The effect of habitat type on farmland bird populations
In
Tarnava Mare Natura2000 reserve, Romania
Krisztina Csiki
Degree project in biology, Master of science (2 years), 2020 Examensarbete i biologi 45 hp till masterexamen, 2020
Biology Education Centre and the Department of Ecology and Genetics, Uppsala University Supervisor: Professor Anssi Laurila
External opponent: Dr. William Jones
Abstract
Widespread extinction is a critical threat to biodiversity and is largely caused by human overexploitation of habitat and populations. A widely used and hence well studied organism group for indication of biodiversity is birds. In Europe especially, farmland specialists have suffered from intensified agricultural practices such as increase of monoculture, use of pesticides, and heavy machinery. This has been shown to be partly caused by an EU legislation called the Common Agricultural Policy (CAP). A specific type of farmland, termed High Nature Value (HNV) farmland, seems to be particularly advantageous for farmland specialist birds and makes up an important conservation target.
The current study was done in the Natura2000 reserve Târnava Mare, Romania, to find out which habitat types play an essential role for occurrence of farmland species. Farmlands in Târnava Mare are highly diverse in structure, characterizing a mosaic of grassland, meadows and fields, and low-intensity farming practices. With bird point count survey data from 2015 to 2019, I evaluated the effect of different habitat types on five species listed in the Farmland Bird Indicator (FBI) and as farmland specialists: red-backed shrike (Lanius collurio),
yellowhammer (Emberiza citrinella), Eurasian skylark (Alauda arvensis), Eurasian tree sparrow (Passer montanus), and common whitethroat (Sylvia communis).
I compared habitat proportion in presence and absence of the species for 2019’s data with Mann-Whitney tests. They all showed significant results for meadow proportion. All species except the common whitethroat showed significant results for crop proportion, while only two species (red-backed shrike and yellowhammer ) showed significant results for scrub.
Independent of which habitat type was tested (meadow, crop or scrub), all species with significant result – except for Eurasian skylark with a negative relationship in crop habitat - showed a positive response to a higher proportion of the tested habitat.
The same species except Eurasian tree sparrow were modelled with the generalized N-mixture model of Dail and Madsen (2011) to evaluate what is influencing abundance, recruitment rates, survival probabilities and detectability over five years. The day of the season affected the detectability of almost all species. The effect of habitat on recruitment rate and survival probability, however, could only be shown for yellowhammer. For the latter, proportion of meadow affected recruitment and proportion of reed affected survival.
In conclusion, the presence of species seems to be generally higher in habitats associated with low-intensity farming on the single season scale. Over time, however, a significant effect on population dynamic parameters for the same species could not be shown for most species.
This could be a result of insufficient data for each year, too few years of data, or that the
tested habitat types are not affecting these parameters over time.
2
Table of Contents
Introduction ... 3
Methods and material ... 5
Results ... 12
Discussion ... 16
Acknowledgements ... 20
References ... 21
Appendix I ... 24
Appendix II ... 26
Appendix III ... 29
Appendix IV ... 32
Appendix V ... 35
3
1. Introduction
1.1 Background Loss of biodiversity
We are in the midst of a sixth mass extinction and actions to reverse it are urgent (Ripple et al., 2017). Conserving species-rich habitat types has a growing importance, especially
considering climate change together with land use- and intensity shifts. In a study from 2017, the abundance of vertebrate species worldwide was estimated to have decreased to only 40%
of the number in 1970 (Ripple et al., 2017). In fact, recent species range reductions have been shown to be associated with human activities (reviewed in MacKenzie et al., 2018 p. 32), and in a study of 8’688 species listed as threatened (includes categories VU, EN, and CR) by the IUCN Red List, the drivers of biodiversity loss were found to result largely from human overexploitation (unsustainable hunting, fishing, and logging) and large-scale agricultural practices (Maxwell et al., 2016). Furthermore, the magnitude of extinction debt – damage to biodiversity that is not yet evident due to slow population response – could make the situation even worse, as more habitats are fragmented or destroyed (Tilman et al., 1994). One
suggested action to remedy defaunation is to stop the global trend of converting native, species-rich habitats like grasslands to other types of land. In Europe, grassland biodiversity has dropped dramatically (SOER, 2015). Halting injury and loss of low-intensity agricultural landscapes should therefore be a top priority in biodiversity conservation (Union of
Concerned Scientists, 1992).
Agricultural legislation in the EU
Agriculture makes up approximately 40% of the land cover in Europe (SOER, 2015). In the EU, agricultural practices are regulated by the Common Agricultural Policy (CAP), which was first introduced in 1962 with the main objective to secure food supplies for the human population. This objective has been fulfilled, but with a number of unforeseen consequences from increased use of machinery, monocultured fields, and the use of nutrients and pesticides to maximize efficiency (Andersen et al., 2003) - negatively affecting biodiversity (EEA, 2004).
CAP has developed over the years. The current plan for 2014-2020 has, in theory, become more oriented toward sustainable use of natural resources and less focused on high
productivity. For instance, farmers have the opportunity to receive subsidies for small-scale, organic practices and integrating crop-rotation and flowering strips in their fields (EEA, 2017). However, the legislation design is still heavily criticized by organisations like WWF and Greenpeace, because the funding for subsidies given to intensify crop yield are
nonetheless larger than that promoting biodiversity. Moreover, loopholes in the policy are considered a critical problem.
High Nature Value farmlands
Agricultural intensity is a somewhat ambiguous term but is most commonly defined as “the
ratio of inputs and outputs within an agricultural system, i.e., in terms of yield per land area
and per input unit” (reviewed in Ruiz-Martinez et al., 2015). It is also important to have a
clear definition of High Nature Value (HNV) as “value” can also be interpreted in different
ways. As defined by Andersen et al. (2003), HNV farmlands are “areas in Europe where
4
agriculture is a major (usually the dominant) land use and where that agriculture supports or is associated with either a high species and habitat diversity or the presence of species of European conservation concern or both”. They are often highly diverse in structure and include various landscape types, such as salt marshes and alpine pastures. According to the EEA’s report on HNV farmland (2004), low-intensity agriculture in the form of semi-natural grasslands serves as biodiversity hotspots (likely due to the intermediate disturbance it gives rise to). Especially farming in Central and Eastern Europe is important in this aspect. One of the three major types of HNV farmland are those that consist of a mosaic of habitat and land use (Andersen et al., 2003). My study sites in Romania fall into this category.
Romania has been subject to the CAP since entering the EU in 2007. As a means to maintain plants and animals inhabiting HNV grasslands, Transylvanian farmers are since 2008
rewarded for management that is based on traditional practices according to a number of criteria: “1) use of natural fertilizers such as farmyard manure, 2) avoidance of over-grazing, with low stocking rates (no more than 1 cow or 5 sheep per hectare), 3) beginning mowing at a later date to allow plants to seed, butterflies to emerge, and ground nesting birds to fledge, and 4) encouraging mowing by scythe or the use of small machines, rather than by heavy tractors that damage soil structure and kill young animals unable to escape” (Akeroyd &
Bădărău, 2012).
