...

SC 11419 ADVISORY COMMITTEE ON PESTICIDES ENVIRONMENTAL PANEL

by user

on
Category: Documents
41

views

Report

Comments

Transcript

SC 11419 ADVISORY COMMITTEE ON PESTICIDES ENVIRONMENTAL PANEL
SC 11419
ADVISORY COMMITTEE ON PESTICIDES
ENVIRONMENTAL PANEL
MEETING DATE: 2 MARCH 2006
BIRD AND MAMMAL RISK ASSESSMENT: REFINING THE PROPORTION OF
DIET OBTAINED IN THE TREATED CROP AREA (PT) THROUGH THE USE OF
RADIO TRACKING DATA
Authors: Mrs Elizabeth Finch & Mr Mark Payne
Appendices provided by Dr Joe Crocker
AIM OF PAPER
On behalf of PSD, the UK’s Central Science Laboratory (CSL) has conducted radio
tracking studies on some commonly occurring species of birds in arable and orchard
crops. This paper demonstrates how, by making certain assumptions, the proportion
of diet obtained in treated areas (PT) by birds can be estimated from the recorded
radio tracked ‘active time’ data and discusses the most appropriate percentile value
for regulatory use in the current deterministic non-target bird risk assessment for
pesticides. It is also proposed to include the derived PT values in WEBFRAM
models where the whole distribution will be used to determine the risk in a
probabilistic manner.
It should be noted that the focus of attention in this paper is the estimation of PT for
populations of possible key or focal species of birds which are found to use the crop
situations under assessment and which therefore have the potential to forage in
these areas. Pesticide risk assessments that use these PT estimates will therefore
relate to the risk to the locally occurring population rather than the general farmland
population which may include unexposed individuals that do not occur in the treated
areas. This is considered to be in line with the assessment approach described in
the Guidance Document on Risk Assessment for Birds and Mammals
(SANCO/4145/2000).
It is proposed that the values obtained for PT from this radio tracking data will be
made available to applicants, via the PSD website, as part of the Checklist for Bird
and Mammals risk assessments (SC11418), for use in the refinement of bird risk
assessments. Other key species for which radio tracking is on-going will be added
once these studies have been completed.
INTRODUCTION
The bird and mammal risk assessment is based on a deterministic approach,
whereby the appropriate toxicity end point is divided by the estimated theoretical
exposure (ETE) to produce a toxicity:exposure ratio (TER). If the TER exceeds the
relevant trigger value then it is considered that the risk is acceptable. If the TER is
1
below this value, then refinement of the risk assessment is required (see
SANCO/4145/2000 for details).
The ETE is calculated via the following equation:
ETE = (FIR/bw) x C x AV x PT x PD (mg/kg bw/d)
Where: FIR = food intake rate of indicator species (g fresh weight/d)
bw = body weight (g)
C = concentration of compound in fresh diet (mg/kg)
AV = Avoidance factor (1=no avoidance, 0=complete avoidance)
PT = Fraction of diet obtained in treated area (1=all of diet, 0=none of diet)
PD = Fraction of food type in diet (1=all diet of one type, 0=none of diet)
At the first tier assessment, a worse case approach is taken where AV, PT and PD
are set at 1, i.e. an animal does not avoid any contaminated food, it obtains all of its
diet from the treated area, and all of its diet is made up of one type of contaminated
food. This may significantly over-estimate the exposure.
In refining the risk assessment it is possible to include more realistic values for AV,
PT and PD, provided that appropriate data are available with which to do so. The
following paper concerns itself with the refinement of PT (fraction of diet obtained
within the treated area).
SANCO/4145/2000 (Section 5.6) states that PT can be estimated from the time spent
active in a treated area, assuming that the active time spent in a habitat is a reliable
indicator of the measure of food obtained there. It is considered that one of the most
appropriate means of obtaining these data is through radio tracking, with such
studies also ideally including visual observations of tracked birds / mammals to help
back up and provide further behavioural details in relation to measured ‘active’ and
‘non-active’ time.
CSL have conducted several radio tracking programmes in which the activity of
commonly occurring bird and small mammal species were tracked in arable and
orchard landscapes (DEFRA project codes PN0915 & PN0903). This document is
only concerned with the data collected on birds as it was considered that further work
was required on the wood mouse to take account of the possible presence of
resident populations occurring within crops. The following summarises the radio
tracking methodology used by CSL and describes how these data have been
interpreted with regards to the estimation of PT, including any assumptions made.
SUMMARY OF RADIO TRACKING STUDY METHODOLOGY
Before radio tracking can be carried out it is important to identify those species which
are relevant to the particular crop/situation in question. These are known as the
‘focal’ or key species. In determining which bird species should be chosen for radio
tracking in the arable environment, the prevalence and frequency of species within
this setting were first established by carrying out censuses in different crops and in
the different seasons (Crocker & Irving 1999). From these censuses it was
2
considered that the most relevant UK arable crop small bird species were the
yellowhammer, skylark and linnet.
In fruit orchards, results of mist netting conducted by CSL indicated blackbird,
chaffinch, blue tit and robin were the most frequently occurring and abundant
species. These species were therefore tracked in the orchard study.
The table below gives details of the focal species observed in the separate arable
and orchard (apple) radio tracking studies, including details of their dietary
requirements and the crop habitats for which PT was assessed.
Table 1:
Details of the focal species of birds for which PT was assessed in the
arable and orchard radio tracking studies
Relevant
Crops/situations
Cereals
Focal species
Dietary details*
Skylark
Omnivorous, including insects, vegetation
and seeds
Yellowhammer Predominantly insectivorous (summer
months)/granivorous (winter months)
Beet/potatoes
Skylark
Omnivorous, including insects, vegetation
and seeds
Yellowhammer Predominantly insectivorous (summer
months)/granivorous (winter months)
Linnet
Granivorous including crop and/or weed
seed.
Oilseed rape
Skylark
Omnivorous, including insects, vegetation
and seeds
Yellowhammer Predominantly insectivorous (summer
months)/granivorous (winter months)
Linnet
Granivorous including crop and/or weed
seed.
Blackbird
Omnivorous including insects, earthworms
and seeds
Orchards
Blackbird
Omnivorous including insects, earthworms
and seeds
Blue tit
Insectivorous
Chaffinch
Omnivorous – insects in summer;
predominantly seed at other times.
Robin
Insectivorous
*From Buxton et al (1998).
For birds, radio-tags were mounted on the base of the tail feathers. Tags were
chosen to suit the size of the species being tracked, with tag weight being no more
than 10% of the typical species body weight in the non-orchard studies and 5% in the
orchard studies. Data were collected in orchards from April to September and in
non-orchard situations in both summer and winter months.
The aim of the work was to capture a representative day in the life of each tagged
individual. The day was divided into 2 hour slots, with the aim being to obtain at least
3
one hour’s contact with a tagged individual in each slot. The size of the tags meant
that the battery life was limited to a few days, so once an individual was tagged it was
followed intensively. During a typical monitoring session a single individual was
tracked continuously. The tracker kept close enough to the individual to be sure of its
whereabouts and, if possible, to see it, but not so close as to disturb it. In the
orchard environment this resulted in an observation distance of 20-50 metres. At
frequent intervals, notes were made of the individual’s location, the habitat occupied
and its activity. Weather and temperature were also recorded.
Although for much of the time it was difficult to see what tracked animals were doing,
it was usually possible to determine from small fluctuations in the signal and shifts in
location whether a tracked animal was active or not. For the purposes of the studies,
‘active’ was taken to mean ‘potentially foraging’. This category included all instances
of recorded foraging and excluded all instances where the animal was known to be
performing some other activity (e.g. singing, nest building) or where it was
considered to be inactive.
DATA ANALYSIS
Please refer to SC11411 for details of the raw data used in the following data
analysis.
In order to derive estimates of the proportion of total diet obtained in each scenario in
which bird species were tracked, the following assumptions and criteria were applied:
i)
The ‘active time’ data recorded for each scenario are taken to represent
the time spent foraging in these situations. The observational information
obtained whilst radio tracking generally supports this relationship, with
foraging being the predominant ‘active’ behaviour. Also all instances
where the animal was known to be performing some other activity (e.g.
singing, nest building) were excluded from the ‘active time’ data.
ii)
The derived data for time spent foraging (based on ‘active time’ data) is
considered to provide a measure of the relative amount of diet consumed
in each of the recorded scenarios.
iii)
Inactive time data are excluded from the data set used to calculate PT,
since by definition individuals cannot forage while they are inactive.
iv)
By comparing for each radio tracked individual, the ‘active time’ in a
particular scenario with the total recorded ‘active time’, the proportion of
total ‘active time’ spent in that situation is obtained, this being equivalent
(by extrapolation) to the proportion of diet consumed (i.e. PT).
v)
As is the nature of ecological data collection it was not possible to fill every
2 hour time slot for all radio tracked individuals. In addition, it normally took
2 to 3 days to collect a full day’s contact data. As a result, it was
necessary to determine a minimum time needed to estimate PT. From
assessing the relationship between PT and the total amount of monitored
contact time it was concluded that after an hour or two PT starts to stabilise
4
in the majority of cases. In plotting this relationship it appears that after
about 150 minutes contact time, PT becomes relatively stable for most
individuals (see Appendix 1 for details). As a result, the analyses that
follows excludes individuals for which there was less than 150 minutes
contact time.
vi)
Data on individual tracked birds are only included in the PT estimations
below where they were active for 30 minutes or more in the particular
situation under assessment, i.e. they were considered to be potential
‘consumers’. The full analysis for all individuals, regardless of their use, is
included in Appendix 4 for information only.
vii)
There was found to be very little difference in the distribution of PT
between data collected on a single day for an individual and data collected
over several days. As a result, it was considered that as the multiple days
data gave a fuller coverage of bird habitat use throughout the day this
should be used in preference to data collected for only part of a single day
(See Appendix 2 for details).
viii)
In relation to bird feeding preferences in orchards, PT estimates were
made for each bird species over the ‘summer’ radio tracked period of April
to September. For the non-orchard situations in which bird species were
tracked throughout the year, separate PT estimates were made for the
summer and winter period (with the precise months included in each
varying slightly according to the scenario being assessed).
ix)
The 90th and 95th percentiles for the data sets were estimated by assuming
that the data is a random sample from a parent distribution with known
mathematical properties. As the data are limited between 0 and 1, the
normal distribution does not provide a good fit for these proportional data.
