Importance of climatic and environmental change in the demography of a multi-brooded passerine, the woodlark Lullula arborea


Correspondence author. E-mail:


1.  We examined the influence of local weather conditions on reproductive success, timing of breeding and survival in a population of a multi-brooded ground nesting passerine (woodlark Lullula arborea) over 35 years.

2.  Woodlarks laid larger clutches when rainfall was low and temperature high during the egg-laying and pre-laying period. Nest success increased with higher temperatures during the nesting period. In successful nests, the number of chicks fledged per egg laid was greater when weather was drier during the brood stage.

3.  Although woodlarks bred earlier in years with warmer early spring temperatures, with the onset of breeding varying by 25 days, there was no significant advance in the onset of breeding over the 35 years of study, due to considerable inter-annual variability, and no overall trend, in weather.

4.  Simulation modelling of annual reproductive output demonstrated that earlier breeding could increase productivity by 23·5% in the warmest compared to the coldest year, due to birds having more nesting attempts. Other effects of weather on productivity affected breeding output to a lesser extent.

5.  Effects of weather on productivity were minor compared to an increased rate of nest predation through the period of study, which reduced productivity by 49·8% by 2004 compared to 1971.

6.  Turning points analysis identified three distinct demographic periods: from1971 to 1988 the population grew slowly, during 1988–1999 the population grew rapidly, but after 1999 the population declined. Increased population growth after 1988 was associated with higher first-year survival rates (estimated using a population model). Population decline after 1999 was caused by a combination of reduced productivity (resulting from increased nest failure rates attributed to predation) and lower first-year survival rates, that appear unrelated to winter temperature.

7.  Climate change (long-term changes in weather) did not explain the marked changes observed in the population trajectory over 35 years. We suggest that understanding effects of both climate and habitat change on populations is essential in predictive population modelling.


Climatic conditions, the prevailing weather patterns in an area over a long period, are a key factor affecting many ecological processes (Stenseth et al. 2002; Parmesan & Yohe 2003; Portner & Farrell 2008) and climate change will alter many species’ population dynamics (Root et al. 2003; Pounds et al. 2006; Hughes et al. 2008). Understanding the influences of weather on populations and predicting responses to future climate change is an urgent priority (Sæther, Sutherland & Engen 2004; Sutherland 2006; Sekercioglu et al. 2008). However, human activity has many other environmental impacts that may limit species’ potential to respond to climate change (Sala et al. 2000; Warren et al. 2001).

Weather variability has well-documented effects on bird populations; changes in survival rates caused by winter temperatures are key drivers of population dynamics in many species (Siriwardena, Baillie & Wilson 1998b; Grosbois et al. 2006; Robinson, Baillie & Crick 2007), while during the breeding season, weather affects both phenology (Crick & Sparks 1999; Both & Visser 2001; Reed et al. 2006) and breeding success (Møller 2002; Rodriguez & Bustamante 2003; Both et al. 2006). However, there is a need for greater understanding of the consequences for species’ population dynamics of these responses (Crick 2004; Both et al. 2006). For example, although changes in phenology in response to climate change can reduce breeding success by causing a mismatch between peak food demand and peak food availability (Thomas et al. 2001; Visser & Both 2005; Both et al. 2006), in other cases phenological responses can maintain synchrony (Cresswell & McCleery 2003). Multi-brooded species may be able to increase productivity by having more nesting attempts (Jenni & Kéry 2003), however few studies have tested whether this occurs (Wilson & Arcese 2003; Saino et al. 2004; Møller 2007).

We use data from a 35-year study of a ground nesting, multi-brooded passerine to examine effects that changes in patterns of weather and other environmental changes have had on population dynamics. The majority of studies examining effects of climate change on avian phenology and breeding success have focussed on single-brooded species, often in artificial nestboxes; however, responses of multi-brooded species may be very different. We explore the challenges and problems of assessing consequences of changing weather and environment on the productivity and survival of a multi-brooded species with few marked individuals. Our study investigates the following questions: (i) Does weather affect nesting success? (ii) Does early spring temperature affect the timing of breeding? (iii) Does weather affect annual productivity? (iv) How do responses to weather affect demography? (v) Is there evidence that climatic change has affected demography? (vi) What are the relative effects of other environmental changes, such as changes in nest predation rates, on long-term changes in demography?

Materials and methods

Study site and population

The woodlark, Lullula arborea, is a species of European Conservation Concern due to a widespread decline in population size and range (Burfield & van Bommel 2004). A woodlark population was studied between 1971 and 2004 in Thetford Forest (52°30′ N, 0°60′ W), which covers 185 km2 of Breckland, a biogeographic region of eastern England. The Breckland region supports 25–30% of the UK breeding woodlark population (Wotton & Gillings 2000), largely within Thetford Forest. The forest is divided into 16 geographical blocks (mean area 1174 ha, 548 SD), and comprises pine-dominated plantations managed by rotational clear-felling and replanting of even-aged stands, creating a mosaic of growth stages (Eycott, Watkinson & Dolman 2006). Woodlarks breed in clear-felled and re-planted stands with trees up to 9 years old, but most (98% of all woodlarks across all years) are found on stands less than 6 years old or areas of permanent open space (Wright 2006).