Birds as indicators of biodiversity
Birds are good indicators of biodiversity in agricultural areas since they depend on various animals (e.g. insects, rodents) and plants for survival. Moreover, they are a well-studied group with extensive previous monitoring data (Donald et al., 2002). Between 1970 and 2000, farmland birds in Europe showed a significant declining trend, in contrast to bird assemblages characterizing other habitat types (Donald et al., 2001; Donald et al., 2006). During this period, the biodiversity index fell with 40% for farmland birds in the UK (Gregory et al., 2003). Among others, wheat yield and use of pesticides were negatively correlated with abundance on the European level (Donald et al., 2006). The extirpation of insects with insecticides in these areas is detrimental for insectivorous birds, which in France – the EU country with steepest decline of farmland birds between 2001 and 2017 - is especially evident in meadow pipits and whinchats (Greshko & Chung, 2018). HNV farmlands in particular seem to be important for the survival of such specialists, as the community composition in these habitats consists of a higher ratio of specialist species as compared to non-HNV farmlands (Doxa et al., 2010).
A recent study by Reif and Vermouzek (2019) in the Czech Republic showed that populations
of farmland birds decline as a result of the country’s entry into the EU and CAP in 2004. The
authors compared standardized national inventory data before and after 2004, which showed
that agricultural intensity (quantified as crop yield of five different crops) is the single
significant predictor involved in the negative population trends, even when considering
several life history traits and ecology of the different species (Reif & Vermouzek, 2019). The
same analysis was conducted for forest birds for comparison, with no effect of entry to the
EU, suggesting that this is a farmland bird-specific issue. Similar population trend results
were shown in Sweden. In the annual Swedish Bird Survey of 2018, a significant negative
trend was found for farmland birds in particular (Svensk Fågeltaxering, 2018), of which
almost all species overlap with Romanian avifauna.
5 1.2 Objective and research questions
The main objective of the present study is to evaluate the bird populations inside the Târnava Mare Natura2000 reserve in Transylvania, in terms of how they are affected by different habitat types in a High Nature Value farmland landscape. This will be done by using point count survey data gathered between 2015 and 2019. Previous studies have focused on the Farmland Bird Indicator (FBI) method, which is used both nationally and on the EU level to make estimates of breeding farmland bird populations (Gregory et al., 2005; Scholefield et al., 2011). Here, I will instead use comparison of habitat proportion in presence and absence of species in 2019 and hierarchical modelling on populations between 2015 and 2019. Both analyses will be done on some species that are listed in the FBI. To my knowledge, this has not been done on Transylvanian bird populations before. Thus, it will hopefully add valuable information to the broader picture of farmland specialist populations in the EU.
Questions to be answered are:
1. In which habitat type(s) within the study area in Târnava Mare do we mostly find the species in the single season of 2019?
2. Which habitat type(s) within the study area in Târnava Mare affect population dynamic parameters like recruitment rates and survival probabilities during the time period 2015- 2019?
Recruitment rate in (2) is referring to the rate at which new individuals are born and/or immigrate into the population each year. Similarly, survival probability here refers to how likely extant individuals in the population are to stay (not die or emigrate) in the population each year. These parameters were estimated mathematically by a type of hierarchical model called the N-mixture model, designed by Dail and Madsen (2011). A more precise estimate of recruitment and survival is not possible in this type of study, since individuals are not marked.
Considering extant literature about the declines of farmland birds and the reasons for the decline (i.e. intensification and homogenization of the agricultural landscape), I hypothesize that they favour habitat types associated with low intensity farming, e.g. meadows, orchards and scrubby areas, rather than an environment consisting largely of cropfields. In other words, their occurrence, recruitment rates and survival probabilities should be (positively) influenced by these habitat types.
2. Methods and material
2.1 Study Site
The area lies within the Carpathian basin and is part of the 85’000 ha Natura2000 site called Târnava Mare. The climate in Târnava Mare is temperate with an average min/max
temperature in Sighișoara at 12/25°C, humidity of 72% and precipitation of 86 mm in July (World Weather & Climate Information, 2019). The protected habitat types in Târnava Mare Natura2000 site are “Scrub habitats and semi-dry grassland over limestone or other
calcareous substrates, with important orchid sites” (6210) and “Sub-Pannonic steppic grasslands” (6240) (Akeroyd & Bădărău, 2012).
The 2019 surveys took place in seven villages located in Mureș county, Romania. These were
(in order of survey): Richiș, Meșendorf, Nou Săsesc, Viscri, Criț, Mălâncrav, and Apold (Fig.
6
1). One more village, Daia, was surveyed previous years. The survey thus covers eight locations within an area of about 50 km
2. All survey sites are located both in and near the villages and so cover the characteristic mosaic landscape of more or less intensely farmed land (mostly maize and hop), meadows, reeds, grasslands, orchards, pastures, and woodland (Fig. 2). The closest city is Sighișoara, which is located ca 250 m above sea level. The furthest village (Viscri) is 42 km from the city.
Figure 1. Map of study sites South of Sighișoara, Romania. The surveyed villages are: Richiș, Nou Săsesc, Mălâncrav, Apold, Angofa, Meșendorf, Criț and Viscri. Daia, which was surveyed previous years, is also included.
Figure 2. A characteristic landscape inTârnava Mare Natura2000 reserve, Romania. A mosaic of low-intensity farmland, meadows, grasslands, pastures, and woodland. Photo: © 2019 Krisztina Csiki
7 2.2 Study Species
The Romanian bird fauna consists of about 395 species (Avibase, 2018). The present study will focus on some of which are listed in the European Farmland Bird Indicator (Scholefield et al., 2011) and categorized as farmland specialists by Gregory et al. (2005) (Table 3). Some of these species, which have been frequently recorded in previous years of the study, are yellowhammer (Emberiza citrinella), meadow pipit (Anthus pratensis), and common whitethroat (Sylvia communis). Most of the recorded species have the current IUCN status LC, with the exception of the Northern lapwing (Vanellus vanellus, VU), turtle dove
(Streptopelia turtur, VU), and meadow pipit (Anthus pratensis, NT). However, many of the LC-species are showing declining trends on the European scale (e.g. linnet, common quail, and red-backed shrike). See Fig 3. for examples of study species.
2.3 Data Collection
The villages have been surveyed with bird point counts every season (June-Aug) since 2015.
Each village was surveyed during one week per season, with the same point-transects (except for some locations that were added, removed or moved in the first three years of the study).
Below is a description of the point count survey. The habitat data is purely based on observations, because of lack of fine-grained land cover maps - which would be needed to capture the high variability of the landscape (Dr. Bruce Carlisle, pers. comm.).
