Distribution, habitat, and conservation status of the near-threatened Japanese Paradise-Flycatcher (Terpsiphone atrocaudata periophthalmica) on Lanyu, Taiwan

Abstract

The near-threatened Japanese Paradise-Flycatcher (Terpsiphone atrocaudata) consists of three subspecies, one of which, T. a. periophthalmica, has an important population on Lanyu, Taiwan. After briefly reviewing the species’ conservation status in its breeding range, we describe our field work in Lanyu during the breeding seasons of 2009 and 2010. We first established that the territory size of a male flycatcher is around 1.16 hectares. We then visited 224 1-hectare grid cells randomly distributed across Lanyu and established 120 presence grid cells. We then used these presence grid cells and nine environmental data layers to build an ensemble distribution model using the software Maxent. The model showed that the Japanese Paradise-Flycatcher prefers relatively wet forest habitats at elevations of 50–300 m. Using the model, we estimated that the extent of suitable habitat covered approximately 12.0 km² (or 26%) of Lanyu’s surface area, which could hold approximately 1000 male territories. Forest cover increased between 1948 and 2006 by approximately 7.6 km² (or 16%) of Lanyu’s surface area, which, all other things being equal, should have resulted in a population increase of around 30%. Given the absence of current threats, the Lanyu population is assumed to be relatively stable. Given this new information and our review of the species’ conservation status, we suggest that the species may be down-listed to “least concern.” However, if the distinct subspecies T. a. periophthalmica would be elevated to species status or be considered two independent conservation units (one in Lanyu and one in Batanes, Philippines), its conservation status would be much more precarious given it only occurs in five known localities (Lanyu and four islands in Batanes) of limited geographic range, and a population size of approximately 1000 males in Lanyu and an unknown population size in Batanes. Therefore, we conclude that more information is needed about (1) the species’ status in Batanes, (2) its migration and wintering grounds, and (3) the taxonomic status of the three subspecies.

Full text

VOLUME 13, ISSUE 1, ARTICLE 7
Späth, T., M.-L. Bai, L. L. Severinghaus and B. A. Walther 2018. Distribution, habitat, and conservation status of the near-threatened Japanese
Paradise-Flycatcher (Terpsiphone atrocaudata periophthalmica) on Lanyu, Taiwan. Avian Conservation and Ecology 13(1):7. https://doi.org/10.5751/
ACE-01167-130107
Copyright © 2018 by the author(s). Published here under license by the Resilience Alliance.
Research Paper
Distribution, habitat, and conservation status of the near-threatened
Japanese Paradise-Flycatcher (Terpsiphone atrocaudata
periophthalmica) on Lanyu, Taiwan
Thorsten Späth 1, Mei-Ling Bai 2, Lucia L. Severinghaus 3 and Bruno Andreas Walther 4
1Georg-August-University-Göttingen, Conservation Biology/Workgroup on Endangered Species, Göttingen, Germany, 2Formosa
Natural History Information Ltd., Taipei, Taiwan, 3Biodiversity Research Center, Academia Sinica, Taipei, Taiwan, 4Master
Program in Global Health and Development, College of Public Health, Taipei Medical University, Taipei, Taiwan
ABSTRACT. The near-threatened Japanese Paradise-Flycatcher (Terpsiphone atrocaudata) consists of three subspecies, one of which,
T. a. periophthalmica, has an important population on Lanyu, Taiwan. After briefly reviewing the species’ conservation status in its
breeding range, we describe our field work in Lanyu during the breeding seasons of 2009 and 2010. We first established that the territory
size of a male flycatcher is around 1.16 hectares. We then visited 224 1-hectare grid cells randomly distributed across Lanyu and
established 120 presence grid cells. We then used these presence grid cells and nine environmental data layers to build an ensemble
distribution model using the software Maxent. The model showed that the Japanese Paradise-Flycatcher prefers relatively wet forest
habitats at elevations of 50–300 m. Using the model, we estimated that the extent of suitable habitat covered approximately 12.0 km²
(or 26%) of Lanyu’s surface area, which could hold approximately 1000 male territories. Forest cover increased between 1948 and 2006
by approximately 7.6 km² (or 16%) of Lanyu’s surface area, which, all other things being equal, should have resulted in a population
increase of around 30%. Given the absence of current threats, the Lanyu population is assumed to be relatively stable. Given this new
information and our review of the species’ conservation status, we suggest that the species may be down-listed to “least concern.”
However, if the distinct subspecies T. a. periophthalmica would be elevated to species status or be considered two independent
conservation units (one in Lanyu and one in Batanes, Philippines), its conservation status would be much more precarious given it only
occurs in five known localities (Lanyu and four islands in Batanes) of limited geographic range, and a population size of approximately
1000 males in Lanyu and an unknown population size in Batanes. Therefore, we conclude that more information is needed about (1)
the species’ status in Batanes, (2) its migration and wintering grounds, and (3) the taxonomic status of the three subspecies.
Répartition, habitat et statut de conservation du Tchitrec du Japon (Terpsiphone atrocaudata
periophthalmica), quasi menacé, sur l'île de Lanyu, Taïwan
RÉSUMÉ. Le Tchitrec du Japon (Terpsiphone atrocaudata) quasi menacé, compte trois sous-espèces dont l'une, T. a. periophthalmica,
est bien établie sur l'île de Lanyu, Taïwan. Après une brève revue du statut de conservation de l'espèce dans son aire de nidification,
nous décrivons les travaux réalisés sur Lanyu durant les saisons de reproduction de 2009 et 2010. Dans un premier temps, nous avons
déterminé que la taille du territoire d'un tchitrec mâle est de 1,16 ha environ. Nous avons ensuite visité 224 unités de grille de 1 ha
réparties aléatoirement sur Lanyu, et avons observé que l'espèce était présente dans 120 d'entre elles. Nous avons utilisé les unités de
grille avec présence et neuf couches de données relatives à l'environnement pour bâtir un modèle de répartition d'ensemble au moyen
du logiciel Maxent. Le modèle a révélé que le Tchitrec du Japon se rencontre plus souvent dans les milieux forestiers relativement
humides, à une élévation allant de 50 à 300 m. À partir du modèle, nous avons estimé que les milieux propices couvraient
approximativement 12,0 km² (ou 26 %) de la superficie de Lanyu, étendue qui pourrait héberger à peu près 1000 mâles territoriaux. Le
couvert forestier s'est accru entre 1948 et 2006, d'environ 7,6 km² (ou 16 %) de la superficie de Lanyu, ce qui, toutes choses égales
d'ailleurs, laisse entendre que la population pourrait avoir augmenté de l'ordre de 30 %. Puisqu'il n'existe pas de menaces actuellement,
nous supposons que la population sur Lanyu est relativement stable. À la lumière de cette nouvelle information et de notre revue du
statut de conservation de l'espèce, nous proposons que son statut soit abaissé à celui de « préoccupation mineure ». Toutefois, si la sous-
espèce T. a. periophthalmica était élevée au rang d'espèce ou si l'on considérait qu'il existe deux unités de conservation indépendantes
(l'une sur Lanyu et l'autre dans la province de Batanes, Philippines), son statut de conservation serait beaucoup plus précaire puisque
cette sous-espèce n'est connue pour fréquenter que cinq endroits (Lanyu et quatre îles en Batanes) d'étendue géographique limitée, et
que la population n'est que de 1000 mâles environ sur Lanyu et que sa taille est inconnue en Batanes. Nous concluons que plus de
recherche doit être faite sur : 1) le statut de l'espèce en Batanes; 2) ses migrations et ses aires d'hivernage; et 3) le niveau taxinomique
des trois sous-espèces.
Key Words: breeding population; conservation status; Japanese Paradise-Flycatcher; Maxent; Terpsiphone atrocaudata
Address of Correspondent: Bruno Andreas Walther, Master Program in Global Health and Development, College of Public Health, Taipei Medical
University, Taipei, Taiwan, bawalther2009@gmail.com

