Demographic and environmental causes of the decline of rural Song Thrushes Turdus philomelos in lowland Britain

Authors


*Corresponding author. Email: will.peach@rspb.org.uk

Abstract

We review current knowledge of demographic mechanisms and environmental factors implicated in the population decline of Song Thrushes Turdus philomelos in rural Britain since the mid-1970s, and present new analyses of regional variation in population changes. Increased mortality during the first year of life (from fledging to recruitment) is highlighted as a potential demographic mechanism having driven the population decline, while Song Thrushes in a rapidly declining farmland population were making too few nesting attempts to sustain local numbers. Breeding Song Thrushes are strongly associated with non-cropped habitats such as woodland edge, field boundaries, gardens and scrub; they make substantial use of grassland, but avoid cereals when foraging. Earthworms constitute a key component of Song Thrush diet and the availability of this prey is strongly influenced by moisture levels in surface soils. Several lines of evidence suggest that dry surface soils during summer are deleterious to the productivity and survival of Song Thrushes, and regional variation in the rates of population change in Britain during 1970–86 was negatively correlated with the extent of under-field drainage on farmland (the main function of which is to promote the drying of surface soils). Increasing dryness of agricultural soils and the loss of grassland from eastern arable counties have probably both contributed to the declines of rural Song Thrushes in Britain. Loss of hedgerows and scrub, and the degradation of woodland may also have contributed to population declines but the role of predators remains unclear. Recovery of rural Song Thrush populations requires challenging new policy initiatives that should aim to restore nesting cover (scrub and woodland understorey), grazed grassland in arable-dominated areas and damper soils in summer.

Song Thrushes Turdus philomelos are included on the ‘Red List’ of Birds of Conservation Concern as a consequence of a large and sustained population decline in rural Britain (Gregory et al. 2002). The decline has been greatest on farmland (approximately 70% of pairs lost) but also pronounced in lowland woodlands (about 50% of pairs lost). There is growing evidence that Song Thrushes are also declining in gardens and suburban landscapes (Glue 2003, Snow 2003) but here we focus entirely on factors affecting rural thrushes.

Devising conservation action plans for Song Thrushes has been hindered by a lack of knowledge of the environmental factors that may have driven the decline and that may currently be limiting densities and demography in rural Britain. At the time of the last BOU farmland birds conference, much ecological research into the decline of the Song Thrush was either ongoing or planned (Thomson & Cotton 2000), and a considerable amount of new information has emerged since. Here we review the current state of knowledge relating to the decline of rural Song Thrushes, and we present new analyses of extensive census data testing specific hypotheses relating to environmental causes of population decline. Finally, we briefly discuss conservation action that might benefit this species in rural Britain.

POPULATION CHANGES AND DEMOGRAPHY

The Common Birds Census (CBC) measured a 69% decline of Song Thrushes on farmland between 1968 and 1999 and a 46% decline on woodland (Baillie et al. 2001). Multiplying the respective CBC indices by the average densities of territories in woodland and farmland indicates that Song Thrush numbers in these rural habitats declined from around three million territories during the early 1970s to about one million by the mid-1990s (Robinson et al. 2004; Fig. 1). Song Thrush numbers declined throughout most of this period, with only a modest partial recovery during the late 1990s. Approximately 85% of the total decline occurred between 1975 and 1986.

Figure 1.

Changes in the estimated number of Song Thrush territories in woodland and farmland in Britain between 1964 and 2000 (dots). The solid line represents a smoothed trend with 95% confidence intervals (dotted lines). The vertical tick marks identify significant (P < 0.05) changes in the population multiplication rate (see Robinson et al. (2004) for details).

Extensive demographic data have indicated slight increases in productivity per nesting attempt, with declines in first-year survival being highlighted as a potential demographic mechanism driving the population declines of the 1970s and 1980s (Baillie 1990, Thomson et al. 1997). Recent analyses incorporating ring recoveries of Song Thrushes ringed as chicks confirm the probable importance of first-year survival (from the third to the twelfth months of life), and highlight a potential influence of post-fledging survival (during the 2 months following fledging) on Song Thrush population changes (Robinson et al. 2004).

