See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/225562034

Smoothing and trend detection in waterbird monitoring data using structural

time-series analysis and the Kalman filter

Article in Journal of Ornithology · December 2007

DOI: 10.1007/s10336-007-0176-7

CITATIONS READS

62 167

4 authors, including:

Hans Visser Arco J. Van Strien

PBL Netherlands Environmental Assessment Agency Centraal Bureau voor de Statistiek
88 PUBLICATIONS 1,688 CITATIONS 152 PUBLICATIONS 8,781 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Water, climate and conflict: rethinking their relationship in the context of future changes View project

Trend estimation in global temperatures View project

All content following this page was uploaded by Hans Visser on 01 May 2014.

The user has requested enhancement of the downloaded file.

J Ornithol
DOI 10.1007/s10336-007-0176-7

ORIGINAL ARTICLE

Smoothing and trend detection in waterbird monitoring data

using structural time-series analysis and the Kalman filter
Leo Soldaat Æ Hans Visser Æ Marc van Roomen Æ
Arco van Strien

Received: 22 December 2006 / Revised: 13 June 2007 / Accepted: 13 June 2007

Dt. Ornithologen-Gesellschaft e.V. 2007

Abstract Many wildlife-monitoring programmes have monitoring scheme for migrating and overwintering
long time series of species abundance that cannot be waterbirds. Taking the numbers of overwintering Greater
summarized adequately by linear trend lines. To describe Scaup (Aythia marila) in the Netherlands as an example,
long time series better, generalized additive models may be we demonstrate three applications of the method: (1) trend
used to obtain a smooth trend line through abundance data. calculation and classification for each year in the time
We describe another approach to estimate a smoothed trend series, (2) assessing alerts for alarming population declines
line through time series consisting of one observation per and (3) testing yearly abundance against a population
time point, such as year or month. This method is based on threshold. We discuss the situations where TrendSpotter is
structural time-series models in combination with the to be preferred over other methods.
Kalman filter and is computerized in the TrendSpotter
software. One of its strengths is the possibility to test Keywords Kalman filter
Structural time-series analysis

changes in smoothed abundances between years, taking Population alert
Trend analysis
Generalized additive
into account serial correlation. The trend method is applied model (GAM)
TRends and Indices for Monitoring data
in the Dutch Waterbird Monitoring Scheme (DWMS), a (TRIM)
Aythia marila
Greater scaup

Communicated by F. Bairlein. Introduction

This is a paper on the occasion of a presentation at the ‘‘population
alerts’’ symposium at the 24th IOC Congress in Hamburg. Long time series of wildlife data are becoming more
TrendSpotter is available from the second author widespread now that many countries have well-established
(http://www.hans.visser@mnp.nl) at no cost. monitoring programmes. In the Netherlands nature moni-
toring has been strongly intensified and standardized in the
L. Soldaat (&)
A. van Strien
past decade of the twentieth century. This development was
Statistics Netherlands, PO Box 4000, Voorburg 2270 JM,
The Netherlands mainly driven by the wish to base nature policy on sound
e-mail: lslt@cbs.nl knowledge of trends in animal and plant numbers and by
A. van Strien international regulations like the European Unions’ Habi-
e-mail: asin@cbs.nl tats and Birds Directive. The Dutch Network for Ecological
Monitoring (NEM; van Strien 2006) now contains 14
H. Visser
monitoring schemes for different species groups, ranging
Netherlands Environmental Assessment Agency (MNP),
Bilthoven, The Netherlands form butterflies and plants to birds. Most time series cover
e-mail: hans.visser@mnp.nl periods of 10–15 years, but time series of bird species span
more than 20 years. The longest time series, since 1975,
M. van Roomen
come from the Dutch Waterbird Monitoring Scheme.
SOVON Dutch Centre for Field Ornithology,
Beek-Ubbergen, The Netherlands In most NEM monitoring schemes data analysis is per-
e-mail: Marc.vanRoomen@sovon.nl formed using log-linear Poisson regression, a form of

