to departures. Departure data are sparser than those for arrival and also more problematic. Most LDD time series have a trend close to zero with 60% of series becoming later. The pattern is about the same for 18 European species for which several time series were available ( Table 9.4 ). In contrast, only 36% of MDD series were getting later and the mean more obviously negative, although these were heavily influenced by the Swiss study of Jenni and Kéry ( 2003 ) and more long-distance migrant species. Of the LDD series quoting significance, 4% were significantly earlier and 14% significantly later. Too few series reported significance to justify presenting the equivalent figures for MDD. One prediction concerning autumn migration is that it could be delayed ( Berthold, 1998 ). Things are not that simple, however, since seasonal variation in temperature trends is considerable (see Figure 9.5 ) and annual cycles of different species may dictate other types of responses. Table 9.4 Mean last departure date (LDD) trends for species with at least eight time series in the database. Species N Mean SE Delichon urbicum Cuculus canorus Muscicapa striata Streptopelia turtur Phylloscopus trochilus Riparia riparia Apus apus Hirundo rustica Locustella naevia Phoenicurus phoenicurus Sylvia borin Acrocephalus scirpaceus Sylvia communis Oenanthe oenanthe Motacilla fl ava Sylvia curruca Acrocephalus schoenobaenus Saxicola rubetra Anthus trivialis Species ranked according to overall mean. Practically all data are from the UK. None of the mean LDDs is statistically signifi cant if Bonferroni correction is applied, but those of the wheatear Oenanthe oenanthe (P = 0.011), whinchat Saxicola rubetra (P = 0.018), and tree pipit Anthus trivialis (P = 0.017) are. For LDDs, at least, the influence of population trends and recorder effort may be influential. Another cause for concern for the robust analysis of LDD trends is that it is probably easier for odd individuals to stay and become observed in autumn as lingerers for climatic or condition-dependent reasons or because selection has not yet reduced genetical variance of the new generation. In many species, the migration strategy timing is age dependent and departure distributions are much more affected by this than spring arrivals (for example, see Anthes, 2004 ). The values of spring and autumn first/last observations may differ strongly in this respect, but also other point estimates should be measured by age groups if their timings differ. In any case, there is more reason than in spring to use quantile and median data from standardized observatories and sites. The lack of clear, statistically significant patterns may also have been a contributing reason for the paucity of published summaries of LDD data INDD 102 3/13/2010 7:07:38 PM