Analyses of Survival in Fed and Control Areas

Appendix 7

Objective 3: to quantify the relationship between survival and body condition with food provision and its knock-on effects on breeding populations

Analyses of survival in fed and control areas

Methods

Winter trapping

Trapping was conducted with the aim of individually colour-ringing 30-50% of the yellowhammers and chaffinches present in each study area in each winter. Local winter abundance, as revealed by monthly surveys of the seed-rich habitats within study tetrads, varied a great deal between areas and fluctuations also occurred between months, so both target and actual numbers of birds trapped also varied. A number of different methods were used, adapted according to habitat context and bird behaviour in each area, but principally, mist-nets placed parallel to or in gaps in hedges, baited walk-in traps and elastic-powered “whoosh” nets. In fed areas, most trapping activity was focused on food patches themselves, which acted as long-term bait for bird flocks and facilitated trapping. In control areas, flocks of target species were located during the monthly surveys and short-term targeted baiting (using the same seed types as used for the main feeding activity) was used to concentrate these flocks in particular locations that allowed trapping. Seed bait was only supplied for long enough for the birds to find and use it and was not replenished after trapping occurred to minimize the blurring of the fed-unfed experimental dichotomy. Similar targeted baiting and trapping was also used in some fed areas, where concentrations of birds were found away from permanent food patches.

Trapping was initiated once flocks had built up at feeding sites, mostly from January to March each winter. A few trapping days also occurred in December and the first week in April. Some species will have begun to breed towards the end of the period used for ringing, potentially biasing some analyses, particularly those concerning movements. However, trapping was only conducted where birds were found in flocks foraging on seed, i.e. exhibiting what can be considered to be “winter behaviour”. Further, although chaffinches were regularly observed exhibiting “breeding season behaviour” in March in the study areas, this was rare for yellowhammers, which seemed to start breeding in late April or May.

All chaffinches, yellowhammers, goldfinches and reed buntings caught were fitted with individual colour-ring combinations consisting of one plastic colour-ring and a BTO metal ring on one leg and two colour-rings on the other. In almost all cases, combinations were organized such that the colours on the “non-metal” leg were a unique identifier, alone, for the site concerned. As well as the birds ringed during the current study (2004-05 to 2006-07), additional, individually colour-ringed individuals were present in the local populations at three sites as a result of previous research there in 2002-03 and 2003-04.

Winter resighting

All food patches were monitored by means of twice-weekly 30-minute bird observation periods from November to mid-April, during which any colour-ring combinations seen were recorded. Additional patch watches dedicated to recording ring combinations were also conducted, because it was often difficult to follow and to identify particular individuals in dense flocks while also counting them. A few resightings were also made away from food patches, in both fed and control areas, but it was generally prohibitively difficult to determine ringing status (ringed or unringed) and ring combinations accurately and in a time-efficient manner where birds were not readily visible on predictable food resources on the ground.

Multiple sightings of particular individuals were frequently made during single winters. For the purposes of survival analysis, these were condensed to single records per winter. While some birds undoubtedly died in a given winter after having been sighted in the same winter, considering this in models of survival would add considerable complexity while also adding uncertainty because death in the winter concerned could not have been separated from “emigration” (simply not being seen at food patch again).

Spring resighting

Resighting effort during the breeding season aimed to identify the status (ringed/unringed) of all territorial males and of as many females as possible (when encountered) of the target, colour-ringed species, and to determine the colour-ring combinations worn by all ringed birds seen. A focus was placed on males because their behaviour made identification of all individuals present achievable: females do not often perch in exposed locations and spend significant periods incubating eggs and brooding chicks, so are not visible. Initial data on target species location and ringing status/combination were obtained during regular surveys using Breeding Bird Survey (BBS) methods to measure breeding abundance (see Appendix 9). These surveys were then followed with intensive searches for the target species in May-June. All suitable habitat was searched by walking along field boundaries at a steady pace, recording all target birds encountered on a map and observing these individuals until status and, if ringed, colour-ring combination had been determined or the focal bird flew away. Playback of song using portable cassette recorders or MP3 players was used to draw birds out of cover. A second visit was then made to each study area later in June, during which any apparent territories identified during BBS or earlier resighting visits where male ringed status or combination was uncertain were checked thoroughly. Data from all visits were then plotted on maps to reveal, to the closest approximation possible, all territory locations in each study tetrad and the status/combination of the birds in each territory.