123
 J Ornithol

generalized linear models (GLMs) (McCullagh and Nelder country. For the Greater Scaup, for instance, yearly, more
1989; ter Braak et al. 1994). The programme used is than 90% of all the birds present in the country have
TRends and Indices for Monitoring data (TRIM) (Panne- probably been counted in the DWMS since 1990.
koek and van Strien 2001), which calculates yearly indices To calculate yearly total numbers of Greater Scaup and
per species and also provides an overall linear trend esti- any other species in the DWMS, missing values in the
mate that is based on the yearly indices and is meant to dataset were first imputed in two steps:
describe the change over the entire study period. However,
(1) First, missing monthly counts were imputed with
long time series may show alternating periods of increase
UINDEX (Bell 1995; Underhill and Prŷs-Jones 1994)
and decrease, and linear trends do not summarize these
at the level of so-called monitoring sites, which in-
time series adequately. Furthermore, to separate patterns of
clude all important wetlands. We did not use TRIM
genuine change from annual fluctuations, it is helpful to
here, because TRIM cannot cope with monthly
apply a smoothing algorithm. Although TRIM is able to
counts. Before imputation, the sites were grouped into
assess trends in parts of the time series by using change
11 regional strata in which a more or less comparable
points, smoothing would require other techniques such as
population development was expected.
Loess estimators or generalized additive models (GAMs).
(2) After summation of the imputed monthly counts to
GAMs are among the most widely used methods to
seasonal sums per monitoring site, the seasonal sums
smooth time series. The smoothed trend line of a GAM will
with > 90% imputed birds were assigned a missing
usually look much like a moving average, but, in addition,
value, in order to exclude extremely high imputed
it provides information on the uncertainty of the trend by
values that are occasionally generated in the first step.
calculating confidence limits. GAMs are being used for the
These missing values were imputed again within a
analysis of bird monitoring data, for instance in the
database that was not stratified in regional strata.
Breeding Bird Scheme and the Wetland Bird Survey in the
UK (Fewster et al. 2000; Atkinson et al. 2006). However, After imputation, the seasonal sums of all sites were
in some situations, GAMs are not satisfactory to apply. In aggregated to a country yearly total. In the standard trend
such cases, an alternative smoothing method from the class analysis procedure of the DWMS, yearly totals consisting
of structural time-series analysis, in combination with the of > 90% imputed birds are assigned a missing value, but,
Kalman filter, may be helpful (Visser 2004, 2005, 2007). In for the Greater Scaup, all yearly totals had < 90% impu-
this paper we demonstrate the main characteristics of this tation. The yearly totals were divided by 12 to give mean
method and some applications, using the TrendSpotter monthly numbers, which give better understandable bird
software package. We first give a brief introduction to numbers than seasonal sums or so-called bird-days. The
TrendSpotter and compare characteristics of three methods last step was to apply TrendSpotter to these mean values.
of analysis (TRIM, GAM and TrendSpotter). We then fo-
cus on the application of TrendSpotter to data of over- Characteristics of the methods of analysis
wintering Greater Scaup (Aythya marila) in The
Netherlands. TrendSpotter estimates smoothed population numbers l1,
l2, . . . , lt, . . . , lN for a time series with N equidistant
measurements over time (y1, y2, . . . , yt, . . . , yN). The
Materials and methods expectations of these smoothed population numbers are
denoted as m1, m2, . . . , mt, . . . , mN. TrendSpotter also
Data and data processing estimates the standard deviations of the smoothed popula-
tion numbers: SD1, SD2, . . . , SDt, . . . , SDN. Finally, it
In the Dutch Waterbird Monitoring Scheme (DWMS) estimates the standard deviations of the differences lN–lt,
waterbirds are counted on all important water bodies in the denoted here as SDN–t. The estimation of confidence inter-
Netherlands. These numbers concern mainly migrating and vals is based on the deviations of time point values from the
overwintering birds. The waterbird counts are performed in smoothed line. This emphasizes temporal variation as the
8–12 months per season, depending on the region (Soldaat main source of error, although measurement errors and er-
et al. 2004; van Roomen et al. 2006a, b). rors due to imputation are implicitly incorporated in the
Although there are many missing values present in the yearly measurements. Missing data are easily incorporated
data because not all water bodies are counted each month, into the model. In the present version of TrendSpotter it is
the scheme comes close to a total census for many species. not possible to add information on differences in reliability
For these species a large part of the total population is of individual measurements. In TrendSpotter the flexibility
counted, and the sum of all birds at individual sites is a of a trend is set by the value of a standard deviation of a noise
good approximation of the number of birds in the whole process. This standard deviation (parameter) can be chosen