A more appropriate distribution is the Beta which is bounded between 0
and 1 and is flexible enough to represent the different shapes seen in the
radio tracking data, including unimodal (a peak anywhere between 0 and
1) and bimodal (peaks at 0 and 1).
Two methods were tried for fitting Beta distributions to the radio tracking
data: maximum likelihood and method of moments. In both cases, 90%
confidence intervals were estimated for the distribution, to take account of
sampling uncertainty due to the varying size of the datasets (see Appendix
3 for details of methods used). Visual examination of graphs of the
distributions by fitted maximum likelihood showed that for many datasets
the fit was poor, with many data points falling outside the fitted confidence
interval. Graphs of distributions fitted by the method of moments showed a
satisfactory fit for every dataset (see Appendix 3 for further details and
graphs). Simulation studies by Frey et al (1999) also showed that the
method of moments performs better than maximum likelihood for Beta
distributions. The method of moments was therefore used to generate the
results shown below.
5
Median estimates for the 90th and 95th percentiles, together with their 90%
confidence intervals, were generated for each dataset. Results for
‘consumers only’ (those birds which visited the relevant crop during the
tracking period) are tabulated below (see Appendix 4 for the analysis of the
whole data set).
RESULTS
The following tables show the 90th and 95th percentiles for the proportion of active
time spent by potential ‘consumers’ within the crops identified (see point vi of Data
Analysis), accompanied by the relevant confidence limits for 5% (lower bound) and
95% (upper bound). It is important to note that the confidence limits relate only to
sampling uncertainty (i.e. the effect of limited sample size on estimation of the
distribution). Please refer to the Data Analysis section above for details of how these
figures were derived.
Cereals
Table 2:
Calculated 90th and 95th percentiles for PT values in cereal crops
Season
Species
Summer
(Apr-Aug)
Skylark
No. of
individuals
26
Yellowhammer 17
Winter (Sep- Skylark
10
Mar)
Yellowhammer 10
90th percentile
(CL)
0.97
(0.86 – 1.00)
0.87
(0.67 – 0.99)
0.94
(0.59 – 1.00)
0.14
(0.09 – 0.23)
95th percentile
(CL)
0.99
(0.94 – 1.00)
0.95
(0.79 – 1.00)
0.98
(0.72 – 1.00)
0.18
(0.11 – 0.30)
The figures in bold highlight those scenarios where the upper confidence limit is 1.00, which is the
maximum value for PT.
CL – upper and lower 95% confidence limits for each percentile value.
Skylark and yellowhammer forage intensively in cereal crops during the summer.
During the winter, only the skylark makes extensive use of cereal crops. As a result,
these are considered to be the appropriate species for refinement of the risk
assessment.
Beet/Potatoes
Table 3:
Calculated 90th and 95th percentiles for PT values in beet/potato crops
Season
Species
Summer
(Apr-Nov)
Skylark#
No. of
individuals
18
Yellowhammer 13
Linnet*
11
6
90th percentile
(CL)
0.88
(0.68 – 0.99)
0.94
(0.76 – 1.00)
0.59
(0.38 – 0.84)
95th percentile
(CL)
0.95
(0.80 – 1.00)
0.98
(0.85 – 1.00)
0.69
(0.47 – 0.92)
#
Skylark data are very limited for potatoes and given the lack of crop palatability to herbivorous birds
an insectivorous bird such as the yellowhammer is considered a more appropriate focal species for
risk assessment purposes.
*Linnet is totally granivorous and therefore it is only appropriate to consider this species when a risk to
seed-eating birds has been identified.
CL – upper and lower 95% confidence limits for each percentile value.
The figures in bold highlight those scenarios where the upper confidence limit is 1.00, which is the
maximum value for PT.
The selection of focal species for use in the regulatory refined risk assessment will
depend on the feeding group of birds identified at potential risk in the first tier risk
assessment. The skylark would be a suitable herbivorous species and the
yellowhammer a suitable insectivorous species.
Oilseed rape
Table 4:
Calculated 90th and 95th percentiles for PT values in oilseed rape
Season
Species
Summer
(Apr-Jul)
Skylark
No. of
individuals
7
Yellowhammer 7
(Apr-Aug)
Linnet*
6
Blackbird**
9
(May-Jul)
Winter (Aug- Skylark
4
Mar)
Yellowhammer 4
(Sep-Mar)
90th percentile
(CL)
0.57
(0.42 – 0.80)
0.86
(0.65 – 1.00)
0.99
(0.61 – 1.00)
0.98
(0.84 – 1.00)
0.98
(0.38 – 1.00)
0.61
(0.15 – 1.00)
95th percentile
(CL)
0.64
(0.47 – 0.87)
0.92
(0.73 – 1.00)
1.00
(0.73 – 1.00)
0.99
(0.90 – 1.00)
1.00
(0.51 – 1.00)
0.79
(0.22 – 1.00)
The figures in bold highlight those scenarios where the upper confidence limit is 1.00, which is the
maximum value for PT.
*Linnet is completely granivorous and therefore should only be considered where a risk to seed-eating
birds has been identified at tier 1.
**Blackbirds were observed by research workers at CSL to be frequently associated with maturing
oilseed rape crops and for this reason a limited radio tracking study was conducted including a total of
ten tracked individuals all caught and tagged close to or within oilseed rape fields. PT values derived
from the radio tracked active time data collected between the months of May to July are presented in
the table above.
CL – upper and lower 95% confidence limits for each percentile value.
During the summer months, blackbirds, linnet and yellowhammer intensively foraged
in oilseed rape. However, during the winter, more use of this crop was made by
skylark. As stated above, the selection of focal species for use in the regulatory
refined risk assessment will depend on the feeding group of birds identified at
potential risk in the first tier risk assessment.
7
Orchards
Table 5:
Calculated 90th and 95th percentiles for PT values in orchards
Season
Species
Summer
(Apr-Sep)
Blackbird
No. of
individuals
28
Blue tit
16
Chaffinch
29
Robin
24
90th percentile
(CL)
0.75
(0.60 – 0.89)
0.58
(0.42 – 0.79)
0.76
(0.63 – 0.88)
0.56
(0.44 – 0.70)
95th percentile
(CL)
0.84
(0.71 – 0.95)
0.68
(0.51 - 0.88)
0.85
(0.73 – 0.95)
0.66
(0.53 – 0.80)
CL – upper and lower 95% confidence limits for each percentile value.
N.B. These data differ from those in SC11411 as the previous analysis was conducted using not only
individuals using the orchard but also those using the perimeter. In this subsequent paper it is
considered that only those individuals using the orchard are relevant.
In the orchard scenario it was observed that the blackbird and chaffinch were more
active, compared with the blue tit and robin which used the orchards for a similar
amount of time. However, these species use orchards differently, e.g. if there is a
risk to earthworm-eating birds then it is appropriate to use blackbird as the focal
species, but if the risk is to insectivorous species then the chaffinch, blue tit and robin
would be more appropriate.
DISCUSSION/CONCLUSIONS
Using the radio tracking data the proportion of diet likely to be obtained in treated
crops (PT) has been estimated for several commonly occurring bird species. It is
considered that radio tracking accompanied by observational data is considered the
most appropriate methodology for refining PT at this stage. However, it may be in
the future that there are other techniques available which are just as valid. Whatever
methodology is used for the refinement of PT, the quality of the data must be
comparable with that collected as part of the studies detailed in this paper.
Choice of percentile value for PT
The selection of the most appropriate percentile value for PT for use in the regulatory
risk assessment is considered to be a risk manager’s decision according to what
level of protection needs to be afforded. At present there is no agreed value for the
required or desired level of protection. It is proposed that a 90th percentile PT value
is used for regulatory purposes.
There are several approaches which can be taken in dealing with the chosen end
point. The following outlines these:
Approach 1: Decide on a percentile and use the mean value for PT
Although this approach is very simple as you can simply pick the PT values from the
tables presented above, there is no consideration of any of the uncertainties
surrounding PT.
8
Approach 2: Decide on a percentile and use the 95% confidence limit (upper
confidence bound)
Again, this approach is very simple and the values are provided in the tables
presented above. As the upper confidence bound has been chosen, uncertainty
regarding PT in terms of sample size has been accounted for. It does not, however,
take into account any other uncertainties which may attribute to PT.
Approach 3: Decide on a percentile and consider all of the uncertainty factors
Table 1 in Appendix 3 gives details of all the types of uncertainty which may affect
PT. From this table it can be seen that the uncertainties have not been quantified,
and to do so would be very subjective, difficult and time consuming, but their
combined impact on PT has been qualitatively assessed.
Of the three approaches presented above, it is felt that Approach 1 is not appropriate
as it does not consider any of the uncertainties regarding PT. Approach 3 does
address all of the uncertainties, however, due to the state of knowledge it can only at
present be considered qualitatively, and therefore is not a viable choice. As a result,
it is proposed to use Approach 2 as this deals with some of the uncertainty regarding
PT in terms of sample size.
As stated above, as there is no agreed value it is proposed to select a 90th percentile.
If Approach 2 is applied, then this will mean that the risk to 90% of the population will
be assessed with 95% confidence. Using this approach, a summary of the values
which are considered appropriate for risk assessment purposes are detailed in Table
6 below.
It should be noted that this paper only considers PT and does not discuss the use of
different percentiles with regards to refining other factors within the equation
surrounding the calculation of ETE. Ideally we should refine all parameters equally
whilst maintaining the same level of protection and this will be considered further as
part of the WEBFRAM 3 project. However, in selecting the 95% confidence limit of
the 90th percentile it is felt that this should provide an appropriate level of protection,
as long as other factors are refined appropriately to maintain this level of protection.