Figure 3. Examples of species commonly seen in agricultural landscapes in Europe. Upper left: Red-backed shrike (Lanius collurio), upper right: common whitethroat (Sylvia communis), bottom left: Eurasian tree sparrow (Passer montanus), bottom right: yellowhammer (Emberiza citrinella). Photos: © 2019 Krisztina Csiki
8
Three point-transects were surveyed in all villages for 42 days between the 20
thof June and the 12
thof August 2019 (Fig. 4). The transects were selected so that as much representative land covers as possible was captured for each village, to allow for inference about the larger area of interest (MacKenzie et al., 2002). The central point-transect was on a 4-4.5 km long route on the valley floor which crossed the village itself and some more intensely farmed land. The East and West or North and South point-transects were semi-circular routes of ca 6 km, reaching from the valley floor and up to the sides of the valley. The surveyed points (7- 12) were located 500 m apart on each point-transect. These points were surveyed for 10 min (similar to Alves et al., 2019), with standing point counts of every seen and/or heard bird. The data was collected from dawn until about 11 AM, when bird activity usually dropped. To avoid bias as the day progressed, the survey was temporally replicated for each transect in opposite direction, on separate days. Single points or whole surveys were cancelled in the case of heavy rain or if sheepdogs were known to be nearby.
Figure 4. Map of surveyed points on each point-transect in the villages: Richiș, Nou Săsesc, Mălâncrav, Apold, Meșendorf, Criț and Viscri. Daia, which was surveyed previous years, is also included.
Land cover was determined within a 100 m radius surrounding each point along the transect, by dividing a surrounding circle into North, East, South, and West quadrants and estimating proportion of seven habitat categories: reedbed, meadow, crop, pasture, settlement, wood, and orchard. Cloud cover (0-4), wind intensity (Beaufort scale 0-5), and main observer was also recorded. The latter is frequently used in analysis of avian point counts since the detection skill and experience between survey leaders can vary (reviewed in MacKenzie et al., 2018 p.
13).
9 2.4 Data analysis
Correlation tests
To choose the variables included in the analyses, I conducted multiple correlation tests on the variables collected during surveys from 2015 to 2019 – one for all years and one for the single year 2019 – to remove one of each pair showing correlation. This was done separately for the habitat (the proportion of a specific habitat type at a point) and weather variables, respectively (Appendix I). See Table 1 for a complete list of variables tested.
Table 1. Table of detection and habitat variables collected in the study. These were tested for correlation and one of each correlated pair were removed, to avoid interdependency of explanatory variables in the statistical tests in which they will be used. Since “observer” is a factor, this was not tested.
Detection variables Habitat variables
• Day Crop
• Observer Meadow
• Wind intensity Orchard
• Cloud cover Pasture Reed Scrub Village Wood
The habitat variables remaining for 2019 were meadow, crop and scrub. For 2015-2019, they were reed and meadow. The weather variables remaining for the 2015-2019 were survey day, cloud and observer.
Occurrence of species in CAP-habitats 2019
To evaluate the occurrence of species in different habitat types in 2019, I compared the proportions of habitats (meadow, crop and scrub) in presence and absence of each species.
Because the data were not normally distributed, I chose an unpaired non-parametric Mann- Whitney test.
Background to Hierarchical Modelling
Hierarchical modelling allows for analysis of various kinds of ecological data regarding abundance, occupancy or population changes over time (Royle, 2004). The major advantage of these types of models is that they allow for modelling both the ecological state
process/development and how that state is observed. In other words, the observational process is considered, recognizing for example false absences during data collection (i.e. detectability
< 1) (MacKenzie et al., 2002). False absence occurs when a species is present but fails to be detected in the survey - leading to an underestimation of presence. This can result from pure chance or that a particular species is shy or cryptic. In that latter case, the song/call might be the only way to detect it (Royle, 2004).
One can choose to model the populations as either open or closed, on single- or multiple
season scale, with either single- or multiple species. Analyses on multiple season scale are
sometimes called “dynamic” models, because they allow for approximations of population
dynamic parameters like recruitment and survival.
10
Recruitment in hierarchical modelling refers to the number of individuals which are born and/or immigrate into the population (N) each year (births + immigration). Similarly, survival refers to the number of individuals which stay (do not die or emigrate) in the population each year (N
t-1- deaths - emigrants). The same is thus also true for the rates of the respective population dynamic parameters. This means that the terms are used more liberally than in mark-recapture studies where a more precise estimate of actual recruitment (births) and survival (non-deaths) is possible.
Here, I will use data on single species and multiple season scale to include all five years of the study. This is done with generalized N-mixture models that allow for open population
analysis of point count data (Dail & Madsen, 2011). N-mixture models can give fairly good abundance estimates even with small data sets, when counts are low, and when there is an over-representation of zeros (i.e. zero-inflation) (Royle, 2004). Such an open population model for point count data is made up of four main formulas:
1. Mean abundance at each site (λ)
2. Recruitment – reproduction and immigration (R), estimated through recruitment rate (gamma, γ)
3. Survival – deaths and emigration (S), estimated through survival probability (omega, ω) 4. Detectability – probability of detection (p)
where R ~ Poisson(N×γ) and S ~ Bin(N, φ) under the auto-regressive model. N refers to the total abundance in the survey area and φ is the per capita survival probability. Mean
abundance, recruitment rate, survival probability and detectability are Maximum Likelihood Estimators (MLE) obtained from numerical optimization techniques (Dail & Madsen, 2011).
These four formulas represent population dynamic parameters with or without covariates collected during survey, which are potentially influencing the particular process (Dail &
Madsen, 2011). For detectability, covariates which might influence the observational process are tested (e.g. weather variables). For the rest of the three components of the model,
covariates that might influence the ecological processes are tested (e.g. habitat variables). The covariate models are then compared with the Akaike Information Criterion against the null model, where the latter states that the individuals are distributed randomly because nothing is influencing neither the observational nor the ecological processes. This could be true in a homogenous landscape but can cause problems in the estimates in heterogeneous habitats (Royle, 2004).
Effects of habitat type on population dynamic parameters
Data were analysed with the generalized N-mixture (hierarchical) model of Dail and Madsen (2011) in the open-source software R version 3.5.1 (R Core Team, 2018). Packages used were mainly ggplot2 (version 3.2.1), PerformanceAnalytics (version 1.5.3), and unmarked (version 0.12-3). The generalized N-mixture model is specifically developed for point count data (Dail
& Madsen, 2011). As background reading on hierarchical models, I used chapter 3-5 of
“Applied hierarchical modelling in ecology” (Kéry & Royle, 2015), the article introducing the
unmarked package (Fiske & Chandler, 2012), and the unmarked package documentation.
11 Workflow:
A) Prepare the data for each species so that it matches the required structure for the R package “unmarked”.
B) Create “unmarked dataframes” adapted for open population, multiple season point count data. This includes the recorded frequencies in the surveys between 2015 and 2019 (y1.1, y1.2, …, y5.2), the variables for detection (obsCovs) and the variables for recruitment rate and survival probability (yearlySiteCovs). There are no covariates for mean abundance in the first sampling season (siteCovs).
C) Check the distribution of data for each species with a histogram.
D) Create N-mixture models by first inserting different combinations of the obsCovs, yearlySiteCovs and the null model for siteCovs.