123
J Ornithol

by hand, or can be optimized by maximum likelihood or other areas with exceptionally high numbers of partic-
optimization (the default choice). If the parameter is set to ular species are selected several times per bootstrap sam-
zero, the estimated trend equals the well-known regression ple. Underhill and Prŷs-Jones (1994) have already
trend line (straight line). If the parameter is set to a very large acknowledged that there are logical difficulties with
number, the trend will go through all measurements y1, y2, incorporating between-site variation in the case of a census.
. . . , yt, . . . , yN. All values for mI, SDI, and SDN–t are In such a situation, we prefer to apply TrendSpotter in
estimated by use of the Kalman filter. This filter is attractive, combination with an imputing method.
because it gives estimates with optimal statistical properties. All three trend methods discussed can be used to iden-
Although normally distributed residuals (in jargon innova- tify change points (Table 1), but the methods differ in their
tions) are not a necessary condition for the proper use of the ability to test differences between years. In TRIM, each
Kalman filter, it still is a desirable property. For example, all yearly index can be statistically tested against the base
confidence limits assessed by ±1.96 · standard error can be year, taking into account serial correlation. TrendSpotter is
interpreted as 95% confidence limits. Normality of residuals also able to test indices between years. More particularly,
can be tested by a so-called normality plot. For mathematical TrendSpotter tests the difference between the smoothed
details about structural time-series analysis and the Kalman population number of the last year against each of the
filter please refer to Harvey (1989). preceding years, thereby taking into account serial corre-
Table 1 lists characteristics of three methods of analysis. lation between years. GAM indices may also be compared
Both TRIM and GAMs can be used to estimate missing between years, but a statistical test is limited because serial
counts in the raw data. TrendSpotter, however, is not able correlation cannot be taken into account. Substantial po-
to impute missing values in the raw data. Thus, for the sitive serial correlation may be present in bird monitoring
DWMS, TrendSpotter can only be used to smooth indices data, because partly the same individual birds are counted
obtained by other methods, e.g. TRIM or UINDEX (Un- each year at the same sites. If such serial correlation is not
derhill and Prŷs-Jones 1994; Bell 1995). In contrast, one of taken into account, confidence intervals may be underes-
the strengths of a GAM is its ability to smooth, using the timated, leading to more type I errors.
raw data (Fewster et al. 2000), which makes it a promising
method, although it is very computer-intensive. Both
TRIM and TrendSpotter apply analytical methods to cal- Results
culate confidence intervals. GAMs typically compute
confidence intervals using bootstrapping, because of diffi- Trend calculation and classification
culties in applying an analytical method (Fewster et al.
2000). However, bootstrapping of sites is not adequate in The mean monthly numbers of overwintering Greater
the case of a census, such as the DWMS. Bootstrapping of Scaup in the Netherlands nowadays are slightly lower than
sites emphasizes the between-site variation, i.e. the varia- in the 1970s and early 1980s, and they are much lower than
tion in changes between sites, as the main source of in the second part of the 1980s and the early 1990s (Fig. 1).
uncertainty, whereas this is not relevant for a census.
Bootstrapping might even lead to unrealistic high upper 70000 Greater Scaup in the Netherlands

limits of confidence intervals when, e.g., the Wadden Sea 60000

mean monthly number

50000
Table 1 Characteristics of the three methods of analysis 40000
Characteristic TRIM GAM TrendSpotter 30000

Imputation Yes Yes No 20000

method?
10000
Smoothing No Yes, using the Yes, using indices
method? raw data of other models 0
Computer No Yes No 1975 1985 1995 2005
intensive?
Fig. 1 Time series analysis of the mean number of Greater Scaup in
Confidence Analytical Bootstrapping Analytical the Netherlands from 1975 to 2004. The black dots indicate the
intervals method seasonal sums (after imputation, see text). The black line is the
Detection of Yes Yes Yes smoothed population number estimated by TrendSpotter. The dashed
change points? black lines indicate the upper and lower limits of the 95% confidence
Test of indices Yes Limited Yes interval. The solid grey line is the favourable conservation status of
between years? the Greater Scaup (set to a mean monthly number of at least 25,000
birds in The Netherlands)