Woodlark territory surveys were conducted annually. Several visits were made to every patch of suitable habitat during March–May and numbers and locations of singing males recorded following Wotton & Gillings (2000). Nests were located by observation of adults. In repeat visits (mean interval 4·6 days ± 3·1 SD) nest status and numbers of eggs and chicks were recorded (following Crick, Baillie & Leech 2003) as well as notes on fledging, adult behaviour and signs of predation of failed nests. As human scent may provide either a cue or deterrent to mammalian predators, observers did not kneel at nests during monitoring. The methodology and frequency of nest monitoring has remained constant, with most nests monitored by a single individual (RAH). Frequency of nest visits did not affect nest survival rates of woodlarks in Dorset UK, when nests were monitored using the protocols of this study (Mallord et al. 2007b). Nests were assumed to be successful if chicks had reached fledging age (10–12 days), the nest was empty but intact with signs of success such as trampled droppings in or next to the nest, adults were alarming or feeding fledged young nearby, or fledged young were seen. If cold eggs or dead chicks were found in or near the nest we assumed the nest had failed due to desertion or chick starvation. We assumed the nest had failed due to predation if there was nest damage or if remains of eggs, young or adult birds were found nearby. Sometimes there were no obvious signs of predation but the clutch or brood disappeared from the nest before chicks had reached fledging age and the usual signs of a successful nest were not present. In these cases, we assumed that predation was the cause of nest failure but the predator had not damaged the nest. For all failed nests, we noted whether the nest lining was torn out or whether the nest cup was empty but intact.

Woodlarks are multi-brooded, making many repeated nesting attempts following nest failure and up to two further broods following successful fledging of an earlier brood (Mallord et al. 2008). Analyses here are based on a sample of nests located throughout the breeding season. Successive nesting attempts of known pairs were not monitored, so it was not possible to differentiate between first and later clutches.

From 1986 chicks were ringed with unique colour combinations at 5–8 days old. In each subsequent year, resightings of marked individuals were recorded.

Territory surveys were not completed in 2001, and nest monitoring and resighting of birds were reduced, as an outbreak of foot and mouth disease restricted access.

Data analyses

The woodlark population growth curve, measured by observed numbers of territorial males, was smoothed using a thin-plate spline with 11 d.f. in SAS proc tpspline (SAS Institute, Inc 1999). The second derivatives of the smoothed trend were used to identify turning points in the population trend (following Siriwardena et al. 1998a; Fewster et al. 2000; Robinson et al. 2004).

Nest success was modelled as the daily nest survival rate using logistic generalized linear models (GLMs) with binomial errors and logit link (Aebischer 1999), constructed using proc genmod in SAS and compared using likelihood ratio tests assuming a chi-squared distribution. Daily nest survival rate did not differ between clutch and brood stages (χ21 = 0·69, = 0·41) which were pooled in subsequent analyses. Nest survival was estimated over the entire nesting period of 28 days, comprising 3 days laying (modal clutch size 4, found in 67% of = 518 nests, laying one egg per day, with incubation commencing on the last day), 14 days incubation and 11 days to fledging (Cramp & Perrins 1988). First egg dates were observed, or estimated from clutch increments or hatch dates.

To investigate effects of weather during the breeding season on nest success, we tested the effects of mean daily minimum temperatures and total rainfall from 1 March to 30 June of each year. We also investigated the effect of weather during the period specific to each nesting attempt, using the mean of daily minimum temperatures and total rainfall from the date the first egg was laid in the nest to the date the nest either fledged or failed. Daily minimum (rather than maximum) temperatures were used as we assumed that in this location, at the north-western limit of their range, woodlark productivity was likely to be limited by the coldest not the warmest temperatures. Weather variables were measured at Santon Downham, located centrally in the study region (52°46′ N, 0°67′ W).

Whether the proportion of nests that failed with the lining removed, or failed with the lining left intact, changed through time was tested using GLMs with logit link and binomial error with each nest as a binomial trial.

Clutch size was modelled in relation to weather using GLMs with Poisson error, log link, a scale parameter estimated by the square root of Pearson’s chi-square/d.f. ratio to account for under dispersion of the data, and nest-specific weather covariates (means of daily minimum temperatures and rainfall during the egg-laying period and the 4 days prior to laying the first egg). All models controlled for significant effects of lay date by incorporating it as a covariate. The number of fledglings per egg was modelled using GLMs with binomial errors and logit link, in relation to means of daily minimum temperatures and rainfall during a 12-day period from the day prior to hatching (in case of error in estimated hatch date) to fledging.

The relation between timing of breeding and spring temperatures was investigated using linear regression. The start of the breeding season was estimated from the fifth, 10th and 25th percentiles of the distribution of first egg dates, calculated for all years with more than 21 nests found (= 18 years). Spring temperature for each year was measured as the mean of daily minima between 15 February and 15 March.

Models of reproductive output

Total annual productivity (the number of fledglings produced in one breeding season by one pair) could not be directly measured, as successive nesting attempts of individual pairs were not monitored throughout the breeding season (most adults are not marked, a new nest is constructed for each repeat attempt, and territories of neighbouring pairs may overlap or shift as the breeding season proceeds). Instead we simulated breeding activity of woodlarks throughout the breeding season to estimate mean annual productivity, using a similar modelling approach to Beintema & Muskens (1987), Bowden & Green (1992), Ratcliffe, Schmitt & Whiffin (2005) and Pearce-Higgins et al. (in press) (Supporting Information Fig. S1). This allowed estimation of annual breeding output, and assessment of the relative contribution of different components (variation in the timing of breeding, daily nest survival, clutch size and chick survival caused by weather and inter-year variation in daily nest survival caused by predation) to annual variation in total reproductive output.