E) Select the type of mixture to be used in the model (“Poisson”, “negative binomial” or
“zero-inflated Poisson distribution”) by looking at the distribution of data in the histogram.
F) Finally, select the type of population dynamics (“constant”, “auto-regressive”, “no trend”,
“trend”, “Ricker-logistic” or “Gompertz-logistic”) and whether immigration should be allowed within seasons.
G) Fit the N-mixture models and test them against the null models, which state that the population dynamic parameters and detectability are not influenced by any covariates (Joseph et al., 2009). The models are tested with the Akaike Information Criterion (AIC) to evaluate parsimony. I will consider as top models those that have a ΔAIC < 2 (Burnham
& Anderson, 2002 p. 97-99). If the null model is included in the best set, I will interpret it as there is no considerable effect of any covariates.
H) Use the most parsimonious model for each species. Check the estimated population size with the “best unbiased predictor” (uses the estimated posterior distribution of N for each site). Check the estimated λ, γ, ω, and p by back-transformation of each parameter.
After plotting a histogram of the collected data, I chose to use zero-inflated Poisson distribution (ZIP) as input for “mixture” in the models (Appendix II). According to Joseph and colleagues (2009), this is advantageous because it “models an ecological mechanism rather than a statistical phenomenon”, as opposed to the widely used negative binomial distribution. Moreover, it is ideal for bird point count surveys because non-independence can arise from clumped observations (flocking birds like starlings) (Joseph et al., 2009).
Regarding immigration within seasons I chose not to allow it, following the logic of Royle
(2004) stating that considering the time in the year when the study took place (June-Aug), the
observed birds are probably local breeders which have established territories. Thus, no foreign
individuals will appear with time within a season. As input for population dynamics, I chose
to use auto-regression which assumes that the response variable at time t is dependent on the
value at time t-1. The detectability covariates used for model creation were survey day,
observer and cloud cover. For recruitment rate (gamma, γ) and survival probability (omega,
ω), the habitat covariates meadow and reed were used. No covariates in the study were
assumed to be the same for all years and therefore I used the null model for the mean
abundance-formula (lambda, λ) (Table 2).
12
Table 2. Summary of input for the Single species-Multiple season generalized N-mixture models for open populations. Two covariates were tested for recruitment rate and survival probability, and two covariates plus one factor were tested for detectability. ZIP = Zero-inflated Poisson, Autoreg = Auto-regression.
Lambda (λ)
Mean abundance
Gamma (γ)
Recruitment rate
Omega (ω)
Survival probability
Detectability (p)
Detection probability
Mixture
Distribution of data
Dynamics
Population dynamics
• Null • Meadow • Meadow • Day • ZIP • Autoreg
• Reed • Reed • Observer
• Cloud cover
3. Results
3.1 Data summary
A total of 19’088 birds and 90 species were recorded in 2019’s survey (see Table 3 and Appendix III for focal species, and Appendix IV for the full list of species). New species for 2019 year were the common coot and wheatear. The single most frequently heard or seen species was the starling, whereas the Eurasian hoopoe, barred warbler, tawny owl, and common redstart were some of the rarest.
Some data from the sites were excluded: points which were only surveyed for one year, and observations that were not identified to species level (some records in 2015). Out of the remaining data, I selected only the observations with species listed in the Farmland Bird Indicator (Scholefield, 2011) and/or as farmland specialists (Gregory et al., 2005) (Table 3).
Table 3. Bird species recorded in the bird point count (BPC) surveys in Târnava Mare during the field season of 2019. Only species which are listed in the Farmland Bird Indicator (FBI) (Scholefield, 2011) and/or listed as farmland specialists (Gregory et al., 2005) are included here.
Species (scientific) Species FBI Specialist BPC freq.
Alauda arvensis Eurasian skylark Yes Yes 107
Anthus campestris Tawny pipit Yes No 14
Anthus pratensis Meadow pipit Yes No 0
Carduelis cannabina Linnet Yes Yes 19
Carduelis carduelis European goldfinch No Yes 246
Ciconia ciconia White stork Yes No 68
Corvus frugilegus Rook Yes No 284
Coturnix coturnix Common quail No Yes 12
Emberiza calandra Corn bunting Yes Yes 28
Emberiza citrinella Yellowhammer Yes Yes 249
Falco tinnunculus Common kestrel Yes No 17
Lanius collurio Red-backed shrike Yes Yes 840
Passer montanus Eurasian tree sparrow Yes Yes 814
Saxicola rubetra Whinchat Yes No 33
Saxicola rubicola Stonechat No Yes 57
Serinus serinus Serin Yes No 3
Streptopelia turtur Turtle dove Yes Yes 18
13
Sturnus vulgaris Common starling Yes Yes 4050
Sylvia communis Common whitethroat Yes Yes 94
Upupa epops Eurasian hoopoe No Yes 2
Vanellus vanellus Northern lapwing Yes Yes 34
Total
21 17 14 6989
3.2 Occurrence of species in CAP-habitats 2019
Three habitat types remained after the multiple correlation test for 2019. These were meadow, crop and scrub. They were tested with five different species separately: red-backed shrike (Lanius collurio), yellowhammer (Emberiza citrinella), Eurasian skylark (Alauda arvensis), Eurasian tree sparrow (Passer montanus), and common whitethroat (Sylvia communis).
The unpaired Mann-Whitney tests showed significant results (p < 0.05) for all five species when testing meadow. For crop, all species except common whitethroat showed significant results. And lastly, red-backed shrike and yellowhammer showed significant results for scrub (Table 4). Independent of which habitat type was tested (meadow, crop or scrub), all species with significant result – except for Eurasian skylark in crop habitat – occurred in higher proportion of that particular habitat, i.e. the mean value for the habitat was higher where the species was present compared to where it was absent (Fig. 5-7).
Table 4. Statistical results fromnon-parametric unpaired Mann-Whitney tests on the proportion of habitat types for five species: red-backed shrike, yellowhammer, Eurasian skylark, Eurasian tree sparrow and common whitethroat. Data from 2019 were used. Significant results are marked with *.