In addition to the intensive searches of study tetrads, up to four, additional ring-resighting transects reaching up to about 5km away from focal tetrads were walked during late June and early July in 2006 and 2007, with the aim of sampling more distant populations for birds that had dispersed from winter feeding areas were ringing had taken place. Transect routes were walked at a steady pace and the status of any birds seen was recorded, as before, using song playback if necessary. Territory locations could not be confirmed using data from single visits, but because territorial behaviour contributed to the detection of most individuals’ presence and facilitated the determination of their status/combination, most data from the transect searches will have related to territorial birds. Note that the timing of these surveys was such that chaffinches were probably under-recorded: yellowhammers were generally still strongly territorial into July, with males singing actively, but chaffinches had generally stopped singing and were much more furtive, often with attendant fledged juveniles.

Analyses of survival rates

The combination of trapping and colour-ring resighting in all 20 tetrads in winter and spring each year allowed analyses of variation in survival rates between fed and control treatments. Analyses are reported for chaffinches and yellowhammers only because these species had the largest colour-ringed samples and therefore the largest resighting samples. Reed buntings were only rarely resighted in spring and goldfinches not at all. Analyses using recaptures only may be possible for other commonly caught species, such as blackbird, dunnock, robin, greenfinch, blue tit and great tit, but have not been attempted to date because of time constraints, and are unlikely to yield useful results for comparisons of fed and unfed areas because of low recapture rates in control areas (see below).

Ring recoveries and external reports

During the course of the study, reports of dead ringed birds and sightings of colour-ringed birds were received from members of the public for several species. In addition, several birds ringed during the study were caught elsewhere by other ringers and several birds trapped were found to have already been ringed elsewhere. These records provide anecdotal information about scales of movement outside the study tetrads but cannot easily be incorporated in survival analyses because “capture” dates did not necessarily fall within the periods used for formal sampling, so inclusion of these data would also add considerable heterogeneity. Further, although extensions to standard capture-recapture survival models that allow the inclusion of dead recoveries (as well as captures and sightings), such extended models are considerably more complex, involving the estimation of several additional parameters, so there are unlikely to be significant benefits to fitting them to the dataset here, in which a maximum of only six recoveries were available for any single species.

Survival model structure

For analysis, data were formatted as individual encounter histories, with each bird being assigned a value of one (seen or captured) or zero (not encountered) for each period considered. “Winter” ringing was conducted between mid-December and the first week of April each year; winter resighting nominally took place from November, but effectively occurred from January onwards because this was when finches and buntings were attracted to food patches in significant numbers, making resighting possible. Encounters in the breeding season consisted of resighting events between May and July, while encounters in winter consisted of both recaptures and resightings. In reality, resighting was much more likely in fed areas because many birds were sighted during semi-weekly observations of food patches, concentration around these food sources is very likely to have affected capture probabilities and resighting in general probably had a different associated probability to recapture. However, all models incorporated a term allowing dependence of encounter probability on site type (treatment), which (i) controlled for this variation in analyses of variation in survival and (ii) allowed its significance to be tested formally. Further, all models incorporating both winter and spring encounters allowed explicitly for different encounter probabilities in spring and winter.

Models were, therefore, structured as Cormack-Jolly-Seber mark-recapture models for open populations (Cormack 1992). The basic parameterization consisted of apparent survival probabilities () and “recapture” (encounter) probabilities (p). A parameter represented survival from one trapping/sighting occasion to the next and a p parameter the probability that a bird alive at the end of a given survival period was encountered (trapped or sighted). Note that survival was “apparent” because permanent emigration from study areas is indistinguishable from mortality in models of this type.

Having two discrete periods in each year when birds were encountered by sighting and/or trapping allowed a range of approaches to modelling survival. At the highest level of detail, the study considered two biologically different periods within each year, one between winter trapping/resighting (mostly January-March) and spring resighting (May-July) and one between spring and the following winter (Fig. 7.1). Survival between winter and spring (a short period, but potentially when ambient food availability was lowest) was fundamentally different from that between spring and winter (a much longer period), so these two forms of survival rate were always allowed to vary independently in fitted models.