The dates that woodlarks start and finish breeding, daily nest survival rates and clutch sizes under different conditions were known from our data. We measured the intervals between a nest either fledging or failing, and the start of the next nesting attempt, for a small number of intensively monitored pairs. For each modelled pair (each of which was considered for 1 year only with no pair-specific values carried over across years), unique values of each parameter were drawn at random from normal distributions based on means and standard deviations derived from empirical data, such that variation in the total productivity between modelled pairs should be representative of variation in total productivity between real pairs of woodlarks in the population. Values and derivation of model parameters are shown in online Supporting Information (Table S1). For each combination of year-specific parameters and each of 34 years (1971–2004), 10 000 breeding pairs were simulated to estimate the mean productivity of, and the variation between, pairs of woodlarks in each year.

The effect of change in nest success on productivity was assessed first without considering weather effects. This model incorporated stochastic variation in the mean first lay date, varying between pairs but not years, with an overall mean first laying date of 28 March. Further models assessed the effect of weather-induced variation between years in the timing of breeding, and effects of weather on nest success and clutch size. Here, the mean first lay date was predicted from spring temperatures in each year. For all models, the first lay date for each pair was drawn from a distribution around the predicted mean first lay date, with SD of 12 days (Supporting Information, Table S1).

The productivity model assumes that the relationship between first laying date and early spring weather derived from 1986 onwards also held for years prior to 1986, and that the mean date at which females stop laying is the same in all years. However, if inter-annual variation in the end of the breeding season covaries with the timing of the start of the season, or differs among periods, this could introduce systematic error into the simulation model results. To test this, we split years into three groups according to turning points in the population growth curve (1971–1987, 1988–1998 and 1999–2004) and between two groups (each of = 17 years) according to early spring temperature (with mean daily minima between mid-February and mid-March either <0·5 or >0·5 °C). Within the observed data, the numbers of nests laid in each week from 1st May onwards were compared among groups using GLMs with Poisson error terms and scale parameters estimated by the square root of the deviance/degrees of freedom ratio, with laying week as a covariate. In each instance, the interaction between week and the temporal group variable was examined to determine whether the timing of the tail of the laying date distribution differed among these a priori groups.

Survival models

Survival rates were modelled in two ways. Capture–mark–resighting data were used to model survival from 1986 onwards. This method could not be used to estimate survival for 1971–1986 when there were no marked birds. Therefore, a demographic model was used to provide estimates of survival in all years of the study, from observed changes in population size between years and estimated annual productivity.

In the first method, survival rates were modelled for 1664 colour-ringed birds (of which 1624 were marked as nestlings and 40 as full-grown birds) with 367 resightings between 1986 and 2004, using MARK (White & Burnham 1999) with logit link functions. Thirteen alternative models were considered. First winter and adult survival was modelled separately, except in one model where they covaried. Models with year-specific or constant survival rates, with linear or quadratic trends in survival, or with survival rates varying between three periods with differing rates of population growth defined by turning points analysis (1986–1988, 1988–1999 and 1999–2004) were compared using Akaike Information Criterion (AICc) (Anderson, Burnham & White 1994). Effects of winter temperature were examined by modelling year-specific survival in relation to the mean of daily minimum temperatures (°C) from 1st December to 28th February at Santon Downham. Further details are given in the Supporting Information. To test whether resighting rates differed between returning juveniles (1 year old when resighted) and older birds, as may occur if there is delayed territorial settlement, we compared the ratio of birds recorded to those not seen in any one year, for all birds known to be alive in that year (i.e. seen in a subsequent year). Likelihood of resighting did not differ between first-years (39% of = 83 resighted) or older adults (42% of 77 resighted, Fisher exact test, = 0·7489). We therefore modelled resighting as constant across age classes. Resighting rates may differ between males and females, however sex was not confirmed for 49·7% of resighted adults therefore combined resighting rates and pooled survival estimates were used for both sexes. This will reduce precision of survival estimates, but will not create any bias unless mortality rates also differ markedly between sexes.

In the second method, survival rates were estimated using a population model. Numbers of woodlarks in the Breckland region were known for each year up to 1999, and breeding productivity was estimated from the simulation model, allowing annual survival rates for 1971–1999 to be estimated from a demographic model. To extend the analysis beyond 1999 we also calculated survival rates for the forest population only. We assumed an equal sex ratio and that all pairs attempted to breed in each year. We considered a demographic model in which we assumed adult survival, A, was constant throughout the study period and estimated using mark–resighting analysis for 1986–2004. Yearly survival rates of first-year birds, Ft, were estimated from:


where Nt is the population in year t, and Pt the year-specific breeding productivity per pair. We also produced a second model, in which year-specific adult (At) and first-year (Ft) survival rates covaried. To allow this to be solved for Ft, we assumed that in any one year the ratio of adult to first-year survival is fixed as the ratio of mean adult (A) to mean first-year survival (F) estimated by mark–resighting analysis. In this model:


These models represent two extremes. In the first, environmental effects on winter survival act solely on vulnerable first-year birds, with adult survival constant. This is likely to over-estimate the amplitude of variability in first-year survival. In the second adult and first-year survival covary; their ratio is fixed so that they have the same inter-annual variance. As adult survival is likely to be less variable than that of first-year birds (Siriwardena et al. 1998b), this will underestimate the variability in first-year survival.