Avian Conservation and Ecology 13(1): 7
http://www.ace-eco.org/vol13/iss1/art7/
INTRODUCTION
Island bird species and populations are especially threatened by
extinction (Johnson and Stattersfield 1990), and habitat
destruction and invasive species are the two main causes (Veitch
and Clout 2002, Reaser et al. 2007). Therefore, there is a growing
focus on the value of conserving island populations as
independent conservation units (Kier et al. 2009, Pruett et al.
2017). Because conservation and management of species is
dependent on knowledge of a species’ distribution, ecology, and
population size (Sutherland et al. 2004), an understanding of the
main factors determining a species’ presence will assist
conservation managers in prioritizing actions and increasing the
efficacy of any applied strategy (Sutherland et al. 2004). The East
Asian region has a large number of island bird species that remain
little known and understudied (Collar et al. 2001, Brazil 2009,
Ando et al. 2014).
One such species is the near-threatened Japanese Paradise-
Flycatcher (Terpsiphone atrocaudata), also known as Black
Paradise-Flycatcher. This species has been divided into three
subspecies, which breed in mature evergreen broadleaf, deciduous,
or mixed forests, and sometimes in plantations up to 1000 m
elevation (Kennedy et al. 2000, Coates et al. 2006, Brazil 2009,
Severinghaus et al. 2010, 2017, Jeyarajasingam and Pearson 2012,
BirdLife International 2015). The three subspecies are distributed
as follows: (1) T. a. atrocaudata in Japan and Korea (according
to Ding et al. 2014 and R.-S. Lin, 2016, personal communication,
this subspecies does not breed in Taiwan, contra Coates et al.
2006); (2) T. a. illex in Ryukyu Islands (or Nansei Shoto), Japan;
(3) T. a. periophthalmica in Lanyu (or Orchid Island), Taiwan,
and the Batanes Islands, which includes Batan Island, northern
Philippines (Fig. 1; Nuytemans 1998, Kennedy et al. 2000, Coates
et al. 2006, Duckworth and Moores 2008, Brazil 2009,
Severinghaus et al. 2010, 2017, Jeyarajasingam and Pearson 2012,
BirdLife International 2015; Batanes Islands biodiversity survey,
unpublished manuscript). T. a. periophthalmica is very distinct from
the other two subspecies and has been considered a full species
by some authors (e.g., McGregor 1907, Alcasid 1965). T. a.
atrocaudata is fully migratory, T. a. illex is partially migratory (or
perhaps resident), and T. a. periophthalmica in Lanyu is mostly
migratory (Nuytemans 1998, Kennedy et al. 2000, Coates et al.
2006, Gonzalez et al. 2008, Brazil 2009, Severinghaus et al. 2010,
2017, BirdLife International 2015; J. C. T. Gonzalez, 2016,
personal communication; G.-Q. Wang, personal communication;
Appendix 1). Whether the Batanes population of T. a.
periophthalmica is partially or completely migratory or even
completely resident remains an open question (J. C. T. Gonzalez,
2016, personal communication). The migratory populations winter
in forests and mangroves up to 700 m elevation in the northern
Philippines, southern Thailand, peninsular Malaysia, Singapore,
and Sumatra (Nuytemans 1998, Kennedy et al. 2000, Coates et
al. 2006, Oliveros et al. 2008, Brazil 2009, Jeyarajasingam and
Pearson 2012, BirdLife International 2015).
The near-threatened conservation status of the Japanese
Paradise-Flycatcher appears to be based mostly on older Japanese
studies (Appendix 1) and an assumed habitat loss on the wintering
grounds (BirdLife International 2015). However, BirdLife
International (2015) proposed that more studies and careful
monitoring are needed in order to (1) determine its current
distribution and abundance on both the breeding and wintering
range and (2) increase knowledge of its habitat requirements.
Fig. 1. Distributions of the three subspecies of the Japanese
Paradise-Flycatcher (Terpsiphone atrocaudata): (green) T. a.
atrocaudata in Japan, South and North Korea; (blue) T. a. illex
in Ryukyu Islands, Japan; (dark red) T. a. periophthalmica in
Lanyu Island, Taiwan, and Batanes Islands, Philippines.
To further improve our knowledge of this species, we (1) assessed
the species’ current conservation status using the available
literature, (2) used the species’ ecological preferences to determine
the geographic distribution on Lanyu, (3) estimated Lanyu’s
population size, and (4) estimated the habitat change on Lanyu
over the last half century.
METHODS
Conservation status assessment
To assess the current conservation status of the Japanese Paradise-
Flycatcher, we gathered information from all recent publications
and reports about the species’ status in Japan, Korea, Taiwan,
and the Philippines. Publications were found by (1) searching
Google, Google Scholar, and Web of Science using the
appropriate search terms (e.g., English and Latin species name,
country names, etc.); (2) checking standard references (e.g., Brazil
1991, Brazil 2009, Coates et al. 2006, Severinghaus et al. 2010,
2017) and websites (e.g., BirdLife International 2015, HBW Alive
2016) for information and further references; (3) emailing any
researcher whose email was given in these references and asking
for further information and publications (any researcher who
responded is mentioned in the Acknowledgements).
Study site
Lanyu (21°58′-22°06′N, 121°29′-121°35′E), Taitung County,
Taiwan, is located 64 km east off the southern tip of Taiwan. It
is a 46.3 km² mountainous volcanic island of mainly wooded
habitats, generally shallow soil, and mostly steep topography. The
tallest mountain is Mt. Hongtoushan at 552 m a. s. l., and nearly
10 peaks are > 400 m (Fig. A1.1). The island is bounded by narrow
strips of coastal flatlands (Fig. A1.4), and seven villages are
located within these strips. The mean annual temperature is 22.7º
C, and the mean annual rainfall is approximately 3000 mm with
an average of about 220 rainy days (Taiwan Central Weather
Bureau). Because of year-round heavy winds and frequent
typhoons (usually several per year), the slopes are largely covered

Avian Conservation and Ecology 13(1): 7
http://www.ace-eco.org/vol13/iss1/art7/
by dense shrubs and stunted trees while the occurrence of taller
forests is confined to wind-sheltered plains and valleys (Chao et
al. 2010). Although the most protected forests can reach a height
of 20 meters, the forest canopy on slopes is usually around eight
meters and, depending on wind exposure, even lower. Much of
the taller forests of the coastal flatlands have been transformed
into agricultural fields or frequently burned grasslands, while
other accessible forests are structurally altered for the purpose of
fruit production or construction wood (T. S., personal
observation).
Territory size estimation
To estimate the territory size of male Japanese Paradise-
Flycatchers, we selected five study plots that represented the
various forest types found on the island and that were sufficiently
accessible for frequent and intensive observations (some steep
slopes are virtually inaccessible). From early March to late
September 2009 and 2010, we censused each plot for the
flycatcher’s presence using conspecific playback at control points
if we had not made spontaneous observations within the first few
minutes after reaching the plot. The majority of these territories
were occupied from late March, and no individuals left the island
before September. Our observations also indicated that territorial
borders became less distinct and that vocal activity or response
to playback decreased as the breeding season progressed. We
therefore chose April and May in 2010 as the period of highest
territorial and vocal activity during which to carry out our island-
wide census.
In 2009, we also captured flycatchers using mist nets at the same
five study plots and color-banded each individual. We could
furthermore distinguish uncaptured males in these study plots by
differences in body characteristics, such as the length of tail
feathers and aspects of plumage coloration on the throat, breast
belly, mantle, greater coverts, and tail feathers. We then mapped
out individual territories within each study plot between late
March and late May. Once a week, each study plot was searched
in the morning; points of encounters were recorded on a map
based on high-resolution satellite images, and the bird was
followed until we lost sight of it. These points were then used to
plot the trajectory of the bird. The outermost points were later
connected as a minimum convex polygon using Hawth’s Tools for
ArcGIS. The mean and standard deviation of 32 mapped
territories was 1.16 ± 0.43 hectares (some mentioned in
Severinghaus and Bai 2009, unpublished data cited in Appendix 1).
Environmental data layers
We assembled 22 environmental layers with a resolution of 100
m (or 0.1 x 0.1 km or 1 hectare grid cell) for Lanyu (Table A1.1).
The two normalized differenced vegetation index (NDVI) layers
were generated from a satellite image taken by FORMOSAT-2 in
April 2007 (Center for Space and Remote Sensing Research;
http://www.csrsr.ncu.edu.tw/) with an 8 m multispectral
resolution. We calculated the NDVI for each 8 x 8 m pixel and
summarized the mean and standard variation of the NDVI values
in each 1 hectare grid cell (Fig. A1.2).
A digital terrain model (DTM) with 40 m resolution generated
by the Aerial Survey Office (http://www.afasi.gov.tw/) was used
to calculate four topographical data layers, namely elevation,
aspect, slope, and solar irradiation for each 1 hectare grid cell.
These four data layers were then used to generate the data layers
3-13 (Table A1.1).
Based on the manual interpretation of aerial photographs taken
in 1948 by the U.S. Airforce and in 2006 taken by the Aerial Survey
Office, we created two layers of land cover types using ArcGIS
that included the following categories: (1) roads; (2) bare coastal
land (beach, rock); (3) other built-over land (buildings, parking
spaces, etc.); (4) farmland or grassland; (5) hilltop shrub; (6)
stunted forest; (7) tall forest (either mixed mature forests or forests
dominated by mature Pometia pinnata; Fig. A1.3), (8); other types,
e.g., landslides, inland water, etc. Stunted forest contained both
low secondary forests and primary shrubby woodland on wind
exposed slopes because these two types could not be differentiated
in the aerial photographs. These two data layers were used to
generate the data layers 14-22 (Table A1.1). To characterize the
land cover types of each grid cell, we calculated the percentage
value of each type for each cell (data layers 14-21 in Table A1.1),
and finally used the same data to generate a categorical data layer,
which corresponded to the dominant land cover type of each grid
cell (data layer 22 in Table A1.1). Only this data layer 22 of the
dominant land cover types was generated for both 1948 and 2006
in order to assess habitat change.
Modeling species distribution
Because we had established that male territories were about one
hectare, we used ArcGIS to establish 1 hectare (100 m x 100 m)
grid cells as sampling units across the entire island. Any grid cells
that were not entirely covered by land surface were a priori
excluded. We then used Hawth’s Tools for ArcGIS to randomly
select 224 grid cells among the remaining grid cells, which
represent about 5% of Lanyu’s total land surface. Because of our
limited resources and the island’s rugged terrain that renders many
parts inaccessible, we a priori chose this 5% threshold, which was
the maximum number of grid cells that we could reasonably
survey during the period of the flycatcher’s highest territorial
activity, i.e., April and May.
Each grid cell was visited once between sunrise and noon during
April and May 2010. Once the center of the grid cell had been
reached, we waited for spontaneous calls or songs for up to three
minutes; if none were detected, we used playback consisting of
recordings of calls and songs of the local population and played
them up to three times at three minute intervals to elicit a response.
The species was defined to be present if it was detected visually
or acoustically during this time, and was defined to be absent from
the respective grid cell if it could not be heard or seen during this
one visit.
This protocol for detecting presence was based on our preliminary
field work. During our regular controls of the five study plots but
also when checking additional transect lines, we used playback at
fixed control points to check for the presence or absence of the
territorial male if it could not be detected by visual or acoustic
detection. We found that there was no significant effect of time
of day on the playback response, and that we never failed to detect
the territorial male even if the playback was conducted only in
the grid cell’s center. Therefore, we are confident that our method
of detection was reliable.
However, we are aware that absence is much harder to establish
than presence, and that some of the absence records in Fig. 2 may