Multiple capture or sighting events within a single encounter period were not separated, i.e. any record of a bird in a period was assumed to indicate that that individual was alive at that time, effectively assuming for analytical purposes that all data referred to a notional single “capture” occasion. This is standard practice in analyses of survival, with the proviso that these periods should be kept as short as possible, effectively assuming that (re)captures occur at the mid-point (or median capture date) of a given encounter period, such that estimated survival probabilities refer to the period between one encounter period mid-point and the next (e.g. Siriwardena et al. 1998, 1999, Thomson et al. 1997).

Full survival models

The most detailed models fitted incorporated different apparent survival and encounter probabilities for each of the winter-spring and spring-winter periods for each of fed and control areas (Fig. 7.1). Models were fitted both omitting and including the sample of birds ringed before 2004-05. Where the latter were included, a dummy survival rate for the period before the first winter of the current study was included in the model. This allowed previously ringed birds that were encountered later to contribute to estimates of survival probabilities for periods before they were first encountered in the main experimental study, without introducing bias.

With “type” denoting the experimental comparison of fed and control areas, “WS” winter-spring, “SW” spring-winter, and “pre04” the period before 2004-05, the most complex models fitted were:

pre04,WS(type), SW(type), ppre04,pWS(type), pSW(type) (model A1)

and

WS(type), SW(type), pWS(type), pSW(type) (model B1).

Each term (except pre04, which was always just a constant) was simplified from a two-level factor (type=fed or type=control) to a constant (denoted c) in turn and models featuring all combinations of parameters (type-dependent or constant) were fitted. In reality, there must have been some spatial and temporal heterogeneity (variation between sites and years) in both forms of survival probability and probably also encounter probabilities, but the data would be unlikely to support more highly parameterized models, such as those incorporating annual survival probabilities. These sources of variation just represent noise around the experimental comparison, however, rather than bias, and are not relevant to the effects of supplementary feeding on survival.

Other survival models

The full model parameterization described above involved the estimation of multiple survival and recapture probabilities and did not necessarily provide the most efficient approach to investigating the variation in annual survival between fed and control areas from the available data. In particular, it could have been beneficial to estimate single, annual survival probabilities from all of the available data. This meant combining encounter data from winter and spring, assuming that the two periods together constituted a single encounter period in each year, and effectively assuming that there was no mortality within this period. The most obvious way to do this was to combine winters with the subsequent spring periods, because there was a relatively small period between them. This allowed survival estimation to be based on three encounter occasions and two full years (plus a dummy “pre04” survival rate if the earlier data were also included; Fig. 7.1). The most complex model considered here was, therefore the relatively simple

pre04,(type), ppre04,p(type) (model A2)

or

(type), p(type) (model B2).

However, this meant discounting any possible mortality during the late winter (February-April) period that is a strong candidate for the major period of food stress for farmland seed-eaters (Siriwardena et al. 2008). A second analysis was therefore also conducted in which each spring period was combined with the subsequent winter period (Fig. 7.1). This analysis was weaker, because the three periods over which survival could be estimated were of different effective lengths (1-3, Fig. 7.1), so could not be combined (whilst retaining biological sense) to estimate a single survival parameter and thus maximize statistical power. (Note, however, that two of the encounter periods were equivalent in terms of duration and range of encounter types considered, so could be modelled as one parameter.) Thus, the model used here was

pre04,1(type), 2(type), 3(type), ppre04, p1(type), p2(type) (model A3)

or

1(type), 2(type), 3(type), p1(type), p2(type) (model B3).

More simple modelling approaches, which involved the estimation of fewer parameters and avoided the use of lengthy encounter periods, but also omitted significant proportions of the data, considered only the winter or spring encounter periods (Fig. 7.1). Both of these approaches based survival estimation two full years of data and three encounter occasions (plus the pre-2004-05 data, where included), but the definitions of the annual period over which survival was measured differed (Fig. 7.1). In both cases, the most complex model considered was the simple

pre04,(type), ppre04,p(type) (models A4 [winter] and A5 [spring])

or

(type), p(type) (models B4 [winter] and B5 [spring]).

As with the full survival models, all terms in each model were simplified to a constant in turn to remove the various postulated effects of the experimental dichotomy, except for models A3 and B3, where all apparent survival parameters f and recapture parameters p were each all set either to vary with site type or to remain constant.