Both models assume a closed population; while this may not be strictly true, the Breckland woodlark population was the largest regional population by far throughout the period of study. In 1997, the relative size of nearby populations was: Suffolk Sandlings (∼70 km away) 209 territories, North Norfolk (∼35 km) 11 territories, Lincolnshire (∼145 km) 30 territories, Nottinghamshire (∼120 km) 29 territories, compared to at least 420 territories in Breckland (Wotton & Gillings 2000). Between 1986 and 2004, there were only a small number of resightings of colour-ringed birds demonstrating emigration from Breckland to the Suffolk Sandlings (three individuals), Nottinghamshire (four individuals) and Lincolnshire (six individuals), despite significant resighting effort. Although in 1997, the Suffolk Sandlings held a substantial population of woodlarks, numbers have subsequently declined in this area (Macklin 2004). Immigration and emigration thus likely involved only a small proportion of the population.

To ensure consistency, only records of woodlarks from sites that were continuously monitored throughout the period of study were included in the annual estimates of Breckland population size. These comprised all of Thetford Forest (18 786 ha) and 87% (7032 ha) of all heathland remaining in Breckland (considered as 8119 ha of heathland Sites of Special Scientific Interest). The maximum proportion of total recorded woodlark numbers in any one year that was located on sites outside our continuously monitored area was 6·5% in 1999.

First-year survival rates calculated from the population model were related to winter temperature (mean of daily minima between December and February) using linear regression. Weather variables were taken from Santon Downham, as many woodlarks winter in Breckland (Atkinson 2001; Dunmore 2006). Although locations of wintering sites beyond Breckland are not known, annual variation in winter temperature is broadly correlated across large areas of central and southern England (Jones & Hulme 1997). We also ran GLMs incorporating a categorical variable comparing first-year survival rates before and after a turning point in the population trajectory.


Population trends

Woodlark populations increased more than 10-fold during the period of study (Fig. 1). Turning points were identified in 1988, when the rate of population growth accelerated, and 1999, when the rate of growth declined (Fig. 1a). Observed annual population growth rates for the three periods, estimated from the population time series using autoregressive models of order 1, were: 1971–1988, λ = 1·13; 1988–1999, λ = 1·22; 1999–2004, λ = 0·92. Habitat availability in the entire forest increased until 1988, fluctuated during 1989–1998, and subsequently declined (Fig. 1b). However, during the 1990s woodlarks expanded their range within the forest, colonizing previously unoccupied blocks, so that the area of habitat available in occupied blocks generally increased throughout the period 1971–1998 (Fig. 1b). Furthermore, there was a fourfold increase in woodlark density after 1988 (Fig. 1c), suggesting the upturn in population growth rate from 1988 cannot be attributed to a release from habitat limitation. Although the availability of habitat decreased after the 1999 turning point, woodlark density also decreased, again suggesting that factors other than habitat extent contributed to the population decline.

Figure 1.

 Woodlark population size and the availability of suitable breeding habitat in Thetford Forest with respect to time. (a) The smoothed population trend from a thin plate spline with 11 d.f., with turning points marked by vertical dotted lines. (b) Numbers of singing males (points) and area of suitable breeding habitat (re-stocked stands aged 0–6 years and areas of open space), shown for the entire forest (continuous line) and occupied blocks of forest (broken line). (c) The density of woodlarks, measured as the number of singing males per hectare of suitable habitat in occupied blocks.

Success and productivity of nesting attempts

Daily nest survival rate declined significantly through the period of the study (Table 1a), from 0·984 (95% CL 0·976–0·989) in 1975 to 0·949 (0·939–0·959) in 2004. Nest failure over the entire 28-day nest period, from laying the first egg to producing at least one fledgling, increased from 36% (95% CL 26–50%) of nesting attempts in 1975 to 76% (69–83%) in 2004 (Fig. 2a). Few instances of partial brood loss were observed (2·6% of broods) and chick survival was high in surviving nests (95·7% of hatched chicks, in nests that fledged at least one chick, survived to fledging), suggesting starvation was not an important source of chick mortality. Desertion accounted for only 3% of nest failures, with the remaining 97% of failures attributed to predation. The proportion of nesting attempts that failed with the lining torn out, representing a subset of predation events that may relate to particular predator species, increased through time (χ21 = 16·85, = 864, < 0·0001), from 2% (95% CL 0·7–5%) of all nests in 1975 to 17% (12–22%) in 2004 (Fig. 2b,c). In contrast, the estimated proportion of nesting attempts that failed with the lining left intact did not change (χ21 = 0·94, = 864, = 0·33). Colonization year of the forest block in which the nest was located did not affect nest survival (adding colonization date to the minimum adequate model: χ21 = 0·03, = 0·8681, = 709 nests), suggesting spatial density dependence is an unlikely mechanism for the change in nest success. Other analyses showed no effect of local population density (per forest block) on nest survival (Wright 2006).