123
 J Ornithol

The figure makes clear that one overall linear trend esti- Table 2 TrendSpotter results of the analysis of mean monthly
mate for this time series is not very informative, because of numbers of Greater Scaup in The Netherlands. The analysis was
performed on log-transformed data. yt mean population number (un-
the relatively high bird numbers in the middle part of the
transformed), mt smoothed population number, SDt standard deviation
time series. Instead, the smoothed line calculated by of mt, mN–mt difference between mN and mt, SDN–t standard deviation
TrendSpotter gives a much better description of the pop- of the difference between mN and mt
ulation trajectory. TrendSpotter enables us to calculate T yt mt SDt mN–mt SDN–t
trend estimates based on the smoothed curve. We distin-
guish between (1) the total change rates (TCRs), i.e. the 1975 12,757 9.852 0.204 –0.047 0.290
smoothed population number in the last year compared 1976 39,288 10.074 0.147 –0.268 0.254
with the smoothed population number in each year in the 1977 27,174 10.215 0.134 –0.409 0.247
time series and (2) the mean yearly change rate (YCR), i.e. 1978 23,895 10.296 0.134 –0.491 0.247
the TCR expressed as a mean change rate per year. The . . . . . .
YCR corresponds with the multiplicative yearly slope in . . . . . .
TRIM. . . . . . .
If TrendSpotter is applied to untransformed data, the 2004 (N) 18,775 (yN) 9.805 (mN) 0.207 (SDN) 0 0
calculations proceed by first approximating the total
change rate as:
  2004, and a mean yearly decline between 1978 and 2004 of
l
TCR ¼ Expectation N 1.9% [100 · (1.00–0.981) = 1.9].
lt
  ð1aÞ The next step is to take the confidence limits into ac-
lN  lt þ mt mN count in the trend calculation. By using the approximation
 Expectation ¼
mt mt for lN/lt given in Eq. 1a, we find the 95% confidence
limits of the TCR and YCR
We note that the expectation of lN/lt could also be cal-
culated by Monte Carlo simulation. However, this option is 1:96 SDNt
ClYCR  TCR  ð3aÞ
not incorporated into TrendSpotter. mt
For log-transformed data the TCR can be approximated
if TrendSpotter is run on untransformed data, or
by:

em N ClYCR  eðmN mt Þ  1:96 SDNt ð3bÞ

TCR  ¼ emN mt ð1bÞ
em t
in the case of log-transformed data. For both untransformed
For both untransformed and log-transformed data, the YCR and log-transformed data, the confidence interval of the
is calculated as: YCR can be calculated by substituting CITCR in Eq. 4 by
the lower and upper confidence limits from Eq. 3a or b
LnðTCRÞ
YCR ¼ e Nt ð2Þ
LnðClTCR Þ

The smoothed curve values for Greater Scaup numbers ClYCR  e Nt ð4Þ
indicate a decline from 1978 to 2004 (Fig. 1). The
corresponding log-transformed values for mN and mt are YCR estimates and confidence intervals can be used to
9.805 and 10.296, respectively (Table 2). The analysis was classify the trends per year. We applied the trend classifi-
performed on log-transformed data, so substituting these cation scheme given in Table 3. The results for Greater
figures in Eq. 1b gives a total change rate of: Scaup in The Netherlands are shown in Table 4 (fourth
column), indicating that the species has undergone a sig-
nificant moderate decline since 1979–1980 and 1988–1995.
e9:805
TCR ¼ ¼ 0:612
e10:296 Population alerts
and substituting this TCR in Eq. 2 gives a yearly change
rate of An extension of trend calculation is the assessment of
population alerts. A population alert highlights any de-
Lnð0:612Þ
YCR ¼ e20041978 ¼ 0:981 clines in bird numbers that are of conservation concern (de
Nobel et al. 2002). The British Trust for Ornithology
In other words, the population underwent an almost 40% (BTO) alert system (http://www.bto.org/birdtrends2005/
decline [100 · (1.00–0.612) = 38.8], between 1978 and alerts.htm) uses strong declines (>50%) and moderate

123
J Ornithol

Table 3 Classification of trends based on the 95% confidence interval of the YCR (CL confidence limit of YCR, CI confidence interval of YCR)
Trend class Criteria Description

Strong increase Lower CL >1.05 >5% increase/year ( doubling in 15 years)