A recent study comparing stable and declining farmland populations has highlighted much lower annual productivity in the latter, mainly as consequence of fewer nesting attempts each summer (Thomson & Cotton 2000). Although initiation of first clutches occurred approximately simultaneously in the two populations, intervals between subsequent nesting attempts were significantly longer in the declining population, with fewer nesting attempts particularly during the latter part of the breeding season in June and July (Thomson & Cotton 2000, RSPB unpubl. data). Productivity in the stable population was about the level required to offset annual mortality typical of stable thrush populations (taken from Thomson et al. 1997, Thomson et al. 1999), while that in the declining population would, in the absence of immigration, be expected to result in annual population declines of approximately 22–28%. This is close to the observed declines in that population of 19% during 1996–97 and 26% during 1997–98 (Peach et al. 2004). There was also evidence of lower post-fledging survival (during 2 weeks after fledging) in the declining population (Thomson & Cotton 2000).

A study of game management practices in Leicestershire has highlighted the potential role of Corvid depredation of nests on Song Thrush population dynamics (Stoate & Szczur 2001). Variation in nest survival across study areas was negatively correlated with corvid density, and large increases in thrush breeding densities followed the removal of corvids and the coincident improvements in habitat quality. The potential influence of predators on Song Thrushes is discussed further below.

These demographic studies therefore highlight post-fledging and first-year survival as possible demographic drivers of the population declines of the 1970s and 1980s, but also suggest that inadequate productivity may have limited some farmland populations in recent years. Lack of food would seem to be the most plausible environmental cause of fewer nesting attempts (e.g. Verboven et al. 2001), while a wide range of environmental factors might influence survival during the 12 months after fledging.

HABITAT ASSOCIATIONS

During the breeding season, Song Thrush territory distribution on farmland is positively associated with woodland edge, field boundaries, gardens and scrub, and negatively associated with arable crops (Mason 1998, 2000, Murray 2004, Peach et al. 2004). Radiotracking of adult Song Thrushes during the breeding season has shown woodland edge, field boundaries, gardens and grassland to be the most utilized foraging habitats (Peach et al. 2004). The intensity of habitat usage (habitat selection) was greatest in scrub, field boundaries (especially wet ditches with hedges), gardens (especially with tilled earth), woodland edge and tall grass and herbaceous vegetation, and lowest in cereals, other arable crops and permanent pasture (Murray 2004, Peach et al. 2004). Song Thrushes in a declining population on arable farmland made greater use of gardens and arable crops during summer, possibly reflecting a lack of alternative food-rich habitats such as grassland, wet ditches and woodland (Peach et al. 2004). Wintering Song Thrushes are found mainly in field boundaries, gardens, woodland and grazed grassland (Peach et al. 2002); broad-leaved crops such as oilseed rape and kale are used but winter cereals are avoided (Wilson et al. 1997, Henderson et al. 2004).

Farmland Song Thrushes are therefore heavily dependent on non-cropped and grassland habitats for foraging, and may make greater use of arable crops when preferred woodland and grassland habitats are scarce.

EARTHWORMS AND DAMP SOILS

At times of year when earthworms are available to Song Thrushes (typically December to May), they account for a high proportion of Song Thrush diet (Davies & Snow 1965, Gruar et al. 2003). Moisture is one of the main factors determining the abundance of earthworms in the top 5–10 cm of soils (Gerard 1967, Peach et al. 2004). As farmland soils progressively dry out during April–July, earthworms descend deeper into the soil and become unavailable to Song Thrushes, forcing them to increase their proportional intake of snails and spiders (Davies & Snow 1965, Gruar et al. 2003). Bouts of summer rainfall bring short-term increases in soil moisture levels and in the proportion of earthworms in the diet (Gruar et al. 2003, Peach et al. 2004).