Risk assessment implications of derived PT values
The results presented show that in some scenarios, e.g. cereals and oilseed rape,
the radio tracking data has identified very little difference between the PT default
value of 1 and the more realistic values calculated, implying that although this is a
significant refinement of the risk assessment there is little scope for reducing the
estimated theoretical exposure value. This also implies that in these scenarios the
results confirm the appropriateness of using 1 in the tier 1 risk assessment.
Use of the derived end points
From the tables above it is anticipated that Notifiers/applicants will be able to identify
for the assessed crop scenarios the appropriate focal species based on which
‘feeding guilds’ (or generic ‘indicator species’) failed the first tier risk assessment
under SANCO/4145/2000. Therefore, for example, if there is an issue with
9
insectivorous birds in beet/potatoes at tier 1, then an appropriate focal species to
consider in a refined risk assessment would be yellowhammer and if a risk to
granivorous species has also been identified, linnet as well. The following table
summarises the focal species for each scenario and also the derived 90th percentile
upper confidence limit (95%) PT values rounded to two decimal points:
Table 6: Summary of the crop scenarios, focal species and the respective 90th
percentile upper confidence limit (95%) PT values
Scenario
Species
Most relevant feeding
guild/s
Summer cereals
Yellowhammer
Skylark
Skylark
Skylark#
Yellowhammer
Linnet
Yellowhammer
Skylark
Blackbird
Insectivorous
Insectivorous/herbivorous
Insectivorous/herbivorous
Insectivorous/herbivorous
Insectivorous
Granivorous
Insectivorous
Insectivorous/herbivorous
Insectivorous/earthwormeating
Granivorous
Insectivorous/herbivorous
Insectivorous
Earthworm-eating
Insectivorous
Granivorous/insectivorous
Insectivorous
Winter cereals
Summer
beet/potatoes
Summer oilseed
rape
Linnet
Winter oilseed
Skylark
rape
Yellowhammer
Summer orchards Blackbird
Blue tit
Chaffinch
Robin
PT (90th
percentile, 95%
confidence limit)
0.99
1.00
1.00
0.99
1.00
0.84
1.00
0.80
1.00
1.00
1.00
1.00
0.89
0.79
0.88
0.70
#
Skylark data are very limited for potatoes and given the lack of crop palatability to herbivorous birds
an insectivorous bird such as the yellowhammer is considered a more appropriate focal species for
risk assessment purposes.
It should be noted that the selection of focal species for use in the regulatory refined
risk assessment will depend on the feeding group of birds identified at potential risk
in the first tier risk assessment. For example, in beet/potatoes and oilseed rape the
linnet is only an appropriate species where a risk to granivorous birds has been
identified at tier 1.
Applicability to acute, short-term and long-term risk assessments
It is considered that the PT values quoted are appropriate for acute, short-term and
long-term risk assessments. Although these values are based on a theoretical ‘day
in the life of’ a bird, the data were collected over several days and a mean for the 90th
percentile was calculated based on the distribution of the available data. In addition,
if individuals were observed for a longer period of time it is likely that the PT values
could be lower due to a greater diversity of suitable alternative food sources being
available. As a result, using these data for the acute, short-term and long-term
assessments is deemed to be precautionary and appropriate given the available
information.
10
Suitability of extrapolations to other scenarios
One further point, which would be of value to applicants, is the ability to make use of
suitable extrapolations of PT values to other crops. Of course, any extrapolation
needs to be appropriately justified. At the moment, it is considered that
extrapolations to other cereal crops from wheat and barley would be appropriate, as
would extrapolating between different orchard crops. In determining whether further
extrapolations are appropriate it is necessary to consider whether a crop is being
used in the same way, e.g. potato leaves are not palatable to birds and therefore only
insectivorous species are required to be assessed, however, this is not the case for
the majority of leafy crops in which herbivorous species would also be exposed. As a
result, if an extrapolation is proposed, it must be fully justified and may require some
data to confirm that such an extrapolation is acceptable.
It should be noted that it is hoped that some further research work will be undertaken
by CSL with regards to extrapolations from the major crop groups to minor crops in
the future.
Further work and the appropriateness of adding other data sets
CSL are currently collecting data on herbivorous birds (greylag geese, wood pigeon)
and mammals (brown hare) and it is proposed that these data are dealt with in the
same way as described above, although these data will be sufficiently different from
those collected for small birds and therefore may require different statistical
methodologies to be employed. In addition, it is proposed that further work is done
on the wood mouse, ensuring that where resident populations exist these are radio
tracked. When these studies are complete and the data analysed, they will be
presented to the Panel for consideration.
With regards to adding data from other studies to this extensive and comprehensive
data set, it is acknowledged that this is a possibility. However, care will be needed to
ensure that any additional data stands up to rigorous scientific scrutiny with data
being derived from similar well conducted and analysed studies.
Acknowledgements
The authors would like to acknowledge the help and co-operation of CSL colleagues
Dr Phil Prosser, Dr Joe Crocker, Willem Roelefs and Dr Andy Hart in preparing this
document.
References
Anon (2006): SC11418 - Checklist of issues to note when carrying out a bird and
mammal risk assessment according to SANCO 4145/2002/EEC – Version 2.
Buxton JM, Crocker DR & Pascual JA (1998): 1998 Update - Contract PN0919
Milestone Report Birds and Farming: Information for Risk Assessment
11
Crocker DR & Irving PV (1999): Project PN0915 Improving estimates of Wildlife
Exposure to Pesticides in Arable Crops: Milestone report 02/01 Variation of bird
numbers on arable crops.
Frey, HC & Burmaster, DE (1999): Methods for characterizing variability and
uncertainty: comparison of bootstrap simulation and likelihood-based approaches.
Risk Analysis 19 (1) 109-130.
Gurney JE, Perrett J, Crocker DR & Pascual JA (1998): 1998 Update - Contract
PN0910 /PN0919 Milestone Report Mammals and Farming: Information for Risk
Assessment
Pascual J, Crocker J & Hart A (1998): Project PN0919 Improving estimates of the
exposure of non-target wildlife to Pesticides in Arable Crops – A review of existing
data
Payne M & Finch E (2006): SC11411 – Bird and Mammal Risk Assessment: Refining
the proportion of diet obtained in the treated crop area (PT) through the use of radio
tracking data.
SANCO/4145/2000 (2002): Guidance Document on Risk Assessment for Birds and
Mammals Under Council Directive 91/414/EEC.
12
Appendix 1: Justification for a minimum tracking time for the
estimation of PT (details provided by CSL)
We wish to estimate the exposure to pesticides for different wildlife species in
typical UK agricultural habitats. Bird species are relatively mobile and may
visit several different habitats in a single day. Therefore we used radio
tracking to follow birds in the field and estimate what proportion of time they
spent in treated (agricultural crops) compared with untreated (hedgerow, setaside etc) habitats. Our aim was to capture a day in the life of individual bird
and to show how proportion of time spent in a pesticide treated area (PT)
varies between individuals
Because it was rarely possible to keep in constant contact with an active
individual we divided the day into 2-hour slots and attempted to follow a given
bird for 1 hour out of each 2 hour slot. We stopped tracking when we had data
for all slots. Depending on the time of year, this might amount to 4 to 8 hours
of contact time.
However, there are two important shortcomings of the data we collected and
they have contrasting consequences for our estimate of PT.
1. Incomplete data: we didn’t succeed in filling every slot for every bird
Small birds could be fitted only with small radio-tags with limited range and
lifetime, and birds often moved out of range so that that they were “lost” to the
tracker for several hours at a time.
This tends to exaggerate the extremes of the distribution of PT. The degree of
exaggeration will depend on the particular movement habits of the individual
but in the extreme case where we have say only 1 minute of contact time it is
clear, since the individual cannot be in two places at once, that it will either be
in treated habitat or it will not, and PT must be either 1 or 0. With more contact
time it is more likely that that more habitats will be visited and PT will less
extreme.
Ideally we would include only birds where we had a full day’s contact data.
Unfortunately our monitoring schedule precluded this: no birds were
monitored all day in a single day. Therefore we need to decide what is the
minimum amount of contact time that we will accept as a reasonable estimate
of PT.
2. Extended timescale: it usually took 2 or 3 days to collect a full day’s
contact data.
For example, a bird might be caught and tagged in the early afternoon,
tracked for 2 hours in the late afternoon/evening, and followed again the next
day until early afternoon.
Extending the monitoring period across several days will tend to
underestimate extreme values of PT. Again, it depends how often a bird
13
moves between habitats, but it seems likely that a bird is more likely to be
seen in different habitats if it has been watched for 1 hour on each of 5 days,
than if it were observed for 5 hours continuously on a single day. A bird may
be quite likely to spend all day in one field but less likely to spend several
days in the same field. Therefore we need to decide whether PT collected
over several days needs to be adjusted to more closely reflect PT for a single
day.
Minimum time necessary to estimate PT
Birds do not move instantly between habitats, but rather stay for some time in
one place before moving to another. An estimate of PT based on a very short
contact time is likely to be close to 1 or 0. As contact time increases, then for
birds that explore both cropped and non-cropped habitat, PT will tend to swing
between these extremes. The pattern of these swings will depend on the
length of typical residence time in relation to contact time. As contact time
further increases, PT (calculated as the cumulated active time spent within a
crop as a proportion of total active contact time) will inevitably become more
stable. Our focus for risk assessment purposes is the habitat preference
shown in a single day. The length of contact time required for fluctuations in
habitat preference to settle down gives us a criterion for rejecting - because
PT is too unstable - birds with less than this amount of contact time.