Meadow in presence of species
Meadow in absence of species
M SD M SD df W-stat p-value
Red-backed shrike 0.27 0.27 0.12 0.23 188 7532 1.25e-06*
Yellowhammer 0.17 0.28 0.11 0.25 182 7123 4.33e-03*
Eurasian skylark 0.33 0.33 0.19 0.24 46 4084 0.03*
Eurasian tree sparrow 0.26 0.26 0.19 0.27 168 6506 0.01*
Common whitethroat 0.45 0.32 0.16 0.22 50 5406 8.75e-08*
Crop in presence of species
Crop in absence of species
M SD M SD df W-stat p-value
Red-backed shrike 0.18 0.25 0.07 0.16 212 6785 5.46e-4*
Yellowhammer 0.17 0.24 0.11 0.21 182 6740 0.04*
Eurasian skylark 0.08 0.20 0.15 0.23 61 2693 0.02*
Eurasian tree sparrow 0.19 0.23 0.11 0.22 150 6398 0.01*
Common whitethroat 0.16 0.23 0.13 0.22 59 3783 0.4832
Scrub in presenceof species
Scrub in absence of species
M SD M SD df W-stat p-value
Red-backed shrike 0.14 0.13 0.09 0.15 149 7509 3.59e-06*
Yellowhammer 0.15 0.15 0.10 0.13 185 7542 2.43e-04*
Eurasian skylark 0.14 0.16 0.12 0.14 50 3927 0.10
Eurasian tree sparrow 0.11 0.09 0.12 0.16 213 5850 0.24
14
Common whitethroat 0.11 0.10 0.12 0.15 92 3739 0.63
Figure 5. Proportion of meadow habitat in presence and absence of five species. Mann-Whitney tests showed significance (p < 0.05) for all. See Table 4 for detailed statistical test results.
Figure 6. Proportion of crop habitat in presence and absence of five species.Mann-Whitney tests showed significance (p < 0.05) for all except common whitethroat. See Table 4 for detailed statistical test results.
15
Figure 7. Proportion of scrub habitat in presence and absence of five species. Mann-Whitney tests showed significance (p < 0.05) for red-backed shrike and yellowhammer. See Table 4 for detailed statistical test results.
3.3 Effects of habitat type on population dynamic parameters
The same species except Eurasian tree sparrow were modelled here: red-backed shrike (Lanius collurio), yellowhammer (Emberiza citrinella), Eurasian skylark (Alauda arvensis), and common whitethroat (Sylvia communis). A total of 12 models per species were tested against the null model with AIC (Appendix V). This was done in a two-step approach where I first selected the best model for detectability, and then used that when proceeding to model recruitment rate and survival probability. In this way, I didn’t have to create as many models and hence saved some time. The most parsimonious (or “top”) models, meaning ΔAIC < 2, are presented in Table 5.
For red-backed shrike, yellowhammer, and common whitethroat, the day in the season influenced detectability the most. It showed a negative influence on yellowhammer and common whitethroat, meaning that they were more difficult to detect as the seasons progressed. Red-backed shrike, on the other hand, became easier to detect (positive
influence). In contrast, cloud cover influenced skylark detectability the most (Table 5), and in a negative way (the thicker cloud cover, the more difficult to detect). When it comes to recruitment rates and survival probabilities, the only species showing higher parsimony with AIC for any covariates was the yellowhammer. The most parsimonious covariate models were meadow for recruitment rate and reed for survival probability (Table 5). Meadow proportion influenced recruitment positively and the same is true for reed proportion on survival.
Estimates of mean abundance per sampling site varied between 1.94 (common whitethroat)
and 6.53 (red-backed shrike), meaning that there were on average 2 whitethroats and 7 red-
backed shrikes present at each surveyed point (Table 6). Eurasian skylark had the lowest
recruitment rate (0.15) while yellowhammer had the highest (1.63). The latter also showed the
16
highest survival probability of 1.00. Population sizes in the area in the first season were estimated to ~248 skylarks, compared to ~2790 red-backed shrikes (Table 6).
Table 5. Parsimony of the tested models on multiple season scale 2015-2019 for red-backed shrike,
yellowhammer, Eurasian skylark, and common whitethroat. The top ranked model (bold) is presented together with the null model, and the model with the top ranked detectability covariate. If there are no top ranked models with covariates for recruitment rate and/or survival probability, only the two latter are presented.λ = lambda or mean abundance per sample site; γ = gamma or recruitment rate; ω = omega or survival probability;
p = detectability.
Species Model nPars AIC ΔAIC AICwt
cumltvWt
Red-backed shrike
λ(.) × γ(.) × ω(.) × p(Day)6 2202.96 0.00 2.5e-01 0
λ(.) × γ(.) × ω(.) × p(.)5 2234.46 31.50 3.6e-08 1 Yellowhammer
λ(.) × γ(Meadow) ×ω(Reed) × p(Day)
8 1401.12 0.00 2.5e-01 0
λ(.) × γ(.) × ω(.) × p(Day)6 1405.19 4.08 3.3e-02 0
λ(.) × γ(.) × ω(.) × p(.)5 1459.53 58.42 5.2e-14 1 Eurasian skylark
λ(.) × γ(.) × ω(.) × p(Cloud)6 500.39 0.00 0.2565 0
λ(.) × γ(.) × ω(.)× p(.)5 510.27 9.88 0.0018 1 Common
whitethroat
λ(.) × γ(.) × ω(.) × p(Day)
6 688.74 0.00 2.8e-01 0
λ(.) × γ(.) × ω(.) × p(.)5 720.14 31.40 4.2e-08 1
Table 6. Estimates of the population dynamic parameters mean abundance per sample unit, recruitment rate, survival probability, and detection probability. The estimates were obtained with the top ranked models for each species respectively. See models in bold in Table 5.
Lambda (λ)
Mean abundance
Gamma (γ)
Recruitment rate
Omega (ω)
Survival probability
Detectability (p)
Detection probability
Population size (N)
Year 2015 Red-backed
shrike
6.53 1.15 0.01 0.51 2790
Yellowhammer 3.77 1.63 1.00 0.49 1861
Eurasian skylark 2.96 0.15 0.88 0.44 248
Common
whitethroat 1.94 1.01 0.03 0.48 717
4. Discussion
4.1 Occurrence of species in CAP-habitats 2019
This study focused on five species listed in the Farmland Bird Indicator (Scholefield et al., 2011) and as farmland specialists (Gregory et al., 2005): the red-backed shrike,
yellowhammer, Eurasian skylark, Eurasian tree sparrow, and the common whitethroat. The
surveyed locations lie within an area of approximately 50 km
2. I predicted that the occurrence
of these species would be higher at sites where the proportions of meadow and scrub were
17
high, which I investigated with a Mann-Whitney test on data from 2019. I was able to confirm this hypothesis for all species when it comes to meadow habitat and two species (red-backed shrike and yellowhammer) when it comes to scrub habitat. Four species also showed
significant results for crop habitat (exception: common whitethroat, p = 0.48), but the Eurasian skylark showed a negative relationship in contrast to the others. Here it is worth mentioning that since the data is very zero-inflated, the barplots (Fig. 5-7) show many of the counts as outliers.
The results imply that indeed, most species seem to be tied to habitats that characterize low- intensity farming (i.e. a mosaic of meadow, crop and scrub). The data was by far most robust for red-backed shrike and yellowhammer, which could be one reason for finding significant results for those particular species in all three habitats including scrub. That the red-backed shrike had a significantly higher abundance in more scrubby areas than less scrubby areas was not surprising, considering that almost all detections were of individuals sitting in bushes. For meadow and crop habitat, perhaps more surveyed points would have made the picture clearer for the rest of the species, too.