Although the body weights of Song Thrushes were not related to the predominance of earthworms (or any other prey items) in faecal samples, the summer weights of chicks and adults were negatively related to the dryness of surface soils (Gruar et al. 2003). Provisioning adults also foraged further from their nests during dry periods (Peach et al. 2004). More rapid and pronounced drying of surface soils in an intensively managed arable landscape where Song Thrush numbers were rapidly declining, was associated with lower soil surface earthworm biomass in June than on mixed farmland supporting a stable thrush population (Peach et al. 2004; Fig. 2). Given that the timing of egg laying and the number (or volumes) of eggs laid by birds are limited by female body reserves (Jones & Ward 1976) or by dietary intake of protein (Bolton et al. 1992, Houston 1997), it is plausible that reduced re-nesting propensity among thrushes in the arable landscape was a direct consequence of relatively severe food shortages mediated through an effect of dry surface soils on earthworm availability (Peach et al. 2004). The hypothesis that dry soils may limit Song Thrush productivity is supported at the extensive scale by positive correlations between the proportions of juvenile Song Thrushes (and Blackbirds Turdus merula) caught in standardized mist-netting programmes and local summer rainfall totals (Balmer & Freeman 2004, S. Freeman pers. comm.). Furthermore, annual variation in the survival rates of adult Song Thrushes (also measured at the extensive scale) is negatively correlated with the duration of summer droughts (Robinson et al. 2004). Taken together, these various studies provide strong evidence that dry summer soil conditions have negative demographic influences on Song Thrush populations probably by reducing the availability of key soil invertebrate prey such as earthworms. A similar mechanism has been shown to reduce the length of the breeding season of Snipe Gallinago gallinago (Green 1988).

Figure 2.

Seasonal changes in mean soil surface penetrability (a) and relationship between penetrability and earthworm biomass (b) on grass fields in mixed (open circles) and mainly arable (filled circles) farmland. Plotted points represent means for individual fields, and the same patterns were evident for arable fields, ditches, woodland and gardens (Peach et al. 2004).

The more rapid drying of surface soils in our arable study area was not a consequence of lower summer rainfall; on the contrary, total rainfall during March–July was greater in the arable study area than in the mixed study area (Peach et al. 2004). Differences in topography and soil types might have contributed to the faster drying of surface soils in the arable landscape but one factor that will have promoted the more rapid drying of agricultural landscapes in many regions of Britain is the widespread installation of under-field drainage pipes (Robinson & Armstrong 1988). In the following section we test whether regional variation in Song Thrush population changes is correlated with regional variation in the intensity of under-field drainage of agricultural land.

EFFECTS OF UNDER-FIELD DRAINAGE ON SONG THRUSH POPULATION CHANGES IN BRITAIN

By promoting the drying of surface soils, under-field drainage pipes increase crop yields, reduce crop disease problems and extend periods when heavy machinery can be used. Encouraged by the provision of government grants and free technical advice, rates of land drainage in England and Wales increased in every decade after the 1940s, peaking during the late 1970s and declining thereafter (Robinson & Armstrong 1988). Approximately half of all land drainage in Britain during the 20th century occurred during the 1970s, with most of this being concentrated in eastern arable areas of England especially on clay soils. Since the timing of land drainage in England and Wales broadly coincides with the timing of the Song Thrush population decline (Fig. 1), we tested whether pronounced regional variation in under-field drainage intensity (Robinson & Armstrong 1988) was associated with regional variation in changes in Song Thrush populations.

Regional statistics describing the areas of agricultural land drained between 1971 and 1980 were collated for 30 Ministry of Agriculture, Fisheries and Food (MAFF) divisions covering England and Wales (Armstrong 1978, 1981). We took the area of land subjected to ‘normal under-drainage’ (i.e. where pipe ditching and other operations such as pumping or reconditioning accounted for less than 25% of the total drainage scheme cost) and divided this by the area of farmland within that region to give an estimate of the proportion of all farmland drained. No equivalent data are available after 1980, but rates of drainage declined rapidly during the 1980s with the gradual withdrawal of grant aid. Nearly 20% of Essex and Lincolnshire farmland was drained during the 1970s compared with less than 2% in Cornwall, Hampshire and South Wales.

As Song Thrushes are strongly associated with field boundaries as both nesting and foraging habitat (Murray 2004, Peach et al. 2004), we also compiled hedgerow loss statistics for the same 30 MAFF divisions. These data cover the period 1969–85 and are based on repeat aerial photography surveys of a random stratified (by soil type) sample of 140 12-km2 blocks covering 1.1% of the area of England and Wales (Hunting Surveys & Consultants 1986). There were at least two sample blocks in each county. Rates of hedgerow loss were greatest in East Anglia (1.3% per annum) and the East Midlands (1.1%) and lowest in southwest and northwest England (0.5% in both regions). On the finer MAFF division scale, annual rates of hedgerow loss varied between 1.8% in Cambridgeshire and Bedfordshire, and just 0.2% in Surrey and Sussex. Although the extent of hedge loss and the intensity of field drainage were correlated across regions (Pearson's r = −0.37, P = 0.044), the relationship was considered sufficiently weak to provide statistical power to distinguish between the effects of the two factors.