Table 1.   Generalized linear models of nest success for woodlarks breeding in Thetford Forest in relation to a linear trend over years, temperature (mean of daily minimum temperatures, 0·1 °C) and rainfall (daily mean, mm)
 Model detailsLikelihood ratio test results
Deviancend.f.Variable estimatesSE of estimateχ2d.f.P
  1. The relationship with weather was tested during the whole breeding season (1 March–30 June) and during the period specific to each nesting attempt (from the date the first egg was laid in the nest to the date the nest either fledged or failed).

(a) Linear trend over years835·32717715Year−0·04080·009819·211<0·0001
(b) Linear trend over years and March–June weather833·97717714Year−0·04470·010520·401<0·0001
(c) Linear trend over years and nest-specific weather675·61677674Year−0·05100·010825·621<0·0001
Figure 2.

 (a) Annual mean nest failure of Thetford Forest woodlarks from 1974 to 2004. (b) The proportion of all nesting attempts where the lining was torn out. (c) The proportion of all nesting attempts that failed with the nest left intact. Error bars show 95% confidence intervals; continuous and dotted lines show fitted models and their 95% confidence limits respectively.

Nest survival was greater when temperatures were higher during the individual nesting period but was not related to rainfall (Table 1c). There was no relationship between annual nest success and either temperature or rainfall during that year’s breeding season (Table 1b), probably because of high intra-seasonal variability in weather. Year-specific nest survival rates were not correlated with population growth rate for either the forest (= 0·190, = 0·344) or Breckland (= 0·105, = 0·642) population, indicating that changes in nest survival were not responsible for observed population changes.

Clutch size was significantly related to rainfall (negative) and temperature (positive) during the laying and pre-laying period (Table 2). Although highly significant, the effect of variations in weather on clutch size were relatively small, resulting in a difference in predicted clutch size between the best and worst years of 11% for 28th March and 8% for 2nd May (times of peak nesting activity).

Table 2.   Generalized linear models of clutch size and number of fledglings per egg for woodlarks breeding in Thetford Forest
 Model detailsLikelihood ratio test results
Deviancend.f.Variable estimatesSE of estimateFd.f.P
  1. (a) GLM of clutch size; temperature (mean of daily minimum temperatures, 0·1 °C) and rainfall (daily mean, mm) refer to the period from four days before to four days after the date the first egg was laid (lay date). (b) GLM of number of fledglings per egg (in successful nests), related to weather during the 12 days prior to fledging. (c) GLM of number of fledglings per egg in relation to April–June weather.

(a) Clutch size48·97532527Lay date0·00390·001113·3710·0003
Lay date2−2·17 × 1059·03 × 10−65·8210·0162
(b) Fledglings per egg (brood stage weather)414·42330326Lay date−0·00770·00542·0410·1529
(c) Fledglings per egg (breeding season weather)414·12324320Lay date−0·00320·00340·8610·3547

Rainfall during the brood stage had a marginal effect on the number of fledglings per egg (Table 2). Temperature did not affect the number of fledglings per egg (Table 2). The effect of rainfall during the brood stage caused a difference in the predicted number of fledglings per egg between the years with the wettest and driest weather during the brood stage, of 7% for a lay date of 28th March and 11% for 2nd May. Because the effect of rainfall on the number of fledglings per egg was small and only marginally significant, it was not included in simulation models of breeding productivity.

Woodlarks bred earlier in warmer springs (Fig. 3). The timing of the start of the breeding season was strongly negatively related to local temperatures between 15 February and 15 March (fifth percentile of the laydate distribution: r2 = 0·72, F1,16 = 41·85, < 0·001, = 18; 10th percentile r2 = 0·63, F1,16 = 26·73, < 0·001, = 18; 25th percentile r2 = 0·51, F1,16 = 16·60, = 0·001, = 18). The advance in the start of the breeding season per 1 °C increase in temperature was virtually identical whether estimated by the fifth, 10th or 25th percentile (2·6 days ± 0·4 SE; 2·7 days ± 0·5 SE; 2·5 days ± 0·6 SE respectively). As the best-fitting model was obtained for the fifth percentile, we use this for subsequent analyses of spring weather effects on the timing of breeding. Although spring temperature significantly affected the timing of nesting in each year, breeding did not advance over the study period due to considerable inter-annual variability in weather (Fig. 4: r2 = 0·01, F1,16 = 0·015, = 0·904, = 18).

Figure 3.

 Timing of the start of the woodlark breeding season, estimated from the fifth percentile of the lay date distribution, related to early spring temperatures.

Figure 4.

 Timing of the start of the breeding season, estimated from the fifth percentile of the lay date distribution, shown for all years where more than 21 nests were found.

Changes in annual reproductive output

We used the simulation model to examine the relative effect of weather and changing nest success on annual breeding productivity (the total number of fledglings per pair per year). Running the model with constant daily nest survival rate, and varying one reproductive parameter in turn, demonstrated that effects of weather on clutch size altered productivity by 4·7% between the best and worst years. Effects of temperature on nest survival altered productivity by up to 9·0%, while the change in the timing of breeding due to weather altered productivity by up to 23·5%. The combined effect of all these weather variations (without incorporating change over years in nest failure rates) caused a difference in productivity of 24·3% between the best and worst years. Combined weather effects increased potential reproductive output over the study period, so that reproductive output was somewhat higher after 1988. However, this effect was small, with an increase of just 0·3 chicks per breeding pair (7·9%) in the latter period.