That the occurrence of skylark was in fact significantly lower in crop habitat could mean that the species does not thrive in intensely managed areas, even though they are open-land species (Svenning, 2007). An effective measure taken in Sweden to protect the Eurasian skylark is to establish so called “lark squares” (BirdLife Sweden et al., 2018). A lark square is a small area (16-20 m
2) within a field which is left unsown, allowing Eurasian skylarks to land, forage and nest. A three-year study showed that this method can increase the abundance of Eurasian skylarks up to 60% (BirdLife Sweden et al., 2018).
4.2 Effects of habitat type on population dynamic parameters
I also predicted that the recruitment rates (rate of births + immigration) and survival
probabilities (probability to stay or not die) over five years (2015-2019) for the same species (except Eurasian tree sparrow) were influenced by meadow habitat, indicating low-intensity farming. I was able to confirm this hypothesis, too, but only for yellowhammer.
Yellowhammer was the only species with covariate models showing a significant difference (ΔAIC < 2) from the null model using the AIC. A high meadow proportion showed a positive influence on recruitment, while a high reed proportion showed a positive influence on
survival. The latter is not a habitat type that is directly connected to low-intensity farmland, which was the main interest of this study, but a reason for its positive influence on survival could be that reeds can serve as a good hiding place to escape predators, and/or it is simply correlated with freshwater sources which are undoubtedly crucial for survival. For the rest of the species the results could be interpreted as meadow is 1) truly not an important habitat type, 2) important only when interacting with other habitat types, 3) important but the five year period is too short to see the effects, or 4) important over five years but the amount of data for each year was insufficient to show it.
Eurasian skylark had the lowest recruitment rate of 0.15 while yellowhammer had the highest rate of 1.63. Interestingly, the analysis gave a survival probability of 1.00 for the latter. This would mean that all individuals during the five years of the study stayed in the population.
The number seems unlikely, but at least it can be interpreted as an indication that survival
probability is very high for yellowhammers in the area and that they do not tend to move
18
elsewhere. In practice, this result is likely caused by very even recordings across years, in contrast to the other species.
Red-backed shrike showed the lowest survival probability estimate, which was surprising considering that they had the largest mean abundance per survey site (6.53) and population size in the first season (2790), and the second highest recruitment rate (1.15). The overall impression on surveys were also that counts of the species were very high in all five years.
The numerous detections of the species could be a result of more fledged chicks that also stay together with their parents for some time. Putting these together, it seems that the red-backed shrike population in the area experiences either a high turnover rate or a large influx and outflux.
The advantage of using recruitment rates and survival probabilities instead of abundance at sites is that the first two can give a better picture of habitat quality and preferences
(MacKenzie et al., 2018 p. 20). However, the mean abundance per sample unit is still a useful parameter to get an estimate of the population - especially because it takes detectability into account (Royle, 2004). These population parameter estimates can then be useful when planning conservation actions (Joseph et al., 2009).
Because the villages were surveyed in slightly varying order from year to year, survey day was used as a detectability covariate representing time. The time in the season is important for detection of birds since they are most active in terms of territoriality and singing during the mating season in spring and beginning of summer, making them easier to detect by song at that time (Kéry et al., 2005). Since the survey continued until late summer (Aug) it is also highly probable that juveniles were counted towards the end – giving on the other hand an increase of observations with time. This could be highly dependent upon the phenology of the respective species. As expected, all species - except skylark - showed that “day” was indeed the most important covariate for detectability. The red-backed shrike was the only species that showed a positive detection trend as the season progressed. An explanation for this could be that they breed early and, as mentioned, stay together as a family for some time, leading to an increase in number of individuals recorded as the chicks grow out of the nest. The detection of yellowhammer and whitethroat showed the opposite trend, which is expected if most
recordings are by song, as they mostly sing early in the season. Detectability of skylark was negatively influenced by cloud cover. This means that individuals were mostly seen flying and singing in sunny weather, which they might prefer.
When choosing covariates there is always a trade-off between bias and variance, meaning that one has to balance between including too few covariates (underfitting) and too many
(overfitting) (Burnham & Anderson, 2002 p. 58-59). Furthermore, it is important to note that the model(s) giving highest parsimony is only as good as you allow by your choice of
variables to include. In reality, there might exist other variables which are more important than those tested (Burnham & Anderson, 2002 p. 89).
4.3 Design issues
A limitation of the survey design is that there were two different observers in 2019, and a total
of five different observers across all seasons from 2015 to 2019. They had varying years of
experience and one had never been in the country before leading the survey. However, for the
models I tested here, “observer” was not top ranked as a detectability covariate for any of the
19
species. This means that the observer was of minor importance for the final results, because their level of knowledge was not significantly different.
In addition, the number of people who participated in the survey each day also varied, as high-school groups joined some of the days to learn about conservation. During these days, the noise might have disturbed both the birds and the survey leaders, causing a heavier underestimation of bird presence than that of the “human factor” which is inevitable in these types of surveys. However, I couldn’t evaluate this effect because the group size was not recorded.
Another important fact to bear in mind is that the species can have different social behaviours, leading to a heavier underestimation of individual numbers. For example, seeing only a few starlings probably means that there are more nearby, considering their social behaviour, but this cannot be recorded without any more sightings or songs. However, most often they were in fact observed in large groups.
The survey points were located about 500 m apart on the transects. This is not ideal
considering that most birds have home ranges reaching further than this, which might cause the surveyers to record the same birds at multiple points (Jiménez-Franco et al., 2019). So, the question is if the points really can be viewed as spatially independent. Furthermore, the
habitat at which an individual is in on a certain point on the transect might not be a 100%
representative of the habitat of its whole home range. But since the point count needs to be done before noon to catch the peak of bird activity, and in addition is done on foot, it would take too long.
For better reliability of the results from the analysis, repeated surveys should preferably be done more than twice in one season (MacKenzie et al., 2002), which was not the case in this study. As put by the authors: “increasing the number of visits per site improves the precision /…/ and the resulting increase in information improves the accuracy of the estimate when detection probabilities are low”. Therefore, they recommend that in the case of less than three visits the estimated detectability needs to be at least 0.3 (MacKenzie et al., 2002). Luckily, all detection probabilities were estimated to around 0.50 (min = 0.44, max = 0.51), which is well above 0.3. But when facing the trade-off between surveying more units with only a few visits, or fewer units but with more visits, the latter is a better allocation of resources (MacKenzie et al., 2017 p. 37).
4.4 Future development
To develop the study design, the number of visits in one season could be increased to get a higher reliability of analysis results. It should also preferably be led by the same person(s), so that “observer” as a variable would not have to be considered. In addition, some of the points on the transects could be removed and instead having a larger distance between those that are left, to lower the risk of counting the same individuals multiple times. The number of points and length of the routes could be made more alike, or as an alternative, effort could be used as a detectability covariate. The effort would then be the time divided by distance, following Kéry and colleagues (2005). Temperature would also be an option (MacKenzie et al., 2002).
A potential constant covariate to include, which was not recorded in this study but seems to be
common in these types of studies, is elevation (e.g. Kéry et al., 2005). However, it might not
be worth recording considering that the sites are close to each other and so do not vary much.