Avian Conservation and Ecology 13(1): 7
http://www.ace-eco.org/vol13/iss1/art7/
be false absences. For this reason, we chose a modeling software
that only needs presence records, namely Maxent. Therefore, our
absence records were not used further in the modeling analyses
described below.
Fig. 2. Location of presence (red, n = 120) and absence (blue, n
= 104) grid cells for the Japanese Paradise-Flycatcher
(Terpsiphone atrocaudata) across Lanyu, Taiwan, in 2010.
We avoided multicollinearity and overfitting in our Maxent
distribution models by reducing our original set of 21 continuous
variables, i.e. environmental data layers, to eight variables (Table
A1.1). We assessed collinearity by constructing a correlation
matrix for the 21 variables based on Spearman’s correlation
coefficient (rs), and we removed all but one variable if rs was >
0.80 between two or more variables. In each case, we kept only
the variable with the highest percent contributions across the 48
global Maxent models which we initially ran.
With this reduced set of eight continuous variables and one
categorical variable (dominant land cover type), we generated
suitability distributions for the Japanese Paradise-Flycatcher
across Lanyu using the maximum entropy algorithm
implemented in the software Maxent (Phillips et al. 2006, Phillips
and Dudík 2008, Elith et al. 2011). We used Maxent with the
logistic output of probabilities. Maxent finds the probability
distribution of maximum entropy subject to constraints imposed
by the information available from the observed distribution of
the species and environmental conditions across the study area.
Thereby, Maxent transforms environmental variables into feature
vectors and then uses entropy as the means to generalize specific
observations of the species’ presence; therefore, it does not require
absence points within its theoretical framework.
Recommended default values for the Maxent modeling procedure
were used for the convergence threshold (10-5), maximum number
of iterations (500), data being randomly divided into 70% training
and 30% testing data, and cross-validation using jackknife
resampling. The selection of “features” (environmental variables
or functions thereof) was also carried out automatically, following
default rules dependent on the number of presence records.
In order to obtain the most suitable model for our data, we
generated 48 models by varying the following modeling features:
1. The number of background points was varied at four
settings: 500, 1000, 2500, and 5000.
2. The modeling features were varied with three categories:
linear, quadratic, and linear + quadratic (cf. Botero-
Delgadillo et al. 2015a,b).
3. The regularization multiplier (also called regularization
constant or beta-multiplier) was varied at four equidistant
intervals: 0.25, 0.50, 0.75, and 1.00 (cf. Botero-Delgadillo et
al. 2015a,b).
In this way, we generated 4 x 4 x 3 = 48 models for comparison.
First, we generated 48 models using all 22 environmental
variables, but only to score the heuristic estimates of each
variable’s relative model contribution and average each variable’s
mean contribution over the entire 48 models. This average percent
contribution was then used to eliminate highly correlating
variables. We then generated another 48 models using the same
combination of modeling features but with only the remaining
nine variables (eight continuous variables plus one categorical
variable).
We then evaluated the support for each of these 48 competing
models using the likelihood-based methods based on the
information theoretic approach proposed in Burnham and
Anderson (2002). We ranked models based on the Akaike’s
Information Criterion adjusted for small sample size (AICc), and
we accepted only those models with an AICc difference (ΔAICc)
< 2 as having “substantial” support (sensu page 70 in Burnham
and Anderson 2002; see also Majić et al. 2011, Grabowska-Zhang
et al. 2012, Hong et al. 2016). These are also the models with the
largest Akaike weights (Wi) and the smallest evidence ratios
(Burnham and Anderson 2002).
To choose a final model, the best model with the lowest ΔAICc =
zero can be used (e.g., Botero-Delgadillo et al. 2015a,b). However,
we chose to adopt the strategy of Burnham and Anderson (2002)
whereby all models with substantial support should be considered
as having value. Therefore, we generated an ensemble model
(sensu Araújo and New 2007) as our final model by combining
all the models with substantial support (ΔAICc < 2). Although
there are also several ways of combining models (M. Araújo, 2012,
personal communication), we simply calculated the mean of all the
models that were included into the final ensemble model.
Maxent also generates response curves, which show how each
environmental variable affects the prediction. These response
curves reflect the dependence of predicted suitability both on the
selected variable and on dependencies induced by correlations
between the selected variable and other variables (Phillips 2017).
We examined these response curves to check model fit and how
the environmental variable affects the prediction.
As recommended by Botero-Delgadillo et al. (2015a,b), model
significance was also evaluated with threshold-dependent
binomial probability tests applied in each replicate, and model
performance was evaluated with the mean values and standard

Avian Conservation and Ecology 13(1): 7
http://www.ace-eco.org/vol13/iss1/art7/
Fig. 3. Frequency distributions depicting the variation in (A) mean elevation, (B) percentage of tall forest, and
(C) mean normalized differenced vegetation index (NDVI) associated with the 120 presence grid cells for the
Japanese Paradise-Flycatcher (Terpsiphone atrocaudata). Note that the bimodal distribution in Fig. 3B results
from the species’ preference for stunted forest (left peak; see also Fig. A1.2) and tall forest (right peak).
deviations of the regularized training gain values, the threshold-
based omission error rates on training and test data, as well as
the AICc.
The best models based on AICc were also evaluated with the 5-
fold cross-validation technique (Peterson et al. 2011). Four runs
of the cross-validation procedure were obtained through data
reshuffling, which produced a total of 20 model replicates that
were used to assess uncertainty around the estimates of model
performance and significance (Botero-Delgadillo et al. 2015a,b).
Finally, we used the minimum training presence threshold
(Phillips et al. 2006) to obtain a binary spatial projection of
environmental suitability, i.e. to transform the logistic model
output from Maxent into a presence-absence grid map, because
it minimizes the inclusion of commission errors in model testing
(Botero-Delgadillo et al. 2015a,b). All these analyses were carried
out with Maxent version 3.4.1 (Phillips 2017), ArcGIS version
10.1, and ENMTools version 1.4.4 (Warren et al. 2010).
Population size estimate
To estimate population size, we divided the size of the entire area
deemed suitable in the binary distribution map generated using
the minimum training presence threshold by the mean estimated
size of a male’s territory, namely 1.16 ± 0.43 hectares. It should
be noted that this estimation assumes that every territory is
occupied.
Habitat change assessment
Our results below demonstrate that the Japanese Paradise-
Flycatcher occurred almost exclusively in hilltop shrub, stunted
forest, and tall forest. For our historical comparison of habitat
change, we therefore lumped these three land cover types into one
type called “forest.” We then compared the forest area of 1948 to
the forest area of 2006 using data layer 22 (Table A1.1) to assess
the amount of change in the area of habitat that is potentially
suitable for the Japanese Paradise-Flycatcher.
RESULTS
Conservation status assessment
A brief assessment of the current conservation status of the
Japanese Paradise-Flycatcher in Japan, Korea, Taiwan, and the
Philippines is given in Appendix 1. From the available but
relatively sparse information, it appears that the Japanese
population is now stable after a decline between the 1970s and
1990s, that the Korean population is probably stable or slightly
increasing (although some experts contended that the species has
been decreasing), the Taiwanese population is stable, while the
status of the Philippine population is unknown although it was
certainly present in 2006 and 2007.
Modeling species distribution
The field work in 2010 established 120 presence and 104 absence
grid cells (Fig. 2). Almost all presence grid cells were at some
distance from the coastline, i.e., almost no singing males were
found in the coastal lowlands. Away from coastal areas, the
presence grid cells were distributed relatively evenly across the
entire island, from north to south and east to west. The means,
variations, and ranges of the environmental values within the
presence and absence grid cells are given in Table A1.2.
The presence grid cells covered an elevational range from 16–399
m with a mean of 172 m (Table A1.2, Fig. 3A). The area of tall
forests within the presence grid cells ranged from 0–100%, with a
mean percentage of 26% (Fig. 3B). Furthermore, 115 out of the
120 presence grid cells (or 96%) ranged from 0.68–0.77 for mean
NDVI, with a mean of 0.74 (Fig. 3C). The dominant land cover
types for the presence grid cells were farm/grassland (5 grid cells),
hilltop shrub (6 grid cells), stunted forest (77 grid cells), and tall
forest (32 grid cells). Therefore, grid cells containing either mostly
stunted forest (Fig. A1.5) or mostly tall forest (Fig. 3B)
represented 91% of all the presence grid cells. Furthermore, grid
cells containing mostly one of the three “forest” types (namely,
hilltop scrub, stunted forest, or tall forest) represented 96% of all
the presence grid cells. To summarize, the Japanese Paradise-
Flycatcher was detected mostly at midelevations of 50–300 m,
preferred the areas of Lanyu with the highest NDVI values, and
was found almost exclusively in forest, and predominantly in
stunted or tall forest.
We ran Maxent with the nine selected environmental variables
and used them to construct 48 models. Of these 48 models, five
models had a ΔAICc < 2 (Table A1.3).
For these five “best” models, the mean regularized training gains
were 0.49 ± 0.04 (n = 5 * 20 replicates = 100 replicates, same
below), the mean values of the threshold-based training omission