In contrast, increased nest failure rates alone (with other parameters held constant) caused a decline in productivity of 49·8% over the 34 years of the study; from a mean of 5·7 (±2·6 SD) fledglings per pair in 1971 to just 2·9 (±2·5 SD) in 2004 (Fig. 5). Incorporating combined effects of weather on the timing of breeding, nest success and clutch size did not overcome this trend. The full model incorporating increased nest failure (attributed to increased predation) and weather effects, predicted a marked decline in breeding output through time, although inter-year variation in weather caused productivity to fluctuate around this trend (Fig. 5).

Figure 5.

 Predicted mean number of fledglings produced per pair in each year from 1971 to 2004, estimated by iterative simulation modelling (= 10 000 pairs per year). The dotted grey line shows only the effect of increased nest predation rates; the solid grey line shows combined effects of changes in nest predation and timing of breeding; the dotted black line includes these factors, plus effects of temperature on nest success; and the solid black line shows the full model that includes all previous factors plus additional effects of temperature and rainfall on clutch size. The double-dotted black line shows the model output with all three weather effects, but with nest predation rate held constant.

To test the relative importance of different effects on annual productivity, results of the full productivity model were compared with those of models with some effects excluded, by linear regression. Increased nest failure explained 88·2% of the variation in the full model; the combination of increasing failure rates and the effect of variation in spring temperatures on the timing of breeding improved the r2 to 0·975; adding the effect of temperature on nest survival further improved the r2 to 0·996 and the remaining variation was explained by effects of weather on clutch size. The model that included all weather effects but with nest failure rate held constant gave an r2 of just 0·010 and was not significantly related to the full model (= 0·568).

The start and end dates of the breeding season are the least robust parameters in the simulation model of breeding productivity. These values were determined by comparing the model generated lay date distribution to the pattern of observed lay dates at the start and end of the season and the model successfully replicated the ‘tails’ at each end of the season. We tested the sensitivity of the model to start and end dates by shortening the breeding season by 10 days at each end. However, this made virtually no difference to estimates of the effect of predicted climate change.

The date when birds stop breeding is also likely to depend on the fate of their previous nesting attempt, with birds that have successfully fledged young stopping earlier (Bowden & Green 1992; Mallord et al. 2008). We were unable to incorporate this effect into our model as we did not have sufficient data for pairs followed throughout the breeding season.

We tested the validity of assuming that the mean date females stop laying is the same in all years by comparing whether the rate of decline in observed numbers of first egg dates in the latter part of the season differs among groups of years categorized by turning points in the population trend (1971–1987, 1988–1998 and 1999–2004). The interaction between laying week and year group was nonsignificant (F2,24 = 0·05, = 0·948) suggesting no difference in the timing of the end of the breeding season among these time periods. Similarly, the timing of the end of the breeding season did not differ significantly between years with colder vs. milder early spring temperatures (F1,16 = 1·54, = 0·233).

Survival rates (mark–resighting analysis)

The mark–resighting model estimate of mean first-year survival across 1986–2004, was 0·22 (95% CL 0·16–0·28), while estimated mean adult survival rate was 0·60 (95% CL 0·54–0·66). The most parsimonious model (lowest AICc) had year-specific first-winter and constant adult survival estimates. A model with a linear trend in first-winter survival and constant adult survival had an AICc only 0·02 higher and suggested survival declined between 1986 and 2004 [analysis of deviance (Grosbois et al. 2008): = 17·65, = 0·0003]. Models that incorporated winter temperature had higher AICc values and nonsignificant analyses of deviance, suggesting temperature at Santon Downham did not significantly affect survival. Further details are given online (Supporting Information, Table S2).

Survival rates (demographic model)

First-year survival rates increased with winter temperatures (Fig. 6, GLM with identity link and normal errors, r2 = 0·148, = 0·028 ± 0·013 SE, F1,26 = 4·53, = 0·043, = 28). Incorporating a categorical variable contrasting the periods before and after 1988, while controlling for temperature, improved model fit (r2 = 0·307: change in residual variance, F1,25 = 4·88, < 0·05) with higher survival rates in the latter period (= 0·083 ± 0·035 SE, F1,25 = 5·708, = 0·025). However, the effect of temperature was no longer significant (F1,25 = 2·416, = 0·133), suggesting a factor other than temperature caused higher survival rates in the latter period. The temperature effect likely relates to just three particularly cold winters during the earlier time-period with mean temperatures around −2 °C; no years after 1988 were this cold. This may explain why no winter temperature effect was found in capture–mark–resighting models of survival that did not include earlier years. Despite these three early cold winters there was no overall trend in winter temperature between 1971 and 2004 (F1,31 = 1·757, = 0·195).

Figure 6.

 (a) First-year survival rates (calculated from a demographic model that assumes constant adult survival) in the Breckland region in relation to winter temperatures for 1971/1972–1987/1988 (closed circles) and 1988/1989–1999/2000 (open circles). (b, d) First-year survival rates against year for Breckland and Thetford Forest respectively. (c) First-year survival rates in Thetford Forest in relation to winter temperatures, as (a) but also showing first-year survival rates in 1999/2000–2003/2004 (crosses).