Moderate increase 1.00 < lower CL £ 1.05 Increase, but unsure whether > 5% year–1
Stable Confidence interval contains 1.00 and lower Population changes less than 5% year–1
CL ‡ 0.95 and upper CL £ 1.05
Moderate decline 0.95 £ upper CL < 1.00 Decrease, but unsure whether > 5% year–1
Steep decline Upper CL < 0.95 > 5% decrease/year ( halving in 15 years)
Uncertain Confidence interval contains 1.00 AND CI too large for reliable trend classification
(lower CL < 0.95 OR upper CL > 1.05)

declines (>25%) as alert thresholds, over different time partly due to a slight increase in the numbers of Greater
periods (the whole time series and the last 25, 10 and Scaup counted, but also due to the larger confidence
5 years of the time series). The BTO system uses GAMs to intervals at the end of the time series. The widening of
calculate smoothed indices of animal abundance and takes confidence intervals at the start and end of time series is
into account the standard errors of the indices to test for inherent to smoothing techniques.
significance of the alerts.
TrendSpotter can also be used to calculate alerts through
the conversion of the TCR values. The population of Discussion
Greater Scaup has a TCR of 0.375 since 1990 (see Table 4,
fifth column), which is similar to a 62.5% decline Time series analysis with TrendSpotter has proved useful
[100 · (1.00–0.375) = 62.5]. This decline is significantly in the Dutch Waterbird Monitoring Scheme. Time series of
larger than 25%, as the maximal TCR gives an estimated up to 30 years, with alternating periods of increases and
39.3% decline for 1990–2004 [100 · (1.00– declines, were adequately described by a smoothed line.
0.607) = 39.3], but not significantly larger than 50%. If an Changes in population abundance since any year could be
alert threshold of 25% decline is used, a moderate alert for statistically tested and trends could be classified for each
the Greater Scaup is generated for the decline since 1990 individual year and converted into population alerts, as
(Table 4). Alert-thresholds may be any arbitrary value, e.g. well as tested against population thresholds. GAMs offer
>10%, >25% or >50% decline, and may also be based on a similar advantages, but, as said earlier, do not take serial
confidence interval of 90% instead of 95%. correlation into account with respect to trend classification
and population alerts.
Testing against a population threshold TrendSpotter may also be useful in other monitoring
schemes, although there are some restrictions to the mini-
Another application of TrendSpotter is the possibility to mum length of the time series and the maximum percent-
test if a smoothed index is above or below a certain pop- age of missing yearly values at the start of the time series.
ulation threshold. If the standard deviations SDt that are Furthermore, the data should not contain too many zero
estimated by TrendSpotter are used, the smoothed popu- values or values close to zero. Its main limitation, however,
lation number for each year in the time series can be is its inability to take into account the uncertainties of the
compared statistically with some predefined aim, for in- imputing models applied to the raw data, and, in that re-
stance the favourable conservation status in the framework spect, a GAM applied to raw data is to be preferred. Cur-
of the EU Birds Directive. For the Greater Scaup this rently, TrendSpotter is best applied in situations where
conservation status was set at a monthly mean of at least GAMs to smooth the time series would be less satisfactory.
25,000 birds (LNV 2006). The smoothed population This is not only in the case of a census; TrendSpotter may
numbers for 1988–1995 were significantly above this va- also be helpful in assessing confidence limits of composite
lue, as can be concluded from a comparison of the lower indices for species groups, a type of biodiversity indicator
confidence limit in Fig. 1 with the solid grey line. From that is becoming increasingly popular (see, e.g. Gregory
1999 until 2003, however, the upper confidence limit was et al. 2005). TrendSpotter is also the method of choice to
below the favourable conservation status, indicating a smooth time series that contain only one value per time
deviation from the threshold. In 2004 no significant dif- point, e.g. the first laying date of Lapwing in the Nether-
ference from the favourable conservation status was found, lands. Many applications of TrendSpotter to composite

123
 J Ornithol

Table 4 Yearly change rate (YCR), its 95% confidence interval and associated trend classification for the Greater Scaup in the Netherlands, as
well as total change rate (TCR), its 95% confidence interval and associated population alerts
Year YCR CIYCR Trend class TCR CITCR Alerts in 2004