We analysed territory counts for Song Thrushes for all farmland CBC plots in England and Wales during the period 1970–86. Plots recording fewer than two Song Thrush territories over this period were excluded. There were no farmland CBC data for one of the five Welsh divisions (Powys), so analyses are based on data from 347 CBC plots from 29 MAFF divisions covering England and most of Wales.

We estimated average annual rates of change in the numbers of Song Thrush territories for each MAFF division by fitting a log-linear model in which the territory count for each plot in each year depended on a ‘plot’ factor plus a ‘year * division’ interaction (Table 1, Fig. 3). We tested the influence of drainage intensity and hedgerow loss on average rates of Song Thrush population change across MAFF divisions by fitting a second log-linear model in which the territory count for a particular plot in a particular year depended on a ‘plot’ factor plus a ‘year’ covariate plus a ‘year * drainage intensity’ or ‘year * hedgerow loss’ interaction. Both models were fitted using procedure GENMOD in SAS (SAS Institute Inc. 1999) declaring Poisson error structures and a log link. The CBC territory data were not overdispersed.

Table 1.  Summary statistics for log-linear models fitted to the Song Thrush farmland CBC data for the period 1970–86.
ModelaNo. of parametersDevianceAICb
  • a

    Plot and Division are factors (with 347 and 29 levels, respectively), while Year, Drainage and Hedge loss are all fitted as continuous linear covariates. Statistically significant (P < 0.0001) predictors of Song Thrush counts are highlighted in bold; other effects were not significant (P > 0.2). Directions of significant covariates are reported in the main text.

  • b

    b The smallest Akaike's information criterion (AIC) value (bold) indicates the most parsimonious model.

Plot3463024.23716.2
Plot+Year3472844.33538.3
Plot+Year+Year * Drainage3482829.93525.9
Plot+Year+ Year * Hedge loss3482843.53539.5
Plot+Year+Year * Drainage+ Year * Hedge loss3492828.63526.6
Plot + Year + Year * Division3762516.23268.2
Figure 3.

Regional variation in the average rates of change (% per annum) in farmland territory counts of Song Thrushes on CBC plots between 1970 and 1986. Regions are MAFF divisions for which the land drainage statistics were available.

Trends in Song Thrush abundance varied significantly between MAFF divisions (likelihood ratio test comparing models with and without the year * division interaction: inline image = 508.0, P < 0.001), and there was a significant (though weak) negative relationship between rate of population change and the intensity of under-field drainage (year * drainage intensity interaction: inline image= 14.4, P < 0.001; Table 1, Fig. 4). The rate of Song Thrush population change was not related to the extent of hedgerow loss either when modelled on its own or in the presence of the intensity of drainage (year * hedgerow loss interaction: inline image = 0.9, P > 0.3 and inline image = 1.3, P > 0.2, respectively).

Figure 4.

Relationship between average population multiplication rate (PMR, ± 95% confidence intervals) of Song Thrushes in 29 regions of England and Wales during the period 1970–86 and the percentage of farmland in each region that was subjected to under-field drainage improvements during the period 1970–80. The negative relationship is statistically significant (P < 0.0001). Symbols represent the following agricultural landscapes as defined by Robinson et al. (2001): six arable-dominated regions (filled circles), nine grass-dominated regions (filled squares), ten mixed regions (open diamonds) and four upland regions (open circles).

This weak but significant tendency for Song Thrush population changes to be more negative in areas of Britain where under-field drainage was relatively intensive provides further supporting evidence for our hypothesis that drier agricultural soils during summer may have negative impacts on rural Song Thrushes via the availability of their soil-dwelling invertebrate prey. However, we acknowledge that the relationship is weak and not necessarily causative as the intensity of land drainage may reflect a wider suite of changes in agricultural practices. The weakness of the relationship (Fig. 4) may reflect unrepresentative regional CBC data or the influence of other factors on Song Thrush population changes. For example, the large declines of Song Thrushes in Cheshire/Staffordshire and Somerset/Dorset (evident on Fig. 3, and on Fig. 4 as the two grass-dominated regions showing the greatest rates of population decline) where there was relatively little under-field drainage, might be associated with the intensive dairy farming that predominates in those counties. Rapid growth of grass swards following chemical fertilizer applications can promote soil dryness through increased transpiration (Lazenby 1988).