In Figures 1 and 2 the relationship between PT (proportion active time spent
in crop) against the total amount of contact time monitored through radio
tracking has been plotted for all individual birds followed in orchard and arable
habitat. Although not very easy to take in, these plots support the argument
above that: where contact time is very short PT will be close to 0 or 1 and may
fluctuate rapidly. After an hour or two of monitoring PT appears to stabilise for
most individuals under most habitats. Ideally, we would fit separate equations
to each individual plot and estimate the length of contact time by which PT
had reached a fixed proportion (e.g. 90%) of its asymptotic value. We might
then take the average of these values to establish an objective criterion for the
minimum quantity of monitoring that would give a reasonable estimate of PT.
Unfortunately, from my initial investigations, it is not very clear what general
form the equations should take (e.g. logistic, logarithmic) and how suitable is
this approach where individuals have different amounts of contact time? At
this stage it may be as well to judge by eye the point at which PT becomes
stable. Acknowledging a degree of subjectivity I would estimate that after
about 150 minutes contact time, PT appears to be relatively stable for most
individuals. Therefore, the analyses that follow will exclude birds for which we
had less than 150 minutes contact data.1
1
We additionally rejected data for 2 birds for which we had successfully collected contact time > 150
minutes but which had been inactive for nearly all that time. All birds included in the analyses had
contact time > 150 minutes and active time recorded of > 30minutes.
14
1
PT - Propn active time in orchard centre
PT - Propn active time in orchard centre
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
50
100
150
200
250
300
350
400
450
500
0
50
100
Blue tit - Contact time (min)
200
250
300
350
400
450
1
PT - Propn active time in orchard centre
1
PT - Propn active time in orchard centre
150
Blackbird - Contact time (min)
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
50
100
150
200
250
300
350
400
450
500
Robin - Contact time (min)
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
50
100
150
200
250
300
350
Chaffinch - Contact time (min)
Figure 1. PT in orchard centre as a function of time in radio contact.
15
400
450
500
1
0.9
0.9
PT - Propn active time in crop
PT - Propn active time in crop
1
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
100
200
300
400
500
0
100
200
300
400
500
400
500
Yellowhammer - Contact time (min)
1
1
0.9
0.9
PT - Propn active time in crop
PT - Propn active time in crop
Skylark - Contact time (min)
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.1
0
0
0
100
200
300
400
500
0
100
200
300
Blackbird - Contact time (min)
Linnet - Contact time (min)
Figure 2 PT in arable crops as a function of time in radio contact.
16
Appendix 2: Justification for averaging PT over several days (details
provided by CSL)
If birds change their habitat preferences from day to day, then averaging PT
across several days is likely to indicate less extreme preferences than would
restricting analyses to data collected on a single day. We looked for any such
effect by comparing the distribution of PT (collected for birds with a minimum
contact time of 150 minutes (see Appendix 1) based on observation periods
collected in the course of a single day, with PT collected from monitoring
across several days.
Figure 3 to 10 show the distribution of PT for species of birds in orchard and
arable scenarios in which they may be exposed to pesticides. Figures on the
left show distributions for the full dataset collected over several days. Figures
on the right show the distribution of PT collected for the same individuals for
the single day with greatest contact time (where minimum contact time >150
min).
If the extended timescale was affecting PT we would expect to see fewer
birds with PTs close to 0 or 1 in the figures on the left with multiple days data.
We looked for statistical evidence of the effect using the Moses test of
extreme reactions (Siegel, 1956). This is a non-parametric test designed to
highlight experimental conditions that may affect some subjects in one way
and others in the opposite way. In our case we are using it to test whether
restricting data to a single day causes some birds to show a PT closer to 0
and other closer to 1.
Data were also separately analysed for 2 ensembles of interest to the risk
assessor. Figures at the top of each page represent the full population of all
birds caught in the orchard or arable landscape with >150 min data. Figures at
the bottom show the sub-population of birds that additionally spent some
active time in the orchard or arable crop. If the target of the risk assessment
is to ensure that a pesticide has no unacceptable effects on the population of
birds that live in and around orchard or arable crops then the appropriate
ensemble is the full population of birds for which we have sufficient data. If
the focus of interest is in ensuring the safety of any bird that visits a pesticide
treated area then the appropriate ensemble should be ‘consumers-only’
excluding those birds that spent no active time in orchard or crop. The latter
ensemble will generally lead to more conservative assessments since it
assumes that all birds will be exposed to pesticide to some extent.
For the “consumers only” ensemble, statistical tests revealed no significant
differences between the multiple day and single day scenarios. For the “all
birds” scenarios the Moses test showed a significant difference between
multiple and single day scenarios for both yellowhammer and skylark in arable
crops. But this result is compromised by the large number of ties (where PT=
0) between the groups which may invalidate the result.
In general there appears to be relatively little difference in the distribution of
PT between data collected over several days and data collected in a single
17
day. Because the extended timescale over which observations were collected
tends to moderate extreme preferences, this to some extent will
counterbalance the tendency for shortfalls in the data (see above) to
exaggerate extreme preferences. Therefore since the multiple-days data
gives a fuller coverage of a bird habitat use throughout the day we will use
them in preference to data collected for only part of a single day.
18
Single day with most
contact time
Multiple days
Group: Max in one day, Species: BLABI
25
20
20
Frequency
All birds
caught in
orchards
Frequency
Group: All data, Species: BLABI
25
15
10
15
10
5
5
Mean = 23.3104
Std. Dev. = 27.13286
N = 40
0
0.00
20.00
40.00
60.00
80.00
Mean = 21.0658
Std. Dev. = 31.20379
N = 37
0
100.00
0.00
20.00
% active time in centre
40.00
60.00
80.00
100.00
% active time in centre
Species: BLABI, Group: Max in one day
Species: BLABI, Group: All data
25
25
20
Frequency
20
Frequency
Orchard
consumers
only
15
15
10
10
5
5
0
0.00
Mean = 33.3006
Std. Dev. = 26.80275
N = 28
20.00
40.00
60.00
80.00
100.00
0
0.00
Mean = 41.0229
Std. Dev. = 32.91156
N = 19
20.00
40.00
60.00
80.00
100.00
% active time in centre
% active time in centre
Figure 3. Distribution, for Blackbirds in Orchards, of PT (%active time in
centre). Data are presented where individuals have radio contact time >150
minutes.
“All birds caught in orchards” – includes the full data set of birds that were caught in orchards.
“Orchard consumers only” is a subset of the data that excludes those birds that were never
seen active amongst orchard trees.
“Multiple days set” includes PT values calculated from radio tracking data that may have
taken several days to collect.
“Single day with most contact time”- where data was collected over more than 1 day, PT is
calculated for the day that radio contact time was greatest (Birds are rejected where
maximum PT < 150 minutes)
19
Single day with most
contact time
Multiple days
Group: All data, Species: BLUTI
Group: Max in one day, Species: BLUTI
10
10
8
8
Frequency
Frequency
All birds
caught in
orchards
6
4
6
4
2
2
Mean = 21.2884
Std. Dev. = 21.65214
N = 20
0
0
20
40
60
80
Mean = 24.3838
Std. Dev. = 23.26622
N = 20
0
100
% active time in centre
0
20
40
60
80
100
% active time in centre
Species: BLUTI, Group: All data
Species: BLUTI, Group: Max in one day
10
10
8
6
Frequency
Orchard
consumers
only
Frequency
8
4
6
4
2
2
Mean = 26.6105
Std. Dev. = 21.04203
N = 16
0
0
20
40
60
80
Mean = 30.4797
Std. Dev. = 22.07892
N = 16
0
100
0
% active time in centre
20
40
60
80
% active time in centre
Figure 4. Distribution, for Blue tits in Orchards, of PT (%active time in centre).
Data are presented where individuals have radio contact time >150 minutes.
“All birds caught in orchards” – includes the full data set of birds that were caught in orchards.
“Orchard consumers only” is a subset of the data that excludes those birds that were never
seen active amongst orchard trees.
“Multiple days set” includes PT values calculated from radio tracking data that may have
taken several days to collect.
“Single day with most contact time”- where data was collected over more than 1 day, PT is
calculated for the day that radio contact time was greatest (Birds are rejected where
maximum PT < 150 minutes)
20
100
Single day with most
contact time
Multiple days
Group: Max in one day, Species: CHAFF
Group: All data, Species: CHAFF
8
8
Frequency
Frequency
All birds
caught in
orchards
10
10
6
6
4
4
2
2
Mean = 31.8123
Std. Dev. = 27.34907
N = 33
0
0.00
20.00
40.00
60.00
80.00
0
0.00
100.00
Mean = 29.0654
Std. Dev. = 24.75653
N = 28
20.00
Group: All data, Species: CHAFF
10
8
8
6
4
2
0
0.00
60.00
80.00
100.00
Group: Max in one day, Species: CHAFF
10
Frequency
Frequency
Orchard
consumers
only
40.00
% active time in centre
% active time in centre
6
4
2
Mean = 36.0915
Std. Dev. = 26.36499
N = 29
20.00
40.00
60.00
80.00
100.00
% active time in centre
0
0.00
Mean = 33.9096
Std. Dev. = 23.41082
N = 24
20.00
40.00
60.00
80.00
% active time in centre
Figure 5 Distribution, for Chaffinches in Orchards, of PT (%active time in
centre). Data are presented where individuals have radio contact time >150
minutes.
“All birds caught in orchards” – includes the full data set of birds that were caught in orchards.
“Orchard consumers only” is a subset of the data that excludes those birds that were never
seen active amongst orchard trees.
“Multiple days set” includes PT values calculated from radio tracking data that may have
taken several days to collect.
“Single day with most contact time”- where data was collected over more than 1 day, PT is
calculated for the day that radio contact time was greatest (Birds are rejected where
maximum PT < 150 minutes)
21
100.00
Single day with most
contact time
Multiple days
Group: Max in one day, Species: ROBIN
Group: All data, Species: ROBIN
12
12
10
10
8
Frequency
8
Frequency
All birds
caught in
orchards
6
6
4
4
2
2
Mean = 20.3882
Std. Dev. = 20.68095
N = 29
0
0
20
40
60
80
Mean = 20.7601
Std. Dev. = 20.21231
N = 27
0
0
100
20
% active time in centre
60
80
100
Group: Max in one day, Species: ROBIN
7
7
6
6
5
5
Frequency
Frequency
Group: All data, Species: ROBIN
Orchard
consumers
only
40
% active time in centre
4
3
2
4
3
2
1
1
Mean = 24.6358
Std. Dev. = 20.28519
N = 24
0
0
20
40
60
80
100
Mean = 28.0262
Std. Dev. = 18.56594
N = 20
0
0
% active time in centre
20
40
60
80
% active time in centre
Figure 6. Distribution, for Robins in Orchards, of PT (%active time in centre).