1975 0.998 (0.979–1.018) Stable 0.954 (0.540–1.684) No alert for decrease

over the whole
time series
1976 0.990 (0.973–1.008) Stable 0.764 (0.464–1.257) –
1977 0.985 (0.967–1.003) Stable 0.664 (0.409–1.077) –
1978 0.981 (0.963–1.000) Stable 0.612 (0.377–0.993) –
1979 0.979 (0.960–0.998) Moderate decline 0.586 (0.362–0.949) –
1980 0.979 (0.959–0.999) Moderate decline 0.598 (0.369–0.968) No alert for decrease
over the last
25 years
1981 0.980 (0.960–1.001) Stable 0.631 (0.390–1.019) –
1982 0.982 (0.961–1.004) Stable 0.674 (0.417–1.089) –
1983 0.985 (0.963–1.008) Stable 0.732 (0.453–1.183) –
1984 0.989 (0.965–1.013) Stable 0.795 (0.492–1.286) –
1985 0.990 (0.965–1.015) Stable 0.821 (0.508–1.327) No alert for decrease
over the last
20 years
1986 0.986 (0.960–1.012) Stable 0.773 (0.478–1.249) –
1987 0.974 (0.947–1.002) Uncertain 0.641 (0.397–1.037) –
1988 0.958 (0.930–0.988) Moderate decline 0.506 (0.313–0.817) –
1989 0.944 (0.914–0.975) Moderate decline 0.421 (0.260–0.680) –
1990 0.932 (0.901–0.965) Moderate decline 0.375 (0.232–0.607) >25% decrease over
the last 15 years!
1991 0.924 (0.891–0.959) Moderate decline 0.358 (0.222–0.579) –
1992 0.919 (0.883–0.957) Moderate decline 0.365 (0.226–0.590) –
1993 0.920 (0.880–0.961) Moderate decline 0.397 (0.246–0.642) –
1994 0.925 (0.882–0.971) Moderate decline 0.460 (0.285–0.744) –
1995 0.937 (0.888–0.988) Moderate decline 0.554 (0.342–0.898) No alert for decrease
over the last
10 years
1996 0.952 (0.896–1.012) Uncertain 0.676 (0.416–1.099) –
1997 0.970 (0.904–1.041) Uncertain 0.808 (0.494–1.322) –
1998 0.984 (0.906–1.070) Uncertain 0.910 (0.553–1.498) –
1999 0.995 (0.900–1.100) Uncertain 0.977 (0.592–1.614) –
2000 1.003 (0.886–1.135) Uncertain 1.012 (0.616–1.662) No alert for decrease
over the last
5 years
2001 0.996 (0.852–1.164) Uncertain 0.988 (0.619–1.578) –
2002 0.987 (0.809–1.204) Uncertain 0.973 (0.654–1.449) –
2003 0.990 (0.766–1.280) Uncertain 0.990 (0.766–1.280) –
2004 0 – – 0 – –

indices and to single value per year time series are found in many of the advantages of a GAM based on raw data and is
http://www.natuurcompendium.nl. easier to apply than a GAM.
A future development of TrendSpotter will be the Unfortunately, even such a combination may not com-
inclusion of weights for each year in the time series, which pletely solve the conceptual problems in the calculation of
will enable one to incorporate confidence intervals of the confidence intervals of trends in waterbirds. For a census,
indices computed by TRIM or other imputing models. Such currently, there does not seem to be an analysis method
a combination of TRIM and TrendSpotter would offer available to take into account imputation uncertainties