OTHER ENVIRONMENTAL FACTORS

Important habitats for farmland Song Thrushes are field boundaries (especially when damp and with hedge cover), woodland edge, scrub, grassland (grazed and unmanaged) and bare ground in gardens, while cereals are strongly avoided. Approximately half of Britain's hedges were removed between 1947 and 1990, with high losses continuing between 1978 and 1990 (Barr & Parr 1994). Densities of many farmland birds (including Song Thrushes) are known to be sensitive to the density of hedgerows (Lack 1992) and hedgerow removal has almost certainly reduced the carrying capacity of much lowland farmland for Song Thrushes. Our failure to detect an influence of hedgerow loss on Song Thrush population changes (above) probably reflects fine-scale heterogeneity or bias in the regional estimates of Thrush population changes and/or hedgerow loss.

Song Thrushes in the UK probably benefited from a 57% increase in total woodland between 1965 and 2001, and an 88% increase in farm woodlands between 1981 and 2001 (Forestry Commission 2002). The area of broad-leaved woodland in England increased by 36% during the period 1980–98, while the area of coniferous woodland declined by 7% (Forestry Commission 2001). However, the removal of understorey vegetation by increasing deer populations may have reduced the suitability of many lowland woods as nesting and foraging habitats for Song Thrushes (Fuller 2001). Scrub is an important source of cover and may be an important factor influencing Song Thrush distribution on arable farmland lacking woodland (Mason 2000). Approximately 60% of scrub was lost in Great Britain between 1965 and 1980 (Locke 1987), representing a substantial loss of potential nesting cover and foraging habitat for Song Thrushes, as well as for a number of other species of farmland bird (Fuller et al. 2004).

Changes in agricultural cropping patterns have probably negatively affected Song Thrushes. Widespread conversion of permanent pasture to cereal production particularly in eastern Britain during the 1960s and 1970s (O’Connor & Shrubb 1986) amounts to the replacement of a heavily utilized, food-rich resource with a habitat that is strongly avoided by foraging Song Thrushes (Murray 2004, Peach et al. 2004). Moreover, the loss of livestock and organic fertilizers from many arable farms is likely to have reduced the organic content of cultivated soils and their suitability for earthworms and other soil invertebrates (Edwards 1984). We tested the influence of the grass–arable mix on the density of Song Thrushes on arable farmland using data from the BTO/JNCC (Joint Nature Conservation Committee)/RSPB Breeding Bird Survey (BBS) for 1998 using methods similar to those of Robinson et al. (2001). Average Song Thrush densities doubled when the proportion of grassland along each BBS transect increased from zero to 40% (Fig. 5). As the median percentage of grassland in the sample of arable BBS squares was only 10% (75% of squares had less than 20% grass), the area of agricultural grassland may well limit thrush densities in arable regions. Falling densities of thrushes in squares with more than 40% grassland probably reflects genuine avoidance of the few grass-dominated areas of eastern England, because there was no such reduction in Song Thrush densities in grass-dominated squares from western pastoral regions of Britain. The reason for this regional difference in the suitability of grass-dominated landscapes requires further investigation. Inclusion of temporal changes in the grass–arable mix would be a useful extension of the analysis of factors influencing regional variation in Song Thrush population changes (above).

Figure 5.

Relationship between Song Thrush density and amount of grassland habitat for BBS squares from nine arable-dominated counties of eastern England (see Robinson et al. 2001). The number of grass transect sections in each BBS square (maximum = 10) was the sum of the two grassland codes (E1 + E2) plus half of the mixed grass/tilled code (E3 * 0.5). The lines show the predicted model relationship with 95% confidence intervals.