Examining first-year survival rates for the forest population only, still assuming constant adult survival (Fig. 6c,d), showed that forest survival rates were very similar to the overall Breckland survival rates during the period when both populations were monitored (up to 1999). First-year survival rates differed significantly among time-periods defined by the turning points (Kruskal–Wallis test, inline image = 8·59; = 0·014), with first-year survival rates after the 1999 turning point in the population trend considerably lower than during earlier periods (Table 3).

Table 3.   Demographic rates in periods with differing rates of population increase, as identified by turning points analysis
PeriodPredicted lay date of first nestAnnual productivity (per pair)Clutch sizeNest success rate (per attempt)Predicted lay date of last nestPopulation model first-year survival (Breckland)Population model first-year survival (forest)CMR model first-year survivalCMR model adult survivalPopulation growth rate (lambda)
  1. Values shown are mean ± SE, except nest success and CMR-modelled survival which show the mean and 95% confidence limits. Values for clutch size are based on observed data, nest success rates are derived from a GLM of daily nest survival rates in each period, other breeding parameters are values from a simulation model of annual breeding output, and first-year survival rates are derived from population models based on either the entire Breckland population or the Thetford Forest population that assume constant adult and varying first-year survival. Dates are given as days from 1st March = day 1. CMR-modelled survival rates during the first time-period only cover 1986–1987 for first-year birds and 1987 for adults, and thus are not representative of the entire period from 1971–1988. The population growth rate is estimated using eqn 1, with annual productivity estimated from the simulation model of breeding (incorporating all weather and nest-survival effects) and first-year survival values estimated from the forest population model shown in this table.

1971–198830·18 ± 1·135·22 ± 0·124·05 ± 0·060·52 (0·40–0·63)93·78 ± 0·200·21 ± 0·0240·21 ± 0·0300·33 (0·21–0·48)0·76 (0·20–0·98)1·15
1988–199924·06 ± 1·324·45 ± 0·144·01 ± 0·040·40 (0·33–0·47)93·68 ± 0·240·30 ± 0·0210·29 ± 0·0230·20 (0·14–0·26)0·60 (0·53–0·68)1·25
1999–200427·66 ± 1·093·40 ± 0·143·87 ± 0·040·26 (0·19–0·33)92·70 ± 0·33No data0·18 ± 0·0100·09 (0·05–0·17)0·46 (0·34–0·60)0·90

These results must be interpreted with caution as the derived survival estimates have no measure of uncertainty so that models relating these to other variables may be subject to type 1 error. However, the difference between the latter two time periods is consistent with survival estimates from capture–mark–resighting models. Survival estimates in the first time-period cannot be compared to capture–mark–resighting models as colour ringing did not begin until the end of this period. The geometric mean of first-year survival derived from the population model between 1986 and 2004 was 0·23, compared to an estimate of 0·22 from the capture–mark–resighting model for the same period. Annual survival estimates using the two methods were reasonably well matched, but for some years the demographic model predicted lower first-year survival (Supporting Information, Fig. S2). The close agreement between these two independent estimates of survival suggests both models are reasonably accurate.

First-year survival rates predicted by the second model (eqns 2, 3), that assumes adult and first-year survival covary with a fixed ratio (A/F = 2·73), had a similar mean (Breckland population: 0·23 compared to 0·24 for the first model) but were less variable (coefficient of variation = 0·20 compared to 0·41 for the first model). However, relationships between first-year survival and winter temperature were similar to those obtained using the first demographic model.


Effects of climate change versus environmental change

Although variation in weather among and within years affected breeding output, we could find no long-term signature of climate change on the productivity of this woodlark population. Woodlarks nested earlier in years with warmer spring temperatures, as do numerous other species (Crick & Sparks 1999; Dunn & Winkler 1999; Both & Visser 2001; Cresswell & McCleery 2003; Both et al. 2004; Reed et al. 2006). The difference in early spring temperature between the warmest and coldest years of 8·2 °C was sufficient to cause a 25 day advance in breeding, potentially increasing annual productivity by 0·90 chicks per pair (23%). However, across the years of this study (1986–2004) we found no significant trend in laying date due to considerable inter-annual variability, with no overall trend, in local early spring temperature during this period (F1,17 = 1·26, = 0·278). Although many other studies have found recent advances in egg laying (Crick et al. 1997; Winkler, Dunn & McCulloch 2002; Both et al. 2004), Crick & Sparks (1999) only found significant long-term trends for 53% of the 36 species they studied, while Beale et al. (2006) found no evidence that laying dates of ring ouzels Turdus torquatus had advanced despite a significant relationship between spring weather and timing of breeding. When considering nest-specific weather (rather than that for the annual breeding season) nesting success was greater when temperatures were higher, and clutch size was slightly larger when weather was drier and warmer, but these relationships had minor consequences, increasing annual productivity by less than 10% after 1988 compared to the earlier period.