Avian Conservation and Ecology 13(1): 7
http://www.ace-eco.org/vol13/iss1/art7/
Table 1. Heuristic estimates of the relative contribution and relative performance of nine environmental variables to modeling the
distribution of the Japanese Paradise-Flycatcher (Terpsiphone atrocaudata) on Lanyu Island, Taiwan. The variables are ordered by
their mean percent contribution. NDVI = normalized differenced vegetation index.
Environmental variable Mean percent contribution†Training gain without‡Training gain with only§
altitude mean 25.20 ± 5.76 0.401 ± 0.020 0.249 ± 0.013
% tall forest 16.37 ± 2.99 0.475 ± 0.024 0.190 ± 0.012
NDVI mean 12.85 ± 4.80 0.479 ± 0.027 0.196 ± 0.008
% bare coastal land 9.74 ± 3.56 0.507 ± 0.024 0.106 ± 0.005
% hilltop shrub 9.72 ± 1.47 0.487 ± 0.022 0.026 ± 0.004
% farm/grassland 8.30 ± 1.32 0.475 ± 0.022 0.059 ± 0.006
dominant land cover type 7.35 ± 4.03 0.507 ± 0.024 0.194 ± 0.013
altitude std. dev. 5.91 ± 1.00 0.490 ± 0.023 0.133 ± 0.010
% aspect south 4.55 ± 3.93 0.489 ± 0.022 0.022 ± 0.002
†Values represent each variable’s mean percent contribution (%) ± 1 standard deviation.
‡Values represent the model’s training gain achieved with all but the regarded variable ± 1 standard deviation.
§Values represent the model’s training gain achieved with only the regarded variable and excluding the remaining eight variables ± 1 standard deviation.
Fig. 4. The ensemble distribution model of the Japanese
Paradise-Flycatcher (Terpsiphone atrocaudata) across Lanyu,
Taiwan, shown as a probability surface (see text for details).
Probability values are depicted in two colors from lowest (blue)
to highest (red) in 10 equal intervals.
error rates were 0.07 ± 0.02 (n = 100), and the threshold-based
test omission error rates were 0.17 ± 0.05 (n = 100), thereby
confirming that these models performed reasonably well. All
cross-validation replicates of these models were statistically
significant according to the threshold-dependent binomial
probability test (all P < 0.05).
Among these 48 models, the three variables that were almost
consistently chosen as the variables with the highest heuristic
estimates of their relative model contribution were mean
elevation, percentage of tall forest, and mean NDVI, having a
combined contribution of 54.4% (Table 1). Although these three
variables did not yield the highest training gain achieved with all
Fig. 5. The binary distribution model of Japanese Paradise-
Flycatcher (Terpsiphone atrocaudata) across Lanyu, Taiwan,
using the minimum training presence threshold to turn the
probability surface depicted in Fig. 4 into a binary map with
modeled presence (black, n = 1203) and absence (grey, n =
3429) grid cells.
but the regarded variable, these three variables achieved the
highest training gain achieved with only the regarded variable,
with one exception, namely dominant land cover type (Table 1).
We combined these five models into an ensemble model (Fig. 4).
Using the minimum training presence threshold, we then
converted this ensemble model into a binary map of the estimated
ecological niche of the Japanese Paradise-Flycatcher on Lanyu
Island (Fig. 5). This binary map covers 1203 out of a total of 4632
grid cells, or 26.0% of the island’s surface (equaling 12.03 km²).
After modeling environmental suitability as a function of these
nine variables, the individual response curves generated by
Maxent for the three most important environmental variables

Avian Conservation and Ecology 13(1): 7
http://www.ace-eco.org/vol13/iss1/art7/
Fig. 6. The probability of presence of the Japanese Paradise-Flycatcher (Terpsiphone atrocaudata) in relation to
spatial variation in (A) mean elevation (in m); (B) percentage of tall forest; and (C) mean normalized
differenced vegetation index (NDVI; unitless) in Lanyu Island, Taiwan. Each individual response curve was
generated by Maxent, with the curve representing the Maxent model created by using only the respective
variable. Here, we used the curves generated for the best model (ΔAICc = 0). The vertical axis represents the
probability of presence from the logistic output of Maxent model. The red line shows the average of the 100
replicate runs, while the blue lines shows ± 1 standard deviation.
indicated that environmental suitability for the Japanese Paradise-
Flycatcher was highest at midelevations of about 30–350 m (Fig.
6A) and increased with the percentage of tall forest within the
grid cell (Fig. 6B) and with NDVI (Fig. 6C). These results
generated by Maxent therefore correspond reasonably well with
the results from the presence grid cells (Fig. 3).
Population size estimate
Because the average size of a male’s territory was 1.16 ± 0.43
hectares, we extrapolated that the area deemed suitable in the
binary distribution map (namely 12.03 km²) could hold 1037.1
territories. If we use the lower and upper estimates given by the
standard deviation (0.73 and 1.59, respectively), we can calculate
an upper and lower limit of the number of male territories of
1647.9 territories and 756.6 territories, respectively.
Habitat change assessment
Our historical comparison of the availability of “forest” revealed
that very small areas of forest were lost between 1948 and 2006
(specifically, 0.66 km² or 1.43% of the island’s area). Almost all
of these areas are located near the coast (Fig. A1.6). During the
same time period, relatively large areas of forest were gained
(specifically, 8.23 km² or 17.77% of the island’s area). These areas
are distributed across the entire island, but especially in the areas
just south of the island’s center (Fig. A1.7). The net gain of forest
areas was therefore 7.57 km² (or 16.35% of the island’s area).
Some of these gained forest areas are located within our binary
presence model (Fig. A1.8), and this gained area covers 3.76 km²
(or 8.13% of the island’s area). Assuming that the Japanese
Paradise-Flycatcher’s habitat preference for the three “forest”
categories did not change between 1948 and 2006, and that other
important variables, e.g., territory size, also did not change, this
area of potentially suitable habitat gained from 1948 to 2006 could
support an additional 324.3 male territories, or 31.27% of the
entire 1037.1 territories.
DISCUSSION
Our field work indicates that males of the periophthalmica
subspecies of the Japanese Paradise-Flycatcher found on Lanyu
Island prefer to establish territories in midelevational forest
habitats that include both stunted and tall forests and that have
relatively high NDVI values. Using the 120 presence locations
from our field work, we produced an ensemble model of the
flycatcher’s ecological niche that estimates that such suitable
habitat extends over about 12.0 km² or about 26% of Lanyu’s
area. Because our field work also established that the average
size of a male’s territory is 1.16 ± 0.43 hectares, we could
extrapolate that approximately 1037 territories may exist on
Lanyu Island, with upper and lower limits of approximately 1648
territories and 757 territories, respectively.
However, we note that any such population size estimate is
associated with various uncertainties, such as that not all
potential territories may be occupied, that there may be annual
variations in population size, and that the use of different
thresholds (Nenzén and Araújo 2011) than the one we used
(minimum training presence threshold) would have resulted in
different estimates of the area of suitable habitat. Nevertheless,
even with these uncertainties, our estimate is a substantial
increase over previous estimates of < 100 breeding pairs (Brazil
2009) and < 500 individuals (Fang 2005).
Our results can also be used to design conservation measures for
the Lanyu population because our results suggest that the
Japanese Paradise-Flycatcher prefers relatively wet midelevational
forest habitats. Currently, the Lanyu population is probably not
threatened in its preferred habitat because limited land use
conversion has been occurring in Lanyu, with an overall net gain
of 7.57 km² of forest habitats between 1948 and 2006, which, all
other things being equal, should have resulted in a population
increase of around 30%. Despite the fact that the local
population rejected a plan to turn parts of the island into a
national park in the 1980s and 1990s (Huang 1997), there are
currently no socioeconomic factors that would drive forest
destruction or conversion. Therefore, continuous monitoring
and maintenance of this suitable forest habitat should ensure the
long-time survival of this species, assuming no stochastic
catastrophic events, such as typhoons, or long-term changes,
such as climate change.
Consequently, the question arises whether the Japanese
Paradise-Flycatcher should remain in the “near-threatened”