20
For further analysis of this particular data set, I would firstly like to run parametric bootstraps on the most parsimonious models to be able to assess goodness of fit, since parsimony (AIC) does not say much about how well the expected probability distribution obtained from the model matches the observed distribution. In other words: exactly how well does the model explain what we see?
I would also like to model more than these five species and increase the number of models per species. It would enable analysing interaction of habitat types within the categories
“recruitment rate” and “survival probability”. In this study, I built 12 models compared to for example Kéry and colleagues (2005) who had a corresponding number of 120. One could also try out multiple species modelling in both single and multiple seasons, where the latter allows the analysis of guilds rather than specific species. This is done by comparing the local
extinction and colonization rates to see if a group of species show similar patterns (MacKenzie et al., 2018 p. 56).
4.5 Conclusion
Most species in my study are still in IUCN Red List categories LC (all of the five modelled species). However, many of them are in decline in other parts of Europe (Donald et al., 2001;
Andersen et al., 2003; Donald et al., 2006), which with time might change their status towards being endangered. An extensive study of the dynamics of avian extinction modelled with the Red List categories of all recognised bird species from 1988 to 2016 (Monroe et al., 2019) forecasted that, within the next 500 years, about 470 species will be lost. About 110 of these are species which are today classified as LC. Thus, there is a massive accumulation of extinction risk even from species which are not vulnerable today. The study emphasizes that human conservation efforts have considerably lowered the extinction risk, but the efforts have mainly focused on species that are already on the brink of extinction (Monroe et al., 2019).
The authors conclude that while this is of course a necessary approach, one should also focus on preventing species which are not yet in the most vulnerable Red List categories from ending up there because “it would help dampen an ever-building wave of conservation need”
(Monroe et al., 2019).
When it comes to the farmland bird populations in Târnava Mare, they will hopefully continue to thrive with an improving EU legislation and subsidies given to the farmers to improve their land management. The landscape structure in the area itself might also be an advantage in the sense that the Carpathian Mountains and surrounding hills do not allow for such large-scale intensification as seen in other parts of Europe.
Acknowledgements
I would like to thank my University supervisor Professor Anssi Laurila for supporting me in my work as well as giving useful feedback on this report. Fellow student Meagan Tunon and postdoc William Jones have also helped me with good inputs.
For the field work, I would like to thank my supervisor on the Opwall site, Dr. Bruce Carlisle
(field supervisor) and Toby Farman (site manager), for guidance and for letting me take part
in their important conservation project and data. Paul Leafe, whom I spent most of my early
field mornings with and acquired all of my new bird knowledge from, has been invaluable!
21
For the analysis part, I am very thankful for the people who voluntarily helped me with the modelling code in R. These were mostly people on the “unmarked” Google forum.
References
Akeroyd J. & Bădărău S. (2012). The High Nature Value dry grasslands of southern Transylvania. Fundația ADEPT Transilvania
Alves F., López‐Iborra G.M. Stojanovic D., Webb M.H., Langmore N., Heinsohn R. (2019).
Occupancy and density of a habitat specialist and a sympatric generalist songbird species in Tasmania, Austral Ecology. DOI: 10.1111/aec.12817
Andersen E., Baldock D., Bennett H., Beaufoy G., Bignal E., Brouwer F., Elbersen B., Eiden G., Godeschalk F., Jones G., McCracken D., Nieuwenhuizen W., van Eupen M., Hennekens S., Zervas G. (2003). Developing a high nature value farming area indicator. Final Report.
EEA, Copenhagen.
Avibase (2018). Part of BirdLife International https://avibase.bsc-
eoc.org/checklist.jsp?region=RO&list=howardmoore®ion=RO&list=howardmoore Accessed: 12 Feb 2019
BirdLife Sweden, Lantmännen, Swedish University of Agriculture & WWF (2018).
Lantbrukare för lärkor - Unikt samarbete ska vända trenden för en hotad art.
Burnham K.P., Anderson D.R. (2002). Model Selection and Multimodel Inference – A Practical Information-Theoretic Approach, 3rd Edition. p 58-59, 89 (7-99 in pdf-reader) Springer-Verlag New York, Inc. ISBN: 0-387-95364-7
Dail D. & Madsen L. (2011). Models for Estimating Abundance from Repeated Counts of an Open Metapopulation, Biometrics. 67: p. 577–587. DOI: 10.1111/j.1541-0420.2010.01465.x
Demongin L. (2016). Identification Guide to Birds in the Hand. ISBN: 978-2-9555019-0-0 Donald P.F., Green R.E., Heath M.F. (2001). Agricultural intensification and the collapse of Europe's farmland bird populations, Proceedings of the Royal Society B-Biological Sciences.
268:1462 p. 25–29. DOI: 10.1098/rspb.2000.1325
Donald P.F., Pisano G., Rayment M.D., Pain D.J. (2002). The Common Agricultural Policy, EU enlargement and the conservation of Europe’s farmland birds. Agriculture, Ecosystems &
Environment. 89: p. 167–182. Doi: https://doi.org/10.1016/S0167-8809(01)00244-4 Donald P.F., Sanderson F.J., Burfield I.J., van Bommel F.P.J. (2006). Further evidence of continent-wide impacts of agricultural intensification on European farmland birds, 1990- 2000. Agriculture, Ecosystems and Environment. 116: p. 189–196. DOI:
10.1016/j.agee.2006.02.007
Doxa A., Bas Y., Paracchini M.L., Pointereau P., Terres J.M., Jiguet F. (2010). Low-intensity agriculture increases farmland bird abundances in France, Journal of Applied Ecology. 47:6 p. 1348-1356. DOI: 10.1111/j.1365-2664.2010.01869.x
European Environmental Agency (2004). High nature value farmland - Characteristics,
trends and policy challenges. Office for Official Publications of the European Communities,
Luxembourg. ISBN: 92-9167-664-0
22
European Environmental Agency (2017). Landscapes in transition - An account of 25 years of land cover change in Europe. Publications Office of the European Union, Luxembourg. DOI:
10.2800/81075
Fiske I. & Chandler R. (2012). Overview of Unmarked: An R Package for the Analysis of Data from Unmarked Animals, CRAN R Projects. https://cran.r-
project.org/web/packages/unmarked/vignettes/unmarked.pdf Accessed: 12 Jan 2020
Gregory R.D., Noble D., Field R., Marchant J., Raven M., Gibbons D.W. (2003). Using birds as indicators of biodiversity, Ornis Hungarica 12-13: p. 11–24.
Gregory R.D., van Strien A., Vorisek P., Gmelig Meyling A.W., Noble D.G., Foppen R.P.B, Gibbons D.W. (2005). Developing Indicators for European Birds, Philosophical Transactions of the Royal Society B: Biological Sciences. 360(1454): p. 269–88. DOI:
10.1098/rstb.2004.1602
Greshko M. & Chung D. (2018). Industrial Farming a Cause of Plummeting Bird Populations, National Geographic Magazine. Sept 2018 issue.
https://www.nationalgeographic.com/magazine/2018/09/embark-data-sheet-farm-bird- population-decline-europe-infographic/ Accessed: 12 Feb 2019
Jiménez‐Franco M.V., Kéry M., León‐Ortega M., Robledano F., Esteve M.A., Calvo J.F.