Late winter and early spring temperatures in eastern England are predicted to rise by 1–2 °C on average over the next 50 years (Hulme et al. 2002). This would result in an advance in woodlark laying dates of 2·6–5·2 days, increasing annual productivity by 0·1–0·3 fledglings per pair, just 2–7% compared to the 30-year mean of 4·1 fledglings per pair. Climate is also predicted to become more variable, with an increased frequency of ‘extremely warm days’ (Hulme et al. 2002); the effects of this are unknown.

In marked contrast to the weak signature of climate change, an increase in the rate of nest failure (97% of which was attributed to predation) resulted in a substantial decline in annual productivity of 2·8 chicks per pair, or 50%, between 1975 and 2004. Nest failure rates were not related to woodlark territory density, either in Thetford Forest (this study) or in Dorset (Mallord et al. 2007c) and did not differ with lay date, plantation age, nest concealment, vegetation, soil type or distance to forest edge (Wright 2006). Thus increased nest failure is most likely due to changes in predator abundance (Dion, Hobson & Larivière 2000; Luginbuhl, Marzluff & Bradley 2001) not settlement of suboptimal territories at higher density. The proportion of nests predated with the lining torn out increased significantly between 1975 and 2004, while the proportion of nests predated but left intact did not change (Fig. 2), suggesting increased activity of a particular class of nest predator.

Despite the consistent increase in nest failure rates through the period of the study, the population dynamics showed three markedly different phases. A period of slow population increase during 1971–1988 was followed by faster population growth from 1988 to 1999, after which woodlark numbers declined.

Phase of rapid population increase 1988–1999

The observed increase in population growth rate after 1988 cannot be explained by changes in productivity, which declined throughout the study, thus winter survival rates were likely to be responsible. As winter temperature did not explain higher survival rates after 1988, other changes may have contributed to improved survival. Introduction of arable set-aside schemes, trialled in 1989/1990 with full uptake from 1992, and consequent increased availability of over-wintered stubbles that are important for seed-eating farmland birds (Donald et al. 2001; Moorcroft et al. 2002; Gillings et al. 2005), may have benefitted woodlarks. Flocks of up to 95 have been observed on stubble fields in our study area in recent winters (Atkinson 2001; Dunmore 2006).

Phase of population decline after 1999

Reduced productivity, resulting from increased rates of nest predation, is likely to have contributed to the population decline after the second turning point in 1999. Assuming the mean values of first-year and adult survival rates estimated using mark–resighting, productivity must be at least 3·6 fledglings per pair per year for the population to remain stable. According to our simulation model, annual breeding output was close to this level (3·7 fledglings per pair) in 2000, but fell below it from 2001 onwards, providing strong support for the importance of increased nest failure rates to the decline. However, first-year survival rates were also lower after 1999, explaining the overall linear decline in survival between 1986 and 2004 apparent in capture–mark–resighting models. During 1999–2004, the population multiplication rate (PMR, λ) was 0·90 (i.e. the population declines by 10% per year). Substituting the productivity value from the preceding period of growth (1988–1999), but keeping the survival rate from 1999 to 2004 gives a PMR of 0·99 (1% decline per year), while using the first-year survival rate from 1988 to 1999 but keeping the productivity value from 1999 to 2004 produces a PMR of 1·01 (1% increase per year). Thus reduced productivity and lower survival contributed similarly to the observed population decline in the latter period.

Reduced survival rates during 1999–2004 cannot be explained by either winter weather, or the amount of set-aside land or spring sown barley in Breckland, which was similar to that in the mid-1990s. Caution must be exercised in assuming that rates of recruitment predicted from the population model relate directly to survival, although survival rates independently estimated from capture–mark–resighting data were also lower after 1999, supporting the conclusions drawn from these independent methods. During this period of population decline, woodlark density decreased within open and young growth forest habitats, which supports the conclusion that the population decline post-1999 must be driven by extrinsic demographic factors. However, although our measure of habitat extent is applied consistently across years, it does not capture habitat quality (c.f. Mallord et al. 2007a), thus if the suitability of young plantations for woodlarks has reduced then habitat-mediated reduction in settlement could have contributed to observed declines.


This study has demonstrated the importance of considering other factors in addition to climate change in predictive population modelling. In contrast to clear effects of climate on timing of breeding and productivity in single-brooded species, effects on multi-brooded species are more complex. While many recent studies have focused on incorporating density-dependent effects into predictive models (Stillman et al. 2001; Sutherland & Norris 2002; Sutherland 2006; Mallord et al. 2007a) or the effects of changes in predation, landscape (Jackson & Green 2000; Evans 2004; Tyler & Green 2004) or climate (Asbjørnsen et al. 2005; Pearce-Higgins et al. in press), few have incorporated all of these effects (although see Wildhaber & Lamberson 2004; Vucetich, Smith & Stahler 2005). A thorough understanding of such effects is required if accurate predictive modelling is to be possible.


We are extremely grateful to the numerous observers who helped with woodlark population monitoring. We wish to thank the Forestry Commission, Ministry of Defence, Elveden Estate and other landowners for permitting access to land owned or managed by them. The Forestry Commission and the Royal Society for the Protection of Birds funded some of the data collection. We thank the Meteorological Office for providing the climate data through the British Atmospheric Data Centre (, and Phil Atkinson and Rob Robinson for valuable comments on earlier versions of this paper. LJW was funded by a NERC award (NER/S/A/2001/06141).