Data are presented where individuals have radio contact time >150 minutes.
“All birds caught in orchards” – includes the full data set of birds that were caught in orchards.
“Orchard consumers only” is a subset of the data that excludes those birds that were never
seen active amongst orchard trees.
“Multiple days set” includes PT values calculated from radio tracking data that may have
taken several days to collect.
“Single day with most contact time”- where data was collected over more than 1 day, PT is
calculated for the day that radio contact time was greatest (Birds are rejected where
maximum PT < 150minutes)
22
100
Single day with most
contact time
Multiple days
Species: BLABI, Group: Max day
Species: BLABI, Group: All days
6
6
5
5
4
All birds
Frequency
Frequency
4
3
3
2
2
1
1
0
0.00
Mean = 69.6543
Std. Dev. = 32.49059
N = 11
20.00
40.00
60.00
80.00
0
0.00
Mean = 57.2794
Std. Dev. = 45.14096
N = 11
20.00
100.00
40.00
60.00
80.00
100.00
% active time in crop
% active time in crop
Species: BLABI, Group: Max day
Species: BLABI, Group: All days
Crop
consumers
only
6
6
5
5
4
Frequency
Frequency
4
3
3
2
2
1
1
Mean = 76.6197
Std. Dev. = 24.08202
N = 10
0
0
20
40
60
80
0
0.00
Mean = 70.0082
Std. Dev. = 39.30174
N=9
20.00
40.00
60.00
80.00
% active time in crop
100
% active time in crop
Figure 7 Distribution, for Blackbirds in arable crops, of PT (%active time in
crop). Data are presented where individuals have radio contact time >150
minutes.
23
100.00
Single day with most
contact time
Multiple days
Species: LINNE, Group: All days
Species: LINNE, Group: Max day
10
10
All birds
8
Frequency
Frequency
8
6
4
2
6
4
2
Mean = 35.0349
Std. Dev. = 31.47413
N = 23
0
0.00
20.00
40.00
60.00
80.00
0
0.00
100.00
% active time in crop
Mean = 29.8765
Std. Dev. = 33.18998
N = 20
20.00
40.00
60.00
80.00
100.00
% active time in crop
Crop
consumers
only
Species: LINNE, Group: Max day
10
10
8
8
Frequency
Frequency
Species: LINNE, Group: All days
6
6
4
4
2
2
0
0.00
Mean = 42.4106
0
Std. Dev. = 29.67439
0.00
N = 19
20.00
40.00
60.00
80.00
100.00
Mean = 39.8354
Std. Dev. = 32.70999
N = 15
20.00
40.00
60.00
80.00
% active time in crop
% active time in crop
Figure 8. Distribution, for Linnets in arable crops, of PT (%active time in crop).
Data are presented where individuals have radio contact time >150 minutes.
24
100.00
Single day with most
contact time
Multiple days
Species: SKYLA, Group: Max day
Species: SKYLA, Group: All days
30
30
Frequency
Frequency
All birds
20
10
10
0
0.00
20
Mean = 36.6547
Std. Dev. = 37.15709
N = 68
20.00
40.00
60.00
80.00
100.00
0
0.00
Mean = 34.438
Std. Dev. = 40.13643
N = 67
20.00
40.00
% active time in crop
30
25
25
20
20
Frequency
Frequency
30
15
10
5
5
Mean = 51.9275
Std. Dev. = 34.03246
N = 48
20.00
40.00
60.00
80.00
100.00
15
10
0
0.00
80.00
Species: SKYLA, Group: Max day
Species: SKYLA, Group: All days
Crop
consumers
only
60.00
% active time in crop
100.00
0
0.00
Mean = 59.1628
Std. Dev. = 36.00912
N = 39
20.00
40.00
60.00
80.00
% active time in crop
% active time in crop
Figure 9. Distribution, for Skylarks in arable crops, of PT (%active time in
crop). Data are presented where individuals have radio contact time >150
minutes
25
100.00
Single day with most
contact time
Multiple days
Species: YELHA, Group: Max day
Species: YELHA, Group: All days
50
50
40
40
Frequency
Frequency
All birds
30
30
20
20
10
10
0
0.00
Mean = 23.2405
Std. Dev. = 32.8481
N = 72
20.00
40.00
60.00
80.00
0
0.00
Mean = 26.0721
Std. Dev. = 36.1361
N = 65
20.00
40.00
60.00
80.00
100.00
% active time in crop
100.00
% active time in crop
Species: YELHA, Group: Max day
Species: YELHA, Group: All days
50
50
40
40
Frequency
Frequency
Crop
consumers
only
30
30
20
20
10
10
0
0.00
Mean = 41.833
Std. Dev. = 34.15565
N = 40
20.00
40.00
60.00
80.00
100.00
0
0.00
Mean = 49.8437
Std. Dev. = 36.21777
N = 34
20.00
40.00
60.00
% active time in crop
Figure 10. Distribution, for Yellowhammers in arable crops, of PT (%active
time in crop). Data are presented where individuals have radio contact time
>150 minutes
26
80.00
% active time in crop
100.00
Appendix 3: Estimation of the 90th and 95th percentile PTs (details
provided by CSL)
For any radio tracked individual we can ask how much it used a particular
crop at a particular season and hence calculate PT. For our focal species we
can then plot a distribution of PTs. We assume that those individuals that
spend a greater proportion of time in the crop are at greater risk from
pesticide exposure. The most likely exposure for any unknown individual is
given by the mean of the distribution. But for the purpose of risk assessment it
is common practice to choose a value for exposure that is conservative, one
that protects the majority of the population, for example the 90th or 95th
percentile. A simple way of estimating the 95th percentile is to rank the
individuals in increasing order of PT and to choose the value of PT
corresponding to the 90th percentile individual, say. (Where there is no
precise identity between an individual and the percentile of interest, we can
interpolate between values of neighbouring individuals in the sequence.). A
problem with this approach is that when the sample size if small (which it is for
many of our scenarios) the value of the 90th percentile may be very variable
between samples (more variable than the mean would be). A better estimate
of the 90th percentile may be obtained by assuming that our data is a random
sample from a parent distribution with known mathematical properties.
For many real-world measurements, statisticians assume that a sample
comes from a normal distribution with parameters µ and σ estimated by the
mean and standard deviation of the sample. However, the normal distribution
(with infinite upper and lower bounds) does not often provide a good fit for
proportional data (limited between 0 and 1). There are three distributions that
may be easily applied to data bounded between 0 and 1: Uniform (min=0,
max=1), Triangular (min=0, min<most-likely<max, max=1) and the Beta (α, β).
The Uniform distribution is in fact a special case of the Beta where α= β=1.
The Triangular distribution tends to be used when we have no actual data but
expect that values in the middle will be more common than values at the
extremes. We can see from our radio tracking data (Figure 3 to 10) that PT
often peaks at 0 or 1 and that sometimes there are two peaks. Therefore we
have used the Beta distribution as the most appropriate for describing PT. A
further advantage of the Beta is that it is very flexible and can take a wide
variety of shapes, including bimodal distributions (Figure 11).
27
Beta(2, 2)
Beta(10, 10)
1.6
4
1.4
3.5
1.2
3
1
2.5
0.8
2
0.6
1.5
0.4
1
0.2
0.5
0
0
0
0.2
0.4
0.6
0.8
0
1
0.2
Beta(1, 0.5)
0.4
0.6
0.8
1
0.8
1
Beta(0.5, 0.5)
12
8
7
10
6
8
5
6
4
3
4
2
2
1
0
0
0
0.2
0.4
0.6
0.8
1
0
0.2
0.4
0.6
Figure 11. Examples of Beta distributions with different alpha and Beta
parameter values
Goodness of fit. How well does the Beta distribution fit PT?
The Beta distribution has 2 parameters (α and β). We have explored 2
alternative ways of fitting these parameters for our datasets:
• The method of moments
• Maximum likelihood
For many distributions, maximum likelihood is the preferred solution. Figure
12 shows the beta distribution which, according to maximum likelihood
method, best fits the distribution of PTs seen in radio tracking on skylarks
visiting cereal fields in summer. The α and β parameters are both <1 giving a
characteristic U-shaped distribution, reflecting the fact that Skylarks appear to
be divided into 2 groups, one spending very little time in the cereal crop, and
the other spending nearly all their time there. Goodness of fit tests suggest
that it is a reasonable description of the empirical data.
Figure 1313 shows an example where the fit is less good. This dataset is for
72 yellowhammers and includes 45 birds which were successfully radio
tracked but which never visited a cereal crop. The distribution is therefore
strongly weighted toward the 0 PT values. Maximum likelihood has fitted
another U shaped Beta distribution which may be reasonable for the zero
values but is not a very accurate description of the distribution of PT near 1.
Goodness-of-fit tests all indicate significant differences between the empirical
data and the theoretical distribution.
28
Beta(0.30994, 0.24889)
X <= 0.000
0.0%
X <= 0.999
90.0%
0
0
0.2
0.4
0.6
0.8
1
PT
Chi square
4.769
0.3118
Test Value
P Value
Anderson-Darling
2.002
0.05 <= p <= 0.1
Kolmogorov-Smirnov
0.2296
0.0553
Figure 12. Plot of best-fitting Beta distribution (red line) against histogram of
raw PT data (blue bars) for 26 skylarks in cereals from April-August excluding
birds that never visited the crop. Table shows results of goodness of fit tests.