Avian Conservation and Ecology 13(1): 7
http://www.ace-eco.org/vol13/iss1/art7/
category. The main reason given by BirdLife International (2015)
is that declines were noted in parts of Japan’s breeding range that
were presumably caused mostly by habitat loss and degradation
within the wintering grounds. However, as outlined in Appendix
1, this assessment is based mostly on relatively old studies and
only from parts of Japan. Since then, its status in Japan appears
unchanged since the 1990s, including the Ryukyu population (H.
Higuchi, 2015, personal communication). Our study suggests that
the Lanyu population is also relatively safe, and despite some
yearly fluctuations (L. L. S., unpublished data), it may even have
increased over the last half century because of habitat expansion,
although there are no historical data on population numbers to
confirm or reject this supposition.
Finally, the Korean population may have been increasing, which
could be due to a northward expansion due to climate change
(e.g., Kwon et al. 2014, Wu and Shi 2016) and successful
reforestation, although other experts considered this to be an
artificial effect derived from increased sampling effort or even
contended that the species has been decreasing (Appendix 1). If
the Philippine population is also found to be stable (pending
further study), a case could be made for down-listing the species
to “least concern.” However, there remains an urgent need to
find out more about this species’ migration routes and wintering
grounds because habitat loss and degradation in the wintering
grounds were suspected to have caused the original decline of
the Japanese populations (BirdLife International 2015). With
rainforest habitats continuing to decline rapidly across Southeast
Asia (Wilcove et al. 2013, Walther et al. 2016), it is likely that the
Japanese Paradise-Flycatcher is being impacted by these
environmental changes. Given rapid environmental changes, the
continuous reassessment of the conservation status of Taiwan’s
birds (Walther et al. 2011, Wu et al. 2014, Lin et al. 2016) and
East Asia’s birds (Collar et al. 2001, Kirby et al. 2008) must
remain a research priority.
Consequently, we have four main recommendations concerning
the T. a. periophthalmica subspecies:
1. Because there is no reliable population estimate for the
Philippine population (Appendix 1), a study of the size of
the Philippine breeding population should be a
conservation priority.
2. Further monitoring and protection of Taiwan’s breeding
population should be a conservation priority for Taiwanese
researchers and authorities.
3. A study of the species’ migration and wintering grounds is
overdue, including changes and threats to suitable habitats
in these grounds.
4. A study to determine if this subspecies actually deserves
full species status (e.g., McGregor 1907, Alcasid 1965)
should be a priority for taxonomists.
Indeed, further taxonomic and genetic studies are warranted for
all the subspecies (or even all of its independent conservation
units) of the Japanese Paradise-Flycatcher because some or all
of its subspecies may be sufficiently distinct to merit full species
status. If, for example, T. a. periophthalmica was to be elevated
to species status, its conservation status might be “near-
threatened” or even “vulnerable” (IUCN 2012) because of its
existence in only five known locations (Lanyu and the four
Batanes islands), small geographic range (~12 km² on Lanyu and
unknown in Batanes), and its relatively small population size
(~1000 singing males on Lanyu and unknown in Batanes). Under
the IUCN criteria, species with either small populations (< 10,000
individuals) or small areas of occupancy (< 2000 km²) may be
classified as vulnerable, regardless of the trajectory of their
populations. T. a. periophthalmica would fulfil both of these
criteria.
Even if T. a. periophthalmica was not elevated to species status,
these two island populations could justifiably be considered
independent conservation units, especially given that they are the
only populations within either Taiwan or the Philippines. Island
populations as independent conservation units have been
proposed for other island bird populations (e.g., Dudaniec et al.
2011, Garcia-del-Rey et al. 2013, Ando et al. 2014, Forcina et al.
2014, Pruett et al. 2017). Conservation units have also been called
evolutionarily significant units or management units (cf. Moritz
1994, Rayner et al. 2010). However, these studies have routinely
included genetic analyses, which were not part of our study, again
emphasizing the need for more genetic and taxonomic studies of
this species.
The Lanyu and Batanes islands are regularly subjected to
devastating super typhoons (e.g., Fritz 2016), with no information
about the impact of these typhoons on local bird populations
(although see Hong et al. 2016 for an example from Taiwan’s
mainland). This lack of information about the recent fate of these
island populations further underlines the need for more
continuous field work, especially in Batanes, but also in Lanyu.
We therefore recommend that the governments of Taiwan and
the Philippines should support future research on bird
populations on islands that could be designated as independent
conservation units. We further recommend that more continuous
monitoring of such populations is financed, because the only
continuous field work on the Lanyu population is the one reported
in this study which lasted for only two breeding seasons. Although
the recently established Taiwan Breeding Bird Survey (Ko et al.
2015) is a step in the right direction, more long-term monitoring
targeted at specific species and populations (e.g., Lin et al. 2007)
is required.
Responses to this article can be read online at:
http://www.ace-eco.org/issues/responses.php/1167
Acknowledgments:
We are grateful to Gui-Qing Wang for providing valuable personal
field observations from Lanyu, and Chin-Kuo Lee for interpreting
aerial photographs of Lanyu. We thank Mark Brazil, Amy
Chernasky, Chang-Yong Choi, Mike Crosby, Juan Carlos Tecson
Gonzalez, Hiroyoshi Higuchi, Han-kyu Kim, Jin-Won Lee, Ruey-
Shing Lin, and Carl Oliveros for providing references and additional
information, Tsai-Yu Wu for help with modelling, Tsai-Yu Wu and
Yu-Wen Emily Dai for translations, and several reviewers for
comments. BAW was financially supported by Taipei Medical
University. TS was financially supported by a grant from German
Academic Exchange Service (DAAD). Covering the costs for

Avian Conservation and Ecology 13(1): 7
http://www.ace-eco.org/vol13/iss1/art7/
overall logistics, material, and additional support by field assistants
was only possible because of substantial funding from Forestry
Bureau of Council of Agriculture of Executive Yuan and Academia
Sinica.
LITERATURE CITED
Alcasid, G. L. 1965. Terpsiphone atrocaudata on Mindoro,
Philippines. Auk 82:644. http://dx.doi.org/10.2307/4083230
Ando, H., H. Ogawa, S. Kaneko, H. Takano, S.-I. Seki, H. Suzuki,
K. Horikoshi, and Y. Isagi. 2014. Genetic structure of the critically
endangered Red-headed Wood Pigeon Columba janthina nitens
and its implications for the management of threatened island
populations. Ibis 156:153-164. http://dx.doi.org/10.1111/ibi.12120
Araújo, M. B., and M. New. 2007. Ensemble forecasting of species
distributions. Trends in Ecology and Evolution 22:42-47. http://dx.
doi.org/10.1016/j.tree.2006.09.010
BirdLife International. 2015. Species factsheet: Japanese
Paradise-Flycatcher Terpsiphone atrocaudata. BirdLife International,
Cambridge, UK. [online] URL: http://www.birdlife.org/
datazone/species/factsheet/22707151
Botero-Delgadillo, E., N. Bayly, S. Escudero-Páez, and M. I.
Moreno. 2015a. Understanding the distribution of a threatened
bird at multiple levels: a hierarchical analysis of the ecological
niche of the Santa Marta Bush-Tyrant (Myiotheretes pernix).
Condor 117:629-643. http://dx.doi.org/10.1650/CONDOR-15-26.1
Botero-Delgadillo, E., N. Bayly, C. Gómez, P. C. Pulgarín-R., and
C. A. Páez. 2015b. An assessment of the distribution, population
size and conservation status of the Santa Marta foliage-gleaner
Automolus rufipectus: a Sierra Nevada de Santa Marta endemic.
Bird Conservation International 25:451-465. http://dx.doi.
org/10.1017/S0959270914000513
Brazil, M. A. 1991. The birds of Japan. Christopher Helm,
London, UK.
Brazil, M. 2009. Birds of East Asia: China, Taiwan, Korea, Japan,
and Russia. Princeton University Press, Princeton, New Jersey,
USA.
Burnham, K. P., and D. R. Anderson. 2002. Model selection and
inference: a practical information-theoretic approach. Second
edition. Springer Verlag, New York, New York, USA. http://dx.
doi.org/10.1007/978-1-4757-2917-7
Chao, W.-C., G.-Z. M. Song, K.-J. Chao, C.-C. Liao, S.-W. Fan,
S.-H. Wu, T.-H. Hsieh, I.-F. Sun, Y.-L. Kuo, and C.-F. Hsieh.
2010. Lowland rainforests in southern Taiwan and Lanyu, at the
northern border of Paleotropics and under the influence of
monsoon wind. Plant Ecology 210:1-17. http://dx.doi.
org/10.1007/s11258-009-9694-0
Coates, B. J., G. C. L. Dutson, C. E. Filardi, P. Clement, P. A.
Gregory, and C. W. Moeliker. 2006. Family Monarchidae
(monarch-flycatchers). Pages 244-329 in J. del Hoyo, A. Elliott,
and D. Christie, editors. Handbook of the birds of the world. Vol.
11. Old World flycatchers to Old World warblers. Lynx Edicions,
Barcelona, Spain.
Collar, N. J., A. V. Andreev, S. Chan, M. J. Crosby, S. Subramanya,
and J. A. Tobias, editors. 2001. Threatened birds of Asia: the
BirdLife International red data book. BirdLife International,
Cambridge, UK.
Ding, T.-S., C.-S. Juan, R.-S. Lin, C.-Y. Pan, Y.-J. Tsai, J. Wu, and
Y.-H. Yang. 2014. The 2014 CWBF checklist of the birds of Taiwan.
Bird Record Committee, Chinese Wild Bird Federation, Taipei,
Taiwan. [online] URL: http://birdingattaiwan.com/
The_2014_CWBF_Checklist.pdf
Duckworth, J. W., and N. Moores. 2008. A re-evaluation of the
pre-1948 Korean breeding avifauna: correcting a ‘founder effect’
in perceptions. Forktail 24:25-47. [online] URL: http://
birdingasia.org/wp-content/uploads/2012/10/Duckworth-Korea.
pdf
Dudaniec, R. Y., B. E. Schlotfeldt, T. Bertozzi, S. C. Donnellan,
and S. Kleindorfer. 2011. Genetic and morphological divergence
in island and mainland birds: informing conservation priorities.
Biological Conservation 144:2902-2912. http://dx.doi.org/10.1016/
j.biocon.2011.08.007
Elith, J., S. J. Phillips, T. Hastie, M. Dudík, Y. E. Chee, and C. J.
Yates. 2011. A statistical explanation of MaxEnt for ecologists.
Diversity and Distributions 17:43-57. http://dx.doi.org/10.1111/
j.1472-4642.2010.00725.x
Fang, W.-H. 2005. A guide to threatened birds of Taiwan. Owl
Publishing House, Taipei, Taiwan.
Forcina, G., P. Panayides, N. Kassinis, M. Guerrini, and F.
Barbanera. 2014. Genetic characterization of game bird island
populations: the conservation of the Black Francolin (Francolinus
francolinus) of Cyprus. Journal for Nature Conservation 22:15-22.
http://dx.doi.org/10.1016/j.jnc.2013.07.004
Fritz, A. 2016. Remember the island in the super typhoon eye?
Not a single person died, reports say. Washington Post, 20
September. [online] URL: https://www.washingtonpost.com/
news/capital-weather-gang/wp/2016/09/20/remember-the-island-
in-the-super-typhoon-eye-not-a-single-person-died-reports-say/?
utm_term=.bab57a84ab9b
Garcia-del-Rey, E., G. Marthinsen, P. Calabuig, L. Estévez, L. E.
Johannessen, A. Johnsen, T. Laskemoen, and J. T. Lifjeld. 2013.
Reduced genetic diversity and sperm motility in the endangered
Gran Canaria blue chaffinch Fringilla teydea polatzeki. Journal
of Ornithology 154:761-768. http://dx.doi.org/10.1007/s10336-013-0940-9
Gonzalez, J. C. T., L. E. Afuang, and A. V. Lacaste. 2008.
Identifying conservation priorities for terrestrial vertebrate fauna
in the Batanes Islands, northern Philippines. Journal of Nature
Studies 7:1-8.
Grabowska-Zhang, A. M., T. A. Wilkin, and B. C. Sheldon. 2012.
Effects of neighbor familiarity on reproductive success in the
Great Tit (Parus major). Behavioral Ecology 23:322-333. http://
dx.doi.org/10.1093/beheco/arr189
Handbook of the Birds of the World (HBW) Alive. 2016. Japanese
Paradise-Flycatcher (Terpsiphone atrocaudata). Lynx Edicions,
Barcelona, Spain. [online] URL: http://www.hbw.com/species/
japanese-paradise-flycatcher-terpsiphone-atrocaudata