123
J Ornithol

without, at the same time, including between-site variation. kerhout, The Netherlands. Statistics Netherlands, Voorburg/
We have chosen to apply TrendSpotter for the DWMS Heerlen & SOVON, Beek-Ubbergen, pp 663–673
Fewster RM, Buckland ST, Siriwardena GM, Baillie SR, Wilson JD
because we prefer to neglect the imputation uncertainties (2000) Analysis of population trends for farmland birds using
rather than to include them and unavoidably incorporate generalized additive models. Ecology 81:1970–1984
between-site variation as obtained from bootstrapping Gregory RD, van Strien AJ, Vorisek P, Gmelig Meyling AW, Noble
using a GAM. This preference is based on the assumption DG, Foppen RPB, Gibbons DW (2005) Developing indicators
for European birds. Philos Trans R Soc B 360:269–288
that imputation uncertainties in the DWMS are probably Harvey AC (1989) Forecasting structural time series models and the
small, as the number of missing counts is limited and ex- Kalman filter. Cambridge University Press, London
tremely high imputed values are excluded from the trend LNV (2006) Natura 2000 doelendocument. Dutch Ministry of
analysis. Agriculture, Nature and Food Quality, Den Haag
McCullagh P, Nelder JA (1989) Generalised linear models. 2nd edn.
Two additional features of TrendSpotter should be no- Chapman & Hall, London
ted. First, the programme may be run on time series with Nobel P de, van Turnhout C, van der Winden J, Foppen R (2002) An
cyclic patterns, as are usually present in monthly counts of Alert System for bird population changes on a national level and
waterbirds. This could be interesting in the analysis of for EU Bird Directive monitoring: a Dutch approach. SOVON-
onderzoeksrapport. SOVON Vogelonderzoek, Nederland, Beek-
shifts in the seasonal patterns of birds as an effect of, e.g., Ubbergen
climate change. Second, similar to applying covariate Pannekoek J, van Strien A (2001) TRIM (TRends and Indices for
models in a GAM, explanatory variables may be added to Monitoring data). Research paper no. 1020, Centraal Bureau
each record in the dataset, which enables the detection of voor de Statistiek (CBS), Voorburg
van Roomen M, Koffijberg K, Noordhuis R, Soldaat L (2006a) Long-
causal factors for observed changes in population abun- term waterbird monitoring in The Netherlands: a tool for policy
dance. and management. In: Boere GC, Galbraith CA, Stroud DA (eds)
Waterbirds around the world. The Stationary Office, Edinburgh,
Acknowledgements This publication would not have been possible pp 463–467
without the data collected by many volunteer bird counters, and the van Roomen M, van Winden E, Koffijber K, Ens B, Hustings F,
financial support of the Dutch Ministry of Agriculture, Nature and Kleefstra R, schoppers J, van Turnhout C, SOVON Ganzen- en
Food quality, the Ministry of Transport, Public Works and Water zwanenwerkgroep, Soldaat L (2006b) Watervogels in Nederland
Management and Vogelbescherming Nederland. Thanks are also due in 2004/2005. SOVON-monitoringrapport 2006/02, RIZA-rap-
to A. Gmelig Meyling for computerizing the calculations described in port BM06.14, SOVON Vogelonderzoek Nederland, Beek-
the paper and to J. Blew and an anonymous reviewer for helpful Ubbergen
comments on a previous draft of the paper. Soldaat L, van Winden E, van Turnhout C, Berrevoets C, van
Roomen M, van Strien A (2004) De berekening van indexen en
trends bij het watervogelmeetnet. SOVON-onderzoeksrapport
2004/02. Centraal Bureau voor de Statistiek, Voorburg/Heerlen
References van Strien A (2006) Landelijke Natuurmeetnetten van het NEM in
2005. Kwaliteitsrapportage NEM. Centraal Bureau voor de
Atkinson PW, Austin GE, Rehfisch MM, Baker H, Cranswick P, Statistiek, Voorburg/Heerlen
Kershaw M, Robinson J, Langston RHW, Stroud DA, van Underhill LG, Prŷs-Jones RP (1994) Index numbers for waterbird
Turnhout C, Maclean MD (2006) Identifying declines in populations. (I) review and methodology. J Appl Ecol 31:463–
waterbirds: the effects of missing data, population variability 480
and count period on the interpretation of long-term survey data. Visser H (2004) Estimation and detection of flexible trends. Atmos
Biol Conserv 130:549–559 Environ 38:4135–4145
Bell MC (1995) UINDEX4. A computer program for estimating Visser H (2005) The significance of climate change in the Nether-
population index numbers by the Underhill method. The lands. An analysis of historical and future trends (1901–2020).
Wildfowl & Wetland Trust, Slimbridge MNP report 550002007. http://www.mnp.nl/en/publications/
Braak CJF ter, van Strien A, Meijer R, Verstrael TJ (1994) Analysis 2005/The_significance_of_climate_change_in_the_Nether-
of monitoring data with many missing values: which method? In: lands.html
Hagemeijer W, Verstrael TJ (eds) Bird numbers 1992. Distri- Visser H, Petersen AC (2007) Outdoor skating and climate change.
bution, monitoring and ecological aspects. Proceedings of the Climatic Change (in press)
12th International Conference of IBCC and EOAC, Noordwij-

123

View publication stats