BetaGeneral(0.063308, 0.74157, 0.0000, 1.0000)
X <= 0.000
0.0%
X <= 0.289
90.0%
6
5
4
3
2
1
0
0
0.2
0.4
0.6
0.8
1
PT
Test Value
P Value
Chi square
240.2
0
Anderson-Darling
9.608
< 0.001
Kolmogorov-Smirnov
0.3995
0
Figure 13. Plot of best-fitting Beta distribution (red line) against histogram of
raw PT for cereal crops (blue bars) for 72 yellowhammers, including all birds
caught on arable land. Table shows results of goodness of fit tests.
29
The poor fit of the maximum likelihood solution is more clearly visible when
the fitted distribution and data are plotted as cumulative distributions. Figure
14 shows the same dataset and fitted distribution as Figure 12, but clearly
suggests a systematic lack of fit with a number of points outside the 95%
confidence interval at intermediate values of PT. Investigation revealed that
this appears to be caused by the Maximum Likelihood method giving very
high weight to data points very close to zero or one. In Figure 14, high weight
given to the top two points “pulls” the fitted distribution to the right.
Cereals Skylark - Consumers Summer
100%
90%
Cumulative Probability
80%
70%
60%
50%
40%
30%
20%
Data
95% Conf. Interval
Median
10%
0%
0
0.2
0.4
0.6
0.8
1
PT
Figure 14. Cumulative plot of raw PT data and Beta distribution fitted by
Maximum Likelihood for 26 skylarks in cereals from April-August excluding
birds that never visited the crop. Dotted lines show 95% confidence interval
for the fitted distribution. This is the same dataset as shown in Figure 12, but
the cumulative plot suggests a systematic lack of fit which was not obvious in
Figure 12.
Cumulative graphs were plotted for all the datasets, and several of the
“consumers only” datasets showed a deviation of the type shown in Figure 14.
Some “all birds” datasets showed the reverse pattern, with multiple data
points at zero pulling the fitted distribution to the left (e.g. Figure 15).
These results suggest that the Maximum Likelihood method performs poorly
for at least some of our datasets, especially those with data points at (or very
close to) zero or one. We therefore examined the alternative approach,
method of moments, to fit beta distributions to our data.
30
Orchard Chaff - All
100%
90%
Cumulative Probability
80%
70%
60%
50%
Data
95% Conf. Interval
Median
40%
30%
20%
10%
0%
0
0.2
0.4
0.6
0.8
1
PT
Figure 15. Cumulative plot of raw PT data and Beta distribution fitted by
Maximum Likelihood for 33 chaffinches in orchards (including birds that never
visited the orchard). Dotted lines show 95% confidence interval for the fitted
distribution. The cumulative plot suggests a systematic lack of fit at lower
values of PT.
The method of moments estimates the parameters of the Beta distribution (α
and β) from the mean and variance of the raw data, whereas the Maximum
Likelihood method uses all the individual data points. The method of moments
is therefore less subject than Maximum Likelihood to strong influence from
extreme values at 0 and 1. As a result, it provides a much more reasonable fit
to distributions with data at 0 or 1, and in fact visual inspection suggests a
very reasonable fit to every one of our datasets, regardless of sample size.
For example, Figures 16 and 17 shows distributions fitted by the method of
moments to the same datasets as Figures 14 and 15: in both cases, the
method of moments provides a superior fit. All the points fall between the
confidence limits, even though they are narrower (90% intervals are shown in
Figures 16 and 17, 95% intervals in Figures 14 and 15).
Our results are consistent with simulation studies by Frey et al. (1999) who
also showed that the method of moments performs better than maximum
likelihood for Beta distributions.
In the light of these findings we decided to use method of moments for fitting
Beta distributions to all the datasets considered in this paper. Appendix 4
includes graphs of the distributions fitted by method of moments to all the
datasets.
31
Cereals Skylark - Consumers Summer
100%
90%
Cumulative Probability
80%
70%
60%
50%
40%
30%
Data
90% Conf. Interval
Median
20%
10%
0%
0
0.2
0.4
0.6
0.8
1
PT
Figure 16. Cumulative plot of raw PT data and Beta distribution fitted by
method of moments for 26 skylarks in cereals from April-August excluding
birds that never visited the crop. Dotted lines show 90% confidence interval
for the fitted distribution. This is the same dataset as shown in Figures 12 and
14, but the method of moments provides a better fitting distribution.
Orchard Chaff - All
100%
90%
Cumulative Probability
80%
70%
60%
50%
Data
90% Conf. Interval
Median
40%
30%
20%
10%
0%
0
0.2
0.4
0.6
0.8
1
PT
Figure 17. Cumulative plot of raw PT data and Beta distribution fitted by
method of moments for 33 chaffinches in orchards (including birds that never
visited the orchard). Dotted lines show 90% confidence interval for the fitted
distribution. This is the same dataset as shown in Figure 15, but the method
of moments provides a better fitting distribution.
Quantifying sampling uncertainty: confidence limits for the distribution
of PT
Because the number of birds radio tracked is always limited, estimation of the
fitted distribution is affected by sampling uncertainty, which is represented by
the confidence intervals shown in Figures 14-17. Confidence intervals for the
Maximum Likelihood distributions were obtained by Bayesian simulations
32
using the software package WinBugs. When the Beta distribution was fitted by
method of moments, the confidence intervals were obtained by parametric
bootstrapping. Briefly, 1000 samples of size equal to the raw dataset were
sampled at random from a Beta distribution with the α and β parameters fitted
by the method of moments; each sample was used to fit a new Beta
distribution by the method of moments; and the 1000 distributions were used
to derive a 90% confidence interval for each percentile of PT.
Note that:
• The confidence intervals for the method of moments distributions
represent only sampling uncertainty: other types of uncertainty affecting
PT are discussed below.
• 90% confidence intervals were chosen because this means the upper
bound can be interpreted as a one-sided 95% confidence limit for PT.
However, although 95% confidence bounds are conventionally used in
many types of research, they are not necessarily the appropriate choice
for risk management purposes. Other bounds (e.g. 90%, 99%) could be
provided if risk managers required a lower or higher level of certainty.
Evaluation of other uncertainties affecting the estimation of PT
Sampling uncertainty due to the limited numbers of birds studied is quantified
by the confidence intervals provided with the results. Other uncertainties
affecting the estimation of PT are summarised in Table 1, together with a
qualitative assessment of their direction and magnitude.
Although there are a large number of uncertainties, most of them are thought
to be unbiased (about equally likely to cause under- or over-estimation), and
most of them are expected to operate independently so they will tend to
average out, to some extent. Overall, taking all the identified uncertainties into
account, our view is that the higher percentiles for PT could be either underor over-estimated, and that the total uncertainty might be approximately
double that shown by the 90% confidence intervals given with the results.
If desired, it would be possible to investigate the potential combined effect of
the various uncertainties quantitatively using Monte Carlo simulations:
however, as the uncertainties could only be estimated subjectively, the results
would be only indicative (a form of sensitivity analysis).
Extrapolation to other species and crops
If it were desired to extrapolate to other species or crops, additional
uncertainty would be introduced. Briefly:
• Extrapolation to other species: The radio tracking studies were focussed
on species known to be abundant and frequent in crop, so are expected to
over-estimate most (but not all) other species.
• Extrapolation to other crops: PT could be significantly higher or lower,
depending on attractiveness of crop to species concerned. This would
need to be assessed case-by-case.
33
Table 1. Qualitative evaluation of the main sources of uncertainty affecting the
estimation of high percentiles of the between-bird distribution of PT. Key to
direction and magnitude: +, ++, +++ = uncertainty thought to be causing small,
medium or large over-estimation of PT; -, - -, - - - = uncertainty thought to be
causing small, medium and large under-estimation of PT. “Medium” is used
where the uncertainty is thought to be of approximately similar magnitude to
that shown by the 90% confidence intervals provided for sampling uncertainty.
Source of uncertainty
Potential bias in the sample of birds monitored. If the probability of
capture in a habitat is positively correlated with the time an
individual spends in that habitat, as seems likely, then the tracked
birds will be biased towards those that spend more time in the
crop. More likely for orchards and oilseed rape due to differences
in location of trapping sites.
Uncertainty in the determination of bird locations while tracking.
Uncertainty in determining active/inactive time while tracking.
Uncertainty due to possibility that birds are disturbed by presence
of observer (though care was taken to minimise this). This would
most likely cause under-estimation of PT.
Uncertainty about relation between active time and foraging time,
and between foraging time and the proportion of diet consumed in
that habitat (which determines exposure).
Bias and imprecision due to limited duration of tracking per bird
(this was limited by excluding birds with < 150 minutes).
Extrapolation from study sites to other locations.
Uncertainty if data are used to estimate PT for longer periods. High
percentiles are more likely to be over-estimated than underestimated, as day-to-day differences in PT will tend to average out
over time.
Deviations from fitted distributions. Graphical inspection indicates
reasonable fit for all datasets, so deviations from Beta are thought
to add little to the sampling uncertainty shown by the confidence
intervals.
Under-estimation of variance in small samples. This might cause
under- or over-estimation of high percentiles and confidence
intervals. However, this is thought to be of minor practical
significance here as the upper confidence intervals for small
samples are already close to 1 (see Appendix 4).
Sampling uncertainty for sample α and β may over- or underestimate sampling uncertainty for true α and β, especially with
smaller samples. Effect should average out over many datasets
but will be large for some individual datasets.
Simulation uncertainty: repeat runs of the estimation process
change tabulated percentiles and confidence bounds by amounts
in the order of 0.01-0.02 (could be reduced by more iterations).
Qualitative evaluation of overall impact of unquantified
uncertainties: total uncertainty might be approximately double
that shown by the 90% confidence intervals given with the
results.