Avian Conservation and Ecology 13(1): 7
http://www.ace-eco.org/vol13/iss1/art7/
Hong, S.-Y., B. A. Walther, M.-C. Chiu, M.-H. Kuo, and Y.-H.
Sun. 2016. Length of the recovery period after extreme flood is
more important than flood magnitude in influencing reproductive
output of Brown Dippers (Cinclus pallasii) in Taiwan. Condor
118:640-654. http://dx.doi.org/10.1650/CONDOR-16-1.1
Huang, Y.-W. 1997. An analysis of decision-making process of
establishing Lanyu National Park. Journal of Geographical
Science 23:13-31.
International Union for Conservation of Nature (IUCN). 2012.
IUCN Red List categories and criteria: Version 3.1. Second
edition. IUCN, Gland, Switzerland and Cambridge, UK. [online]
URL: http://s3.amazonaws.com/iucnredlist-newcms/staging/public/
attachments/3097/redlist_cats_crit_en.pdf
Jeyarajasingam, A., and A. Pearson. 2012. A field guide to the
birds of Peninsular Malaysia and Singapore. Oxford University
Press, Oxford, UK.
Johnson, T. H., and A. J. Stattersfield. 1990. A global review of
island endemic birds. Ibis 132:167-180. http://dx.doi.org/10.1111/
j.1474-919X.1990.tb01036.x
Kennedy, R. S., P. C. Gonzales, E. C. Dickinson, H. C. Miranda,
and T. H. Fisher. 2000. A guide to the birds of the Philippines.
Oxford University Press, Oxford, UK.
Kier, G., H. Kreft, T. M. Lee, W. Jetz, P. L. Ibisch, C. Nowicki, J.
Mutke, and W. Barthlott. 2009. A global assessment of endemism
and species richness across island and mainland regions.
Proceedings of the National Academy of Sciences of the United
States of America 106:9322-9327. http://dx.doi.org/10.1073/
pnas.0810306106
Kirby, J. S., A. J. Stattersfield, S. H. M. Butchart, M. I. Evans, R.
F. A. Grimmett, V. R. Jones, J. O'Sullivan, G. M. Tucker, and I.
Newton. 2008. Key conservation issues for migratory land- and
waterbird species on the world’s major flyways. Bird Conservation
International 18 (Supplement 1):S49-S73. http://dx.doi.
org/10.1017/S0959270908000439
Ko, C.-J., M.-W. Fan, Y.-X. Jiang, W.-J. Yu, Y.-Y. Lo, R.-S. Lin,
K. Lin, and P.-F. Lee. 2015. 2013 Taiwan breeding bird survey.
Annual Report. Endemic Species Research Center, Council of
Agriculture, Endemic Species Research Institute, Nantou,
Taiwan. [online] URL: https://www.researchgate.net/
publication/273026543_2013_Taiwan_Breeding_Bird_Survey_A
nnual_Report_taiwanfanzhiniaoleidadiaocha2013nianbao_Trad
itional_Chinese_version
Kwon, T.-S., C. M. Lee, and S.-S. Kim. 2014. Northward range
shifts in Korean butterflies. Climatic Change 126:163-174. http://
dx.doi.org/10.1007/s10584-014-1212-2
Lin, R.-S., P.-F. Lee, T.-S. Ding, and Y.-T. K. Lin. 2007.
Effectiveness of playbacks in censusing the Fairy Pitta (Pitta
nympha) during the breeding season in Taiwan. Zoological Studies
46:242-248. [online] URL: http://zoolstud.sinica.edu.tw/Journals/46.2/242.
pdf
Lin, R.-S., Y.-J. Lu, T.-J. Tseng, W.-J. Chen, C.-J. Ko, and, C.-H.
Yang. 2016. The red list of birds of Taiwan, 2016. Endemic Species
Research Institute and Forestry Bureau, Council of Agriculture,
Executive Yuan, Nantou, Taiwan. [online] URL: https://www.
researchgate.net/publication/312045198_The_Red_List_of_Bird
s_of_Taiwan_2016
Majić, A., A. M. T. de Bodonia, Đ. Huber, and N. Bunnefeld.
2011. Dynamics of public attitudes toward bears and the role of
bear hunting in Croatia. Biological Conservation 144:3018-3027.
http://dx.doi.org/10.1016/j.biocon.2011.09.005
McGregor, R. C. 1907. The birds of Batan, Camiguin, Y'Ami,
and Babuyan Claro, islands north of Luzon. Philippine Journal
of Science 2A:337-351.
Moritz, C. 1994. Defining ‘evolutionarily significant units’ for
conservation. Trends in Ecology and Evolution 9:373-375. http://
dx.doi.org/10.1016/0169-5347(94)90057-4
Nenzén, H. K., and M. B. Araújo. 2011. Choice of threshold alters
projections of species range shifts under climate change.
Ecological Modelling 222:3346-3354. http://dx.doi.org/10.1016/j.
ecolmodel.2011.07.011
Nuytemans, H. 1998. Notes on Philippine birds: interesting
records from northern Luzon and Batan Island. Forktail 14:39-42.
[online] URL: http://orientalbirdclub.org/wp-content/uploads/2012/09/
Nuytemans-Philippine.pdf
Oliveros, C., A. T. Peterson, and M. J. C. Villa. 2008. Birds,
Babuyan Islands, province of Cagayan, Northern Philippines:
new island distribution records. Check List 4:137-141. http://dx.
doi.org/10.15560/4.2.137
Peterson, A. T., J. Soberón, R. G. Pearson, R. P. Anderson, E.
Martínez-Meyer, M. Nakamura, and M. B. Araújo. 2011.
Ecological niches and geographic distributions. Monographs in
Population Biology no. 49. Princeton University Press, Princeton,
New Jersey, USA.
Phillips, S. 2017. Maxent software for modeling species niches and
distributions. [online] URL: http://biodiversityinformatics.amnh.
org/open_source/maxent/
Phillips, S. J., R. P. Anderson, and R. E. Schapire. 2006. Maximum
entropy modeling of species geographic distributions. Ecological
Modelling 190:231-259. http://dx.doi.org/10.1016/j.ecolmodel.2005.03.026
Phillips, S. J., and M. Dudík. 2008. Modeling of species
distributions with Maxent: new extensions and a comprehensive
evaluation. Ecography 31:161-175. http://dx.doi.org/10.1111/
j.0906-7590.2008.5203.x
Pruett, C. L., A. Ricono, C. Spern, and K. Winker. 2017. Island
life and isolation: the population genetics of Pacific Wrens on the
North Pacific Rim. Condor 119:131-142. http://dx.doi.
org/10.1650/CONDOR-16-183.1
Rayner, M. J., C. J. F. Carraher, M. N. Clout, and M. E. Hauber.
2010. Mitochondrial DNA analysis reveals genetic structure in
two New Zealand Cook’s Petrel (Pterodroma cookii) populations.
Conservation Genetics 11:2073-2077. http://dx.doi.org/10.1007/
s10592-010-0072-1
Reaser, J. K., L. A Meyerson, Q. Cronk, M. De Poorter, L. G.
Eldrege, E. Green, M. Kairo, P. Latasi, R. N. Mack, J.
Mauremootoo, D. O’Dowd, W. Orapa, S. Sastroutomo, A.
Saunders, C. Shine, S. Thrainsson, and L. Vaiutu. 2007.
Ecological and socioeconomic impacts of invasive alien species