The potential influence of predators such as Magpies Pica pica and Sparrowhawks Accipiter nisus on Song Thrush numbers remains unclear and controversial. Average whole-nest survival rates increased during periods of population decline (Baillie 1990), and have not declined in regions of Britain with increasing Magpie densities (Gooch et al. 1991). Although spatial variations in nest failure rates of Song Thrush and Blackbird are positively correlated with corvid density (Paradis et al. 2000), for neither species are population changes correlated with local changes in Magpie numbers (Thomson et al. 1998). A comparison of stable and declining farmland Song Thrush populations found similar rates of nest failure in each (Thomson & Cotton 2000). Removal of corvids and improvements in habitat quality on farmland in Leicestershire are associated with increased nest survival rates and population levels for several songbirds, including Blackbird and Song Thrush (Stoate & Szczur 2001). However, it is unclear in this study whether predator control or habitat changes have had most influence on songbird populations (see also Whittingham & Evans 2004). Negative correlations between population changes of Song Thrushes and Sparrowhawks on CBC plots (Thomson et al. 1998) suggests that further research may be merited into the potential impacts of this predator. However, detailed studies have not suggested that Song Thrushes experience significant predation pressure from hawks (Opdam 1978, Newton 1986).

Prolonged low winter temperatures and dry summer conditions can depress first-year and adult survival rates, respectively, but do not appear to have driven Song Thrush population declines (Thomson et al. 1997, Robinson et al. 2004). Dry summer soil conditions were associated with reduced body weights of adult Song Thrushes and chicks (Gruar et al. 2003) with reduced availability of earthworm prey being the likely mechanism (Peach et al. 2004). A similar mechanism probably accounts for the lower proportions of juvenile Song Thrushes and Blackbirds caught in standardized mist-netting programmes (implying lower annual productivity) during dry summers (Balmer & Freeman 2004). Increasingly dry summers in southern and eastern England, as predicted under some global climate change scenarios (Hulme et al. 2002), are likely to have negative impacts on Song Thrushes via their earthworm prey.

CONSERVATION PROSPECTS

Breeding Song Thrushes require dense woody vegetation for nesting cover situated close to damp soils providing soil invertebrates. The small home ranges of nesting Thrushes (Murray 2004, Peach et al. 2004) emphasize the need for nesting and foraging habitats to be close together. Current British agri-environment schemes offer relatively few measures that might benefit Song Thrushes on lowland farmland. Planting and restoration of hedgerows under the Countryside Stewardship Scheme (CSS) should provide nesting and feeding sites, particularly when coupled with sympathetic field margin management. Song Thrushes prefer tall, wide, species-rich hedges with trees, ditches and adjacent grassland (Hinsley & Bellamy 2000). Of the new ‘arable options’ recently incorporated into the CSS, Song Thrushes are most likely to benefit from ‘wild bird seed mixes’ incorporating leafy brassicas such as kale (Henderson et al. 2004).

The various CSS grassland management and creation measures are unlikely to encourage arable farmers to take on grazing livestock. Organic farming might be expected to provide benefits for Song Thrushes in arable landscapes because it often involves a return to mixed farming with livestock, because higher levels of organic matter in the soil should promote key invertebrate prey such as earthworms, and because field boundaries on organic farms may be managed more sympathetically for birds. There is some evidence that densities of Song Thrushes are higher on organic farms and this does not appear to be entirely due to the presence of taller, wider hedgerows (Chamberlain & Wilson 2000).

New farm woodlands created under the Farm Woodland Premium Scheme and the Woodland Grant Scheme have been shown to be attractive to Song Thrushes during summer and winter especially where the adjoining hedges are dense and contain trees (Vanhinsbergh et al. 2002), and their utility might be increased by the encouragement of perimeter shrub cover (Hinsley et al. 1995). There are currently no financial incentives for the planting or restoration of scrub on British farmland, despite the importance of this habitat to Song Thrushes and other birds.

New policy initiatives are needed to encourage mixed farming, damper soil conditions and small, uncropped features across significant areas of UK lowland farmland. Given the importance of wet ditches to Song Thrushes and several other priority farmland birds (e.g. Reed Bunting Emberiza schoeniclus), consideration should be given to agri-environment payments for the creation and maintenance of ditches that remain damper for longer during summer, preferably with hedge cover or close to scrub. Agri-environment payments for raised water levels on grass fields have traditionally been aimed at encouraging breeding waders but damp grass fields are likely to benefit a wide range of birds that feed on soil invertebrates.

Acknowledgments

Major contributions to the RSPB Song Thrush study were made by Derek Gruar, Roy Taylor, Kim Fenton, Chris Orsman, Pete Cotton, Ian Hill, Matthew Denny, Dave Barritt, John Mallord and Andrew Impey. Stephen Baillie, David Thomson and Rhys Green contributed to analyses of extensive BTO data. Paul Britton helped produce Figure 3. We are grateful to them all.

Ancillary