34
Direction &
magnitude of
uncertainty
++ orchard and
oilseed rape,
+/- other crops
+/+/+/- -
++/- -
+/++/- ++/- (for longer
term
assessments)
+/-
+/-
+/-
+/-
++/- -
Appendix 4. Results for all datasets (details provided by CSL)
Orchard
(April-September)
All birds
Species
Blackbird
PT
90th percentile
N
Sample
mean
Median
estimate
40
0.23
0.69
Lower
bound
(5%)
0.53
95th percentile
Upper
bound
(95%)
0.86
Median
estimate
0.82
Lower
bound
(5%)
0.67
Upper
bound
(95%)
0.95
Blue tit
20
0.21
0.55
0.40
0.79
0.67
0.51
0.89
Chaffinch
33
0.32
0.74
0.61
0.88
0.85
0.72
0.95
Robin
29
0.21
0.53
0.39
0.70
0.65
0.49
0.83
Orchard consumers Species
only
Blackbird
PT
90th percentile
N
Sample
mean
Median
estimate
28
0.33
0.75
Lower
bound
(5%)
0.60
95th percentile
Upper
bound
(95%)
0.89
Median
estimate
0.84
Lower
bound
(5%)
0.71
Upper
bound
(95%)
0.95
Blue tit
16
0.27
0.58
0.42
0.79
0.68
0.51
0.88
Chaffinch
29
0.36
0.76
0.63
0.88
0.85
0.73
0.95
Robin
24
0.25
0.56
0.44
0.70
0.66
0.53
0.80
Table 1. Orchard. Estimation of 90th and 95th percentile PT values. For each percentile, the median estimate is shown together with the lower and
upper bounds of the 90% confidence interval. Estimates are made separately for the population comprising “All birds” caught in orchards, and for
the sub-population “Orchard consumers only” that excludes birds that never used orchard habitat during radio-tracking.
35
PT
Cereals
N
Sample
mean
Median
estimate
Skylark
68
0.21
Yellowhammer
72
Summer (Apr-Aug)
Skylark
Winter (Sep-Mar)
Median
estimate
0.88
Upper
bound
(95%)
0.99
0.98
Lower
bound
(5%)
0.88
Upper
bound
(95%)
1.00
0.09
0.36
0.18
0.71
0.67
0.45
0.94
44
0.25
0.92
0.73
1.00
0.99
0.90
1.00
Yellowhammer
28
0.21
0.77
0.50
0.98
0.92
0.71
1.00
Skylark
24
0.14
0.67
0.26
1.00
0.91
0.55
1.00
Yellowhammer
44
0.02
0.05
0.02
0.09
0.09
0.05
0.18
PT
Summer (Apr-Aug)
Winter (Sep-Mar)
90th percentile
N
Sample
mean
Median
estimate
Skylark
36
0.39
Yellowhammer
27
Skylark
Yellowhammer
Crop consumers
only
All months
95th percentile
Lower
bound
(5%)
0.67
All birds
All months
90th percentile
95th percentile
Upper
bound
(95%)
1.00
Median
estimate
0.97
Lower
bound
(5%)
0.87
0.99
Lower
bound
(5%)
0.95
Upper
bound
(95%)
1.00
0.24
0.75
0.54
0.96
0.88
0.70
0.99
26
0.42
0.97
0.86
1.00
0.99
0.94
1.00
17
0.34
0.87
0.67
0.99
0.95
0.79
1.00
Skylark
10
0.34
0.94
0.59
1.00
0.98
0.72
1.00
Yellowhammer
10
0.02
0.14
0.09
0.23
0.18
0.11
0.30
Table 2. Cereals . Estimation of 90th and 95th percentile PT values. For each percentile, the median estimate is shown together with the lower and
upper bounds of the 90% confidence interval. Estimates are made separately for the population comprising “All birds” caught in the arable
landscape, and for the sub-population “Crop consumers only” that excludes birds that never used cereal fields during radio-tracking.
36
Beet-Potatoes
PT
All birds
All Months
Summer (Apr-Nov)
N
Sample
mean
90th
percentile
Skylark
68
0.09
Yellowhammer
72
Linnet
Summer (Apr-Nov)
95th percentile
Upper
bound
(95%)
0.82
95th
percentile
0.39
Lower
bound
(5%)
0.17
0.08
0.32
0.09
22
0.13
0.42
Skylark
59
0.11
Yellowhammer
50
Linnet
21
0.74
Lower
bound
(5%)
0.47
Upper
bound
(95%)
0.98
0.86
0.75
0.42
0.99
0.24
0.77
0.59
0.36
0.92
0.47
0.23
0.87
0.79
0.53
0.99
0.12
0.56
0.27
0.96
0.87
0.59
1.00
0.13
0.43
0.26
0.84
0.60
0.38
0.96
PT
Crop consumers
only
All Months
90th percentile
90th percentile
N
Sample
mean
90th
percentile
Skylark
18
0.35
Yellowhammer
14
Linnet
11
95th percentile
Upper
bound
(95%)
0.99
95th
percentile
0.88
Lower
bound
(5%)
0.68
0.95
Lower
bound
(5%)
0.80
Upper
bound
(95%)
1.00
0.43
0.94
0.75
1.00
0.98
0.85
1.00
0.25
0.59
0.38
0.84
0.69
0.47
0.92
Skylark
18
0.35
0.88
0.68
0.99
0.95
0.80
1.00
Yellowhammer
13
0.46
0.94
0.76
1.00
0.98
0.85
1.00
Linnet
11
0.25
0.59
0.38
0.84
0.69
0.47
0.92
Table 3. Beet & Potatoes . Estimation of 90th and 95th percentile PT values. For each percentile, the median estimate is shown together with the
lower and upper bounds of the 90% confidence interval. Estimates are made separately for the population comprising “All birds” caught in the
arable landscape, and for the sub-population “Crop consumers only” that excludes birds that never used beet or potato fields during radiotracking.
37
OSR
N
PT
Sample
mean
Skylark
Yellowhammer
Linnet
Blackbird
68
72
22
10
0.05
0.05
0.12
0.56
Summer (Apr-Jul) Skylark
Yellowhammer
Linnet
Blackbird
41
28
22
10
0.05
0.11
0.12
0.56
0.18
0.60
0.62
0.98
0.08
0.28
0.07
0.84
0.62
1.00
1.00
1.00
0.38
0.82
0.95
1.00
0.20
0.50
0.43
0.90
0.89
1.00
1.00
1.00
Winter (Aug-Mar)
Skylark
Yellowhammer
27
44
0.05
0.01
0.10
0.00
0.00
0.00
1.00
0.88
0.47
0.01
0.06
0.00
1.00
1.00
Skylark
Yellowhammer
Linnet
Blackbird
11
9
6
9
0.31
0.39
0.44
0.44
0.72
0.82
0.99
0.98
0.52
0.58
0.61
0.84
0.93
0.99
1.00
1.00
0.81
0.90
1.00
0.99
0.61
0.67
0.73
0.90
0.97
1.00
1.00
1.00
Summer (Apr-Jul) Skylark
Yellowhammer
Linnet
Blackbird
7
7
6
9
0.33
0.45
0.44
0.44
0.57
0.86
0.99
0.98
0.42
0.65
0.61
0.84
0.80
1.00
1.00
1.00
0.64
0.92
1.00
0.99
0.47
0.73
0.73
0.90
0.87
1.00
1.00
1.00
Winter (Aug-Mar)
4
2
0.36
0.01
0.98
0.61
0.38
0.15
1.00
1.00
1.00
0.79
0.51
0.22
1.00
1.00
All birds
All Months
90th
percentile
Median
Lower
estimate
bound
(5%)
0.17
0.06
0.11
0.02
0.62
0.07
0.98
0.84
Upper
bound
(95%)
0.52
0.55
1.00
1.00
95th percentile
Median
Lower
estimate
bound
(5%)
0.42
0.23
0.40
0.18
0.95
0.43
1.00
0.90
Upper
bound
(95%)
0.90
0.95
1.00
1.00
Crop consumers
only
All Months
Skylark
Yellowhammer
Table 4. OSR. Estimation of 90th and 95th percentile PT values. For each percentile, the median estimate is shown together with the lower and
upper bounds of the 90% confidence interval. Estimates are made separately for the population comprising “All birds” caught in the arable
landscape, and for the sub-population “Crop consumers only” that excludes birds that never used beet or oilseed rape fields during radiotracking.
38
Cereals Skylark - Consumers All
100%
90%
90%
80%
80%
Cumulative Probability
Cumulative Probability
Cereals Skylark - All
100%
70%
Data
90% Conf. Interval
Median
60%
50%
40%
30%
70%
60%
50%
40%
30%
20%
20%
10%
10%
0%
100%0
Cereals Skylark - Consumers All
0.2
0.4
0.6
0.8
0%
100%0
1
Data
90% Conf. Interval
Median
Cereals Skylark - All Summer
0.2
0.4
90%
90%
80%
80%
70%
60%
50%
40%
30%
Data
90% Conf. Interval
Median
20%
60%
50%
30%
10%
Cereals Skylark - All Winter
0.2
0.4
0.6
0.8
0%
100% 0
1
90%
80%
80%
70%
Data
90% Conf. Interval
Median
60%
Cereals Skylark - Consumers Winter
0.2
0.4
0.6
0.8
1
PT
Cumulative Probability
Cumulative Probability
Data
90% Conf. Interval
Median
40%
90%
50%
40%
30%
70%
60%
50%
40%
30%
20%
20%
10%
10%
0
1
70%
PT
0%
0.8
20%
10%
0%
100% 0
0.6
PT
Cumulative Probability
Cumulative Probability
PT
0%
Data
90% Conf. Interval
Median
0
0.2
0.4
0.4
0.6
0.8
Figure0.2
18. Cereals:
skylark.
Data are
plotted as1 cumulative distribution
functions,
soPT
that 0.6
for any
PT
percentile of interest along the Y axis, median value of PT and the 90% confidence
bounds can be read off directly from the X-axis.