Avian Conservation and Ecology 13(1): 7
http://www.ace-eco.org/vol13/iss1/art7/
in island ecosystems. Environmental Conservation 34:98-111.
http://dx.doi.org/10.1017/S0376892907003815
Severinghaus, L. L., T.-S. Ding, W.-H. Fang, W.-H. Lin, M.-C.
Tsai, and C.-W. Yen. 2010. The avifauna of Taiwan. [Translated
from the Chinese]. Forestry Bureau of Council of Agriculture of
Executive Yuan, Taipei, Taiwan.
Severinghaus, L. L., T.-S. Ding, W.-H. Fang, W.-H. Lin, M.-C.
Tsai, and C.-W. Yen. 2017. The avifauna of Taiwan. Forestry
Bureau, Council of Agriculture, Taipei, Taiwan. [online] URL:
http://conservation.forest.gov.tw/0001888
Sutherland, W. J., I. Newton, and R. E. Green, editors. 2004. Bird
ecology and conservation: a handbook of techniques. Oxford
University Press, Oxford, UK. http://dx.doi.org/10.1093/acprof:
oso/9780198520863.001.0001
Veitch, C. R., and M. N. Clout, editors. 2002. Turning the tide:
the eradication of invasive species. IUCN Species Specialist
Group, Gland, Switzerland and Cambridge, UK.
Walther, B. A., C. Boëte, A. Binot, Y. By, J. Cappelle, J. J. Carrique-
Mas, M. Chou, N. Furey, S. Kim, C. Lajaunie, S. Lek, P. Méral,
M. Neang, B.-H. Tan, C. Walton, and S. Morand. 2016.
Biodiversity and health: lessons and recommendations from an
interdisciplinary conference to advise Southeast Asian research,
society and policy. Infection, Genetics and Evolution 40:29-46.
http://dx.doi.org/10.1016/j.meegid.2016.02.003
Walther, B. A., T.-Y. Wu, Y.-H. Chen, R.-S. Lin, and P.-F. Lee.
2011. Using species distribution models to assess the rarity and
conservation status of Taiwanese birds. Taiwan Journal of
Biodiversity 13:295-322. [online] URL: http://tesri.tesri.gov.tw/
files/tesri_protect/tesri_queen_20120614101857/3.PDF
Warren, D. L., R. E. Glor, and M. Turelli. 2010. ENMTools: a
toolbox for comparative studies of environmental niche models.
Ecography 33:607-611. http://dx.doi.org/10.1111/j.1600-0587.2009.06142.
x
Wilcove, D. S., X. Giam, D. P. Edwards, B. Fisher, and L. P. Koh.
2013. Navjot’s nightmare revisited: logging, agriculture, and
biodiversity in Southeast Asia. Trends in Ecology and Evolution
28:531-540. http://dx.doi.org/10.1016/j.tree.2013.04.005
Wu, J., and Y. Shi. 2016. Attribution index for changes in
migratory bird distributions: the role of climate change over the
past 50 years in China. Ecological Informatics 31:147-155. http://
dx.doi.org/10.1016/j.ecoinf.2015.11.013
Wu, T.-Y., B. A. Walther, Y.-H. Chen, R.-S. Lin, and P.-F. Lee.
2014. Reassessment of the conservation status and protected area
coverage of Taiwanese birds: how distribution modelling can help
species conservation. Bird Conservation International 24:223-238.
http://dx.doi.org/10.1017/S0959270913000336
Editor-in-Chief: Keith A.Hobson
Subject Editor: Nicholas JBayly

1
Appendix 1. Brief assessment of the current conservation status of the Japanese
Paradise-Flycatcher.
Supplementary Text
Text A1. Brief assessment of the current conservation status of the Japanese Paradise-Flycatcher.
This brief assessment is based on the information obtained by the fourth and corresponding author
(see Methods in main text).
The Japanese Paradise-Flycatcher has been categorized as near-threatened since 1994 because it is
suspected to have been in a moderately rapid decline, probably because of habitat degradation and
habitat loss on both the breeding and wintering grounds (Higuchi and Morishita 1999, BirdLife
International 2015). However, this status assessment is mainly based on its decrease in Japan where it
was common on Kyushu, Shikoku, and southern Honshu up to Tokyo’s latitude, and scarcer further
north, but has decreased during the last 50 years to become uncommon and local except in western
Japan (specifically, Kyushu, Shikoku, and Tsushima) where it remains locally common (Brazil 1991,
2013). In addition, this assessment is mainly based on relatively old studies from Japan which detected a
steep decline in parts of the Japanese breeding population between the 1970s and 1990s (Hirano 1996,
Higuchi and Morishita 1999), including the complete extinction in a forest plot near Higashimatsuyama
city, Japan, between 1972 and 1995 (Uchida 1996 cited in Higuchi and Morishita 1999), and population
declines detected in Yamaguchi Prefecture between 1973 and 1995 (Yamamoto and Seto 1997) and in
Amami Island, Ryukyu, between 1985 and 2001 (Sugimura et al. 2003). According to H. Higuchi (in
litt. 2015), no further studies have been published since then to assess the species’ current status in
Japan, although its status appears unchanged since the 1990s. Furthermore, the species was recorded as a
common breeder on Nakano-shima Island, Japan (Higuchi and Morishita 1999) and described as
common on Ryukyu Islands (Brazil 1991, Coates et al. 2006). H. Higuchi (in litt. 2015) asserted that the
Ryukyu population remains common and stable.

2
In Korea, the Ministry of the Environment designated the Japanese Paradise-Flycatcher as an
Endangered Bird Species Level II and reported records up to 37°N between 1997 and 2005 (Kim et al.
2010a). The distribution has been moving northwards with several recently confirmed records over 37°N
(H. Kim, in litt. 2016). A recent status report published by Birds Korea (2014) categorized the species as
an amber species which means it has (1) a global status of near-threatened, and (2) its national status
assessed by NIBR (2011, 2012) is “vulnerable.” This report also stated that the species “is increasingly
recorded in the ROK [= South Korea], with recent breeding confirmed north to at least 35°N,” and both
its historical and recent population trends are increasing. The breeding population was estimated to be
between 100 and 999 individuals. However, Kim et al. (2010c) wrote that the species “has rapidly
declined due to habitat loss (i.e. deforestation and industrialization).” This apparent contradiction in the
assessment of the species’ trend is likely due to several factors: (1) increased sampling effort in both
intensity and geographic coverage because of (i) increased survey efforts (C.-Y. Choi in litt. 2016) and
(ii) the “increased number of birders and bird photographers during [the last] 20 years [which] may be
over 100 times more than before” (H. Kim, in litt. 2016); (2) successful reforestation after the Korean
War has steadily increased suitable forest habitat in South Korea while North Korea continues to suffer
from deforestation (C.-Y. Choi, H. Kim, in litt. 2016); (3) however, locally, detrimental pressures may
negatively affect the species. In conclusion, the general impression is that the species trend is stable or
slightly increasing while spreading northwards; however, this impression is somewhat clouded by the
increased observer density and the lack of reliable long-term data on a national basis. C.-Y. Choi (in litt.
2016) even asserted that “any nationwide assessment on the trend of the species is unreliable and
probably biased at this stage.”
A 2010 survey of Jeju Island, which is a 1,848 km2 island of the southern tip of Korea, found a total
of 124 individuals in evergreen forest between 83 and 1106 meters a. s. l. (Kim et al. 2010b). Jeju’s
broad-leaved forests have increased in area and quality, and many parts of them are protected within a
national park and thus serve as a stronghold of the species in Korea (C.-Y. Choi, in litt. 2016).
Within the administrative boundaries of Taiwan, T. a. periophtalmica is known to breed exclusively
on Lanyu, and its conservation status is given as category 2 “rare and valuable” (Severinghaus et al.
2010, 2017, WCAT 2013, Lin et al. 2016). Fang (2005) estimated the population as fewer than 500
individuals and warned that clearance for agriculture, roads and construction is reducing suitable habitat

3
but admitted that “its breeding population in Taiwan is difficult to assess.” Brazil (2009) estimated fewer
than 100 breeding pairs in Lanyu, but without giving any further details. Lin et al. (2016) listed the
species under “Nationally Near-threatened bird taxa in Taiwan.”
During the 2009 breeding season (see also main text), preliminary data on the species’ phenology
was collected (Severinghaus, L. L., and M. L. Bai. 2009. Habitat use and breeding biology of the black
paradise flycatcher (in Chinese). Forestry Bureau of Council of Agriculture of Executive Yuan, Taipei,
Taiwan. Unpublished Report) which established that the Japanese Paradise-Flycatcher was present on
Lanyu from late February until mid-September. After mid-September, T. B. never observed any
individuals. However, a few individuals were occasionally observed in some previous years during the
winter season by L. L. S. (unpublished data) and a local birdwatcher (G.-Q. Wang). Severinghaus et al.
(2017) wrote that after the breeding season “most leave to overwinter somewhere else; only a few
remain on Orchid Island for the winter.” Therefore, the vast majority of individuals are absent during the
winter months.
The Japanese Paradise-Flycatcher breeds on the islands or island groups of the Batanes (namely
Batan, Itbayat, Ivuhos islands, and Sabtang) in the very north of the Philippines (Gonzalez et al. 2008),
with 91 captures in suitable habitat during a 2006-2007 survey (J.C.T. Gonzalez, in litt. 2016) where it
was described as common (Coates et al. 2006) or fairly common (J.C.T. Gonzalez, in litt. 2016).
With only two known breeding populations in Lanyu and Batanes, this subspecies may be
threatened because (1) its range is small; and (2) like the other subspecies, it may suffer from the habitat
destruction of the wintering grounds.
Besides the incomplete and sometimes contradictory information above, the overall conservation
status of the Japanese Paradise-Flycatcher also remains questionable because no reliable global
population size estimate exists. The only estimates for breeding populations are the ones given above for
Lanyu and in the present study and the very wide-ranging estimates of 100-10,000 breeding pairs for
both Japan and Korea (Brazil 2009, BirdLife International 2015). Because of these uncertainties,
BirdLife International (2015) conceded that the data quality for this species was poor and recommended
that therefore the species “should be carefully monitored.”