(2019). Use of classical bird census transects as spatial replicates for hierarchical modelling of an avian community, Ecology and Evolution. 9: p. 825–835. DOI:
https://doi.org/10.1002/ece3.4829
Joseph L.N., Elkin C., Martin T.G., Possingham H.P. (2009). Modeling abundance using N- mixture models: the importance of considering ecological mechanisms, Ecological
Applications. Ecological Society of America 19:3 p. 631-642. DOI:
https://doi.org/10.1890/07-2107.1
Kéry M. & Royle A.J. (2015). Applied hierarchical modelling in ecology, Elsevier. Academic Press Inc. ISBN: 9780128013786
Kéry M., Royle A.J., Schmid H. (2005). Modeling avian abundance from replicated counts using binomial mixture models, Ecological Applications. 15:4 p. 1450–1461. DOI:
https://doi.org/10.1890/04-1120
MacKenzie D.I., Nichols J.D., Lachman G.B., Droege S., Andrew Royle J., Langtimm, C.A.
(2002). Estimating site occupancy rates when detection probabilities are less than one, Ecology. 83:8 p. 2248-2255.
MacKenzie D.I., Nichols J.D., Royle J.A., Pollock K.H., Bailey L., Hines J. E. (2018).
Occupancy Estimation and Modeling: inferring patterns and dynamics of species occurrence, Elsevier. p. 3, 27, 37. Academic Press Inc. ISBN: 9780128146910
Maxwell S.L., Fuller R.A., Brooks T.M., Watson J.E.M. (2016). Biodiversity: the ravages of guns, nets and bulldozers, Nature. 536: p. 143-145. DOI: 10.1038/536143a
Monroe M.J., Butchart S.H.M., Mooers A.O., Bokma F. (2019). The dynamics underlying avian extinction trajectories forecast a wave of extinctions, Biol. Lett. 15:20190633. DOI:
http://dx.doi.org/10.1098/rsbl.2019.0633
23
R Core Team (2018). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/
Reif J., Vermouzek Z. (2019). Collapse of farmland bird populations in an Eastern European country following its EU accession. Conservation Letters. 2019;12:e12585. DOI:
https://doi.org/10.1111/conl.12585
Ripple W.J., Wolf C., Newsome T.M., Galetti M., Alamgir M., Crist E., Mahmoud M.I., Laurance W.F. (2017). World Scientists’ Warning to Humanity: A Second Notice, Bioscience.
67:12 p. 1026-1028. DOI: 10.1093/biosci/bix125
Royle J.A. (2004). N-Mixture Models for Estimating Population Size from Spatially Replicated Counts, Biometrics. 60: p. 108-115. DOI: https://doi.org/10.1111/j.0006- 341X.2004.00142.x
Ruiz-Martinez I., Marraccini E., Debolini M., Bonari E. (2015). Indicators of agricultural intensity and intensification: a review of the literature, Italian Journal of Agronomy. 10:656 p. 74-84. DOI: 10.4081/ija.2015.656
Scholefield P., Firbank L., Butler S., Norris K., Jones L.M., Petit S. (2011). Modelling the European Farmland Bird Indicator in Response to Forecast Land-Use Change in Europe, Ecological Indicators. 11:1 p. 46–51. DOI: https://doi.org/10.1016/j.ecolind.2009.09.008.
SOER (2015). The European environment - State and outlook 2015.
Svenning P. B. (2007). Management Plan for Skylark (Alauda arvensis) 2007-2009, European Communities.
Svensk Fågeltaxering (2018). Övervakning av fåglarnas populationsutveckling - Årsrapport för 2018. Lund Univeristy
Svensson L., Mullarney K., Zetterström D., Grant P.J. (2011). Collins Bird Guide 2
ndEdition.
Collins. ISBN: 9780007449026
Tilman D., May R.M., Lehman C.L., Nowak M.A. (1994). Habitat destruction and the extinction debt, Nature. 371: p. 65-66. DOI: https://doi.org/10.1038/371065a0
Union of Concerned Scientists (1992). World Scientists’ Warning to Humanity
World Weather & Climate Information (2019). Sighișoara, Romania. https://weather-and-
climate.com/sighisoara-July-averages Accessed: 24 Feb 2019
24
Appendix I
Figure 8. Correlation matrix for habitat variables 2019. Red star indicates correlation. The more stars, the stronger the correlation.
Figure 9. Correlation matrix for habitat variables 2015-2019. Red star indicates correlation. The more stars, the stronger the correlation.
25
Figure 10. Correlation matrix for detectability variables 2015-2019. Red star indicates correlation. The more stars, the stronger the correlation.
26
Appendix II
Figure 11. Histogram of the distribution of point count data for red-backed shrike. Data were collected between 2015 and 2019.
Figure 12. Histogram of the distribution of point count data for yellowhammer. Data were collected between 2015 and 2019.
27
Figure 13. Histogram of the distribution of point count data for Eurasian skylark. Data were collected between 2015 and 2019.
Figure 14. Histogram of the distribution of point count data for Eurasian tree sparrow. Data were collected between 2015 and 2019.
28
Figure 15. Histogram of the distribution of point count data for common whitethroat. Data were collected between 2015 and 2019.
29
Appendix III
Figure 16. Bird point count frequencies of the focal species in 2015 years’ survey. Starling and rook are excluded here due to high frequencies. See separate graphs. FBI = species that are listed in the Farmland Bird Indicator.
Figure 17. Bird point count frequencies of the focal species in 2016 years’ survey. Starling and rook are excluded here due to high frequencies. See separate graphs. FBI = species that are listed in the Farmland Bird Indicator.
30
Figure 18. Bird point count frequencies of the focal species in 2017 years’ survey. Starling and rook are excluded here due to high frequencies. See separate graphs. FBI = species that are listed in the Farmland Bird Indicator.
Figure 19. Bird point count frequencies of the focal species in 2018 years’ survey. Starling and rook are excluded here due to high frequencies. See separate graphs. FBI = species that are listed in the Farmland Bird Indicator.
31
Figure 20. Bird point count frequencies of the focal species in 2019 years’ survey. Starling and rook are excluded here due to high frequencies. See separate graphs. FBI = species that are listed in the Farmland Bird Indicator.
Figure 21. Bird point count frequencies of starling (left) and rook (right) by year, between 2015-2019. Note difference in y-axis scale.
32
Appendix IV
Table 7. Bird species recorded in the bird point count (BPC) surveys in Târnava Mare during the field season of 2019. Species included in the Farmland Bird Indicator (FBI) (Scholefield, 2011) and/or listed as farmland specialists (Gregory et al., 2005) are marked.
Species (scientific) Species Freq.
2019
FBI Specialist