Figure 18. Cereals: Skylark
39
0.8
1
Cereals Yelha - All
100%
90%
90%
80%
Data
90% Conf. Interval
Median
70%
Cumulative Probability
Cumulative Probability
80%
60%
50%
40%
30%
70%
60%
50%
30%
20%
10%
10%
0%
0
Yelha - All Summer
0.2 Cereals0.4
0.6
0.8
0%
0
100%
1
PT
90%
80%
80%
Cumulative Probability
90%
70%
Data
90% Conf. Interval
Median
60%
50%
40%
30%
50%
40%
30%
10%
10%
0.2
0.4 Yelha - 0.6
Cereals
All Winter
PT
0.8
0
100%
90%
1
PT
80%
Data
90% Conf. Interval
Median
70%
Cumulative Probability
Cumulative Probability
0.2 Cereals Yelha
0.4 - Consumers
0.6
Winter0.8
90%
80%
60%
50%
40%
30%
60%
50%
40%
30%
20%
10%
10%
0
F
0.2
0.4
0.6
0.8
1
PT
Data
90% Conf. Interval
Median
70%
20%
0%
Data
90% Conf. Interval
Median
0%
1
1
PT
60%
20%
0
100%
Summer
0.2Cereals Yelha
0.4 - Consumers
0.6
0.8
70%
20%
0%
Data
90% Conf. Interval
Median
40%
20%
100%
Cumulative Probability
Cereals Yelha - Consumers All
100%
0%
0
0.2
0.4
0.6
PT
Figure 19. Cereals: Yellowhammer
40
0.8
1
Beet Pots Skylarks - All
100%
90%
90%
80%
80%
Data
90% Conf. Interval
Median
70%
Cumulative Probability
Cumulative Probability
Beet Pots Skylarks - Consumers All
100%
60%
50%
40%
30%
70%
60%
50%
40%
30%
20%
20%
10%
10%
0%
0
0.2
0.4
0.6
0.8
0%
1
Data
90% Conf. Interval
Median
0
0.2
PT
90%
90%
80%
80%
70%
Data
90% Conf. Interval
Median
50%
40%
30%
60%
50%
40%
30%
20%
10%
10%
0.2
0.4
0.6
1
70%
20%
0
0.8
Beet Pots Skylarks - Consumers Summer
100%
Cumulative Probability
Cumulative Probability
Beet Pots Skylarks - All Summer
0%
0.6
PT
100%
60%
0.4
0.8
1
PT
0%
Data
90% Conf. Interval
Median
0
0.2
0.4
0.6
PT
Figure 20. Beet & Potatoes:Skylarks
41
0.8
Beet Pots Yelha - All
100%
90%
90%
80%
80%
70%
Cumulative Probability
Data
90% Conf. Interval
Median
60%
50%
40%
30%
70%
60%
50%
40%
30%
20%
20%
10%
10%
0%
0
0.2
Cumulative Probability
100%
0.4
0.6
Beet Pots Yelha - All Summer
PT
0.8
0%
1
Data
90% Conf. Interval
Median
0
90%
90%
80%
80%
70%
Data
90% Conf. Interval
Median
60%
50%
40%
30%
0.6
0.8
1
PT
30%
0%
Data
90% Conf. Interval
Median
0
0.2
0.4
0.6
PT
Figure 21. Beet & Potatoes: Yellowhammer
42
1
40%
10%
0.4
0.8
50%
10%
0.2
0.6
60%
20%
0
0.4
Beet Pots Yelha - Consumers Summer
PT
70%
20%
0%
0.2
100%
Cumulative Probability
Cumulative Probability
Beet Pots Yelha - Consumers All
100%
0.8
1
Beet Pots Linnets - Consumers All
100%
90%
90%
80%
80%
Cumulative Probability
Cumulative Probability
Beet Pots Linnets - All
100%
70%
Data
90% Conf. Interval
Median
60%
50%
40%
30%
60%
50%
30%
20%
10%
10%
0
0.2
0.4
0.6
0.8
0%
1
PT
Beet Pots Linnets - All Summer
90%
80%
70%
Data
90% Conf. Interval
Median
60%
50%
40%
30%
20%
10%
0
0.2
0
0.2
0.4
0.6
PT
100%
0%
Data
90% Conf. Interval
Median
40%
20%
0%
Cumulative Probability
70%
0.4
0.6
0.8
1
PT
Figure 22. Beet & Potatoes: Linnet
43
0.8
1
OSR Skylarks - All
100%
90%
90%
80%
80%
Data
90% Conf. Interval
Median
70%
Cumulative Probability
Cumulative Probability
OSR Skylarks - Consumers All
100%
60%
50%
40%
30%
70%
60%
50%
30%
20%
20%
10%
10%
0%
0
100%
0.2
OSR Skylarks - All Summer
0.4
0.6
0.8
0%
100% 0
1
PT
90%
Cumulative Probability
70%
60%
50%
40%
30%
0.6
0.8
1
PT
70%
60%
50%
30%
20%
10%
10%
0
0.2
100%
0.4
0.6
OSR Skylarks - All Winter
PT
0.8
0%
1
Data
90% Conf. Interval
Median
40%
20%
Cumulative Probability
0.4
80%
Data
90% Conf. Interval
Median
0
90%
90%
80%
80%
70%
60%
50%
40%
30%
Data
90% Conf. Interval
Median
20%
0.4
0.6
0.8
OSR Skylarks - Consumers Winter
PT
1
70%
60%
50%
40%
Data
90% Conf. Interval
Median
30%
20%
10%
0%
0.2
100%
Cumulative Probability
Cumulative Probability
OSR Skylarks - Consumers Summer
0.2
90%
80%
0%
Data
90% Conf. Interval
Median
40%
10%
0
0.2
0.4
Figure 23.
OSR: skylark
PT
0.6
0.8
1
0%
0
0.2
0.4
0.6
PT
Figure 23: OSR: Skylark
44
0.8
1
OSR Yelha - All
100%
90%
90%
80%
80%
Data
90% Conf. Interval
Median
70%
Cumulative Probability
Cumulative Probability
OSR Yelha - Consumers All
100%
60%
50%
40%
30%
70%
60%
50%
40%
30%
20%
20%
10%
10%
0%
0
0.2
0.4
0.6
0.8
0%
1
Data
90% Conf. Interval
Median
0
0.2
100%
100%
90%
90%
80%
80%
70%
Data
90% Conf. Interval
Median
60%
50%
40%
30%
40%
30%
10%
OSR Yelha - All Winter
0.6
0.8
Data
90% Conf. Interval
Median
0%
100%0
1
OSR Yelha - Consumers Winter
0.2
0.4
PT
0.8
1
90%
80%
80%
Data
90% Conf. Interval
Median
70%
Cumulative Probability
Cumulative Probability
0.6
PT
90%
60%
50%
40%
30%
70%
60%
50%
40%
30%
20%
20%
10%
10%
0%
1
50%
10%
0.4
0.8
60%
20%
0.2
0.6
70%
20%
0%
100%0
0.4
PT
OSR Yelha - Consumers Summer
Cumulative Probability
Cumulative Probability
PT
OSR Yelha - All Summer
0
Figure
Yellowhammer
0.224. OSR:0.4
0.6
0.8
1
PT
0%
Data
90% Conf. Interval
Median
0
0.2
0.4
0.6
PT
Figure 24 OSR: Yellowhammer
45
0.8
1
OSR Linnets - Consumers
100%
90%
90%
80%
80%
Cumulative Probability
Cumulative Probability
OSR Linnets - All
100%
70%
60%
50%
Data
90% Conf. Interval
Median
40%
30%
70%
60%
50%
40%
30%
20%
20%
10%
10%
0%
0
0.2
0.4
0.6
0.8
0%
1
Data
90% Conf. Interval
Median
0
0.2
100%
100%
90%
90%
80%
80%
70%
60%
50%
40%
30%
20%
0.6
0.8
1
Data
90% Conf. Interval
Median
70%
60%
50%
40%
30%
20%
Data
90% Conf. Interval
Median
10%
0%
0.4
OSR BlaBi - PT
Consumers
Cumulative Probability
Cumulative Probability
PT - All
OSR BlaBi
0
0.2
0.4
0.6
0.8
10%
1
PT
0%
0
0.2
0.4
0.6
PT
Figure 25 OSR Linnet & Blackbird
46
0.8
1
Orchard Blabi - Consumers
100%
90%
90%
80%
80%
Cumulative Probability
Cumulative Probability
Orchard Blabi - All
100%
70%
Data
90% Conf. Interval
Median
60%
50%
40%
30%
70%
60%
50%
30%
20%
20%
10%
10%
0%
0
0.2
0.4
0.6
0.8
0%
1
Data
90% Conf. Interval
Median
40%
0
0.2
100%
100%
90%
90%
80%
80%
70%
Data
90% Conf. Interval
Median
60%
50%
40%
30%
40%
30%
10%
0.4
0.6
1
50%
10%
0.2
0.8
0.8
1
PT
Data
90% Conf. Interval
Median
60%
20%
0
0.6
70%
20%
0%
0.4
PT
Orchard BluTi - Consumers
Cumulative Probability
Cumulative Probability
PT
Orchard BluTi - All
0%
0
0.2
0.4
0.6
PT
Figure 26. Orchard: Blackbird & Blue tit
47
0.8
1
Orchard Chaff - Consumers
100%
90%
90%
80%
80%
Cumulative Probability
Cumulative Probability
Orchard Chaff - All
100%
70%
60%
50%
Data
90% Conf. Interval
Median
40%
30%
70%
60%
50%
30%
20%
20%
10%
10%
0%
0
0.2
0.4
0.6
0.8
0%
1
Data
90% Conf. Interval
Median
40%
0
0.2
100%
100%
90%
90%
80%
80%
70%
Data
90% Conf. Interval
Median
60%
50%
40%
30%
40%
30%
10%
0.4
0.6
1
50%
10%
0.2
0.8
0.8
1
PT
Data
90% Conf. Interval
Median
60%
20%
0
0.6
70%
20%
0%
0.4
PT
Orchard Robin - Consumers
Cumulative Probability
Cumulative Probability
PT
Orchard Robin - All
0%
0
0.2
0.4
0.6
PT
Figure 27, Orchard: Chaffinch & Robin
48
0.8
1
Fly UP