4
LITERATURE CITED
BirdLife International. 2015. Species factsheet: Terpsiphone atrocaudata. [online] URL:
http://www.birdlife.org/datazone/species/factsheet/22707151
Birds Korea. 2014. Status of birds 2014. Birds Korea, Busan, Republic of Korea. [online] URL:
http://www.birdskorea.org/Habitats/Yellow-Sea/YSBR/Downloads/Birds-Korea-Status-of-Birds-
2014.pdf
Brazil, M. 2009. Birds of East Asia: China, Taiwan, Korea, Japan, and Russia. Princeton University
Press, Princeton, New Jersey, USA.
Brazil, M. A. 1991. The birds of Japan. Christopher Helm, London, UK.
Brazil, M. A. 2013. The nature of Japan: From dancing cranes to flying fish. Japan Nature Guides,
Japan.
Coates, B. J., G. C. L. Dutson, C. E. Filardi, P. Clement, P. A. Gregory, and C. W. Moeliker. 2006.
Family Monarchidae (monarch-flycatchers). Pages 244-329 in J. del Hoyo, A. Elliott, and D. Christie,
editors. Handbook of the birds of the world. Vol. 11. Old World flycatchers to Old World warblers.
Lynx Edicions, Barcelona.
Fang, W.-h. 2005. A guide to threatened birds of Taiwan. Taipei, Taiwan.
Gonzalez, J. C. T., L. E. Afuang, and A. V. Lacaste. 2008. Identifying conservation priorities for
terrestrial vertebrate fauna in the Batanes Islands, northern Philippines. Journal of Nature Studies 7:1-8.
Higuchi, H., and E. Morishita. 1999. Population declines of tropical migratory birds in Japan. Actinia
12:51-59.
Hirano, T. 1996. Changes in breeding avifauna during the past 25 years at Tomatsuriyama in
Utsunomiya City, central Japan. Strix 14:25-31.

5
Kim, C.-H., J.-H. Kang, Y. Lee, D.-W. Kim, J.-H. Suh, and M. Kim. 2010a. Distribution of the
endangered birds species in South Korea. Korean Journal of Ornithology 17:67-137.
Kim, Y.-H., W.-B. Kim, and H.-S. Oh. 2010b. The distribution of black paradise flycatcher on Jeju
Island and management. Korean Journal of Ornithology 18:141-148.
Kim, Y.-H., H.-S. Oh, Y.-C. Jang, and S.-S. Choi. 2010c. Nest environment selection of black paradise
flycatcher (Terpsiphone atrocaudata). Korean Journal of Ornithology 17:11-19.
Lin, R.-S., Y.-J. Lu, C.-H. Yang, T.-J. Tseng, C.-J. Ko, and W.-J. Chen. 2016. The red list of birds of
Taiwan, 2016. Endemic Species Research Institute and Forestry Bureau, Council of Agriculture,
Executive Yuan, Nantou, Taiwan. [online] URL:
https://www.researchgate.net/publication/312045198_The_Red_List_of_Birds_of_Taiwan_2016
NIBR. 2011. Red data book of endangered birds in Korea. Published in Korean (with English
summaries). National Institute of Biological Resources, Incheon, Republic of Korea.
NIBR. 2012. Endemic species of Korea. National Institute of Biological Resources, Incheon, Republic
of Korea.
Severinghaus, L. L., T.-S. Ding, W.-H. Fang, W.-H. Lin, M.-C. Tsai, and C.-W. Yen. 2010. The
avifauna of Taiwan (in Chinese). Forestry Bureau of Council of Agriculture of Executive Yuan, Taipei,
Taiwan.
Severinghaus, L. L., T.-S. Ding, W.-H. Fang, W.-H. Lin, M.-C. Tsai, and C.-W. Yen. 2017. The
avifauna of Taiwan (in English). Forestry Bureau, Council of Agriculture, Taipei, Taiwan. [online]
URL: http://conservation.forest.gov.tw/0001888
Sugimura, K., F. Yamada, and A. Miyamoto. 2003. Population trend, habitat change and conservation of
the unique wildlife species on Amami Island, Japan. Global Environmental Research 7:79-89.
Uchida, H. 1996. Black paradise flycatchers declined drastically in Higashimatsuyama city (in
Japanese). Yacho 591:13.

6
WCAT. 2013. Wildlife Conservation Act of Taiwan. [online] URL:
http://law.coa.gov.tw/GLRSnewsout/EngLawContent.aspx?Type=E&id=146
Yamamoto, Y., and N. Seto. 1997. Decrease of summer visiting birds in Yamaguchi Prefecture analyzed
from records of regular birding events. Strix 15:15-23.

7
Supplementary Tables
Table A1.1. List of 22 environmental data layers (or variables) used to build distribution models of
the Japanese Paradise-Flycatcher. The abbreviations used for the data layers and their data sources
are: DTM = digital elevation model; LCT = land cover type; NDVI = normalized differenced
vegetation index; SI = solar irradiation; std. dev. = standard deviation (for more details, see
Methods). Continuous or categorical values for each grid cell were calculated as follows: (1) Means
and standard deviations were calculated using only values of the original data source found within
each grid cell; (2) percentage coverage was calculated as the percentage of the area of the entire grid
cell covered by the respective land cover type; (3) the dominant land cover type of each grid cell was
the land-cover type with the highest percentage of coverage within each grid cell. The nine data
layers which we used to construct our final models are marked with an asterisk (*); see Table 1.
$This data layer was generated for 1948 and 2006 (see Methods for details).
Number
Data layer
Data source
Type
Calculation of values
1
NDVI mean*
FORMOSAT-2
continuous
1
2
NDVI std. dev.
FORMOSAT-2
continuous
1
3
elevation mean*
DTM
continuous
1
4
elevation std. dev.*
DTM
continuous
1
5
% aspect flat
DTM
continuous
2
6
% aspect north
DTM
continuous
2
7
% aspect south*
DTM
continuous
2
8
% aspect east
DTM
continuous
2
9
% aspect west
DTM
continuous
2
10
slope mean
DTM
continuous
1
11
slope std. dev.
DTM
continuous
1
12
SI mean
DTM
continuous
1
13
SI std. dev.
DTM
continuous
1
14
% road
LCT
continuous
2
15
% bare coastal land*
LCT
continuous
2

8
16
% built-over land
LCT
continuous
2
17
% farm/grassland*
LCT
continuous
2
18
% hilltop shrub*
LCT
continuous
2
19
% stunted forest
LCT
continuous
2
20
% tall forest*
LCT
continuous
2
21
% other types
LCT
continuous
2
22
dominant land cover type*$
LCT
categorical
3

9
Table A1.2. Mean ± standard deviation (range in brackets) for all environmental data layers calculated across all grid cells (n = 4632), presence grid
cells (n = 120) and absence grid cells (n = 104). For abbreviations, see Table A1.1. Not included is the dominant land cover type because it is a
categorical variable.
Variable
Unit
All grid cells
Presence grid cells
Absence grid cells
NDVI mean
unitless
0.65 ± 0.20 (-0.14 - 0.78)
0.74 ± 0.03 (0.62 - 0.78)
0.67 ± 0.10 (0.35 - 0.77)
NDVI std. dev.
unitless
0.05 ± 0.05 (0.00 - 0.33)
0.02 ± 0.02 (0.01 - 0.14)
0.06 ± 0.05 (0.01 - 0.21)
Elevation mean
m
173.5 ± 140.2 (0.0 - 524.4)
171.9 ± 93.8 (15.9 - 398.5)
133.2 ± 137.4 (3.7 - 517.0)
Elevation std. dev.
m
11.9 ± 8.2 (0.0 - 47.5)
13.2 ± 6.2 (2.4 - 34.8)
10.0 ± 8.3 (0.5 - 39.6)
% aspect flat
%
9.4 ± 20.1 (0.0 - 100.0)
2.7 ± 5.9 (0.0 - 29.5)
11.3 ± 19.2 (0.0 - 85.1)
% aspect north
%
22.8 ± 28.9 (0.0 - 100.0)
21.3 ± 27.4 (0.0 - 99.5)
27.4 ± 27.9 (0.0 - 100.0)
% aspect south
%
25.1 ± 31.1 (0.0 - 100.0)
31.0 ± 32.1 (0.0 - 100.0)
18.8 ± 29.0 (0.0 - 100.0)
% aspect east
%
21.7 ± 29.0 (0.0 - 100.0)
24.4 ± 29.1 (0.0 - 100.0)
23.9 ± 28.5 (0.0 - 100.0)
% aspect west
%
20.8 ± 28.1 (0.0 - 100.0)
20.6 ± 26.7 (0.0 - 99.5)
18.6 ± 27.7 (0.0 - 99.9)
Slope mean
°
22.3 ± 12.4 (0.0 - 57.2)
25.7 ± 8.9 (5.3 - 46.6)
18.7 ± 12.4 (0.9 - 47.2)
Slope std. dev.
°
6.5 ± 3.3 (0.0 - 27.4)
7.2 ± 2.5 (1.7 - 14.1)
6.8 ± 3.9 (1.7 - 25.7)
SI mean
10³kWh/m²
1484.6 ± 180.1 (0.0 - 1739.9)
1469.6 ± 147.7 (920.8 - 1676.6)
1495.1 ± 170.5 (838.3 - 1708.6)
SI std. dev.
10³kWh/m²
81.6 ± 58.5 (0.0 - 492.9)
93.5 ± 52.0 (13.3 - 261.1)
86.6 ± 73.1 (4.6 - 331.5)
% roads
%
0.7 ± 2.2 (0.0 - 19.1)
0.2 ± 1.3 (0.0 - 11.4)
1.8 ± 3.2 (0.0 - 14.3)
% bare coastal land
%
11.2 ± 29.5 (0.0 - 100.0)
0.0 ± 0.0 (0.0 0.0)
3.8 ± 15.6 (0.0 - 99.0)
% built-over land
%
1.6 ± 9.5 (0.0 - 100.0)
0.1 ± 0.7 (0.0 - 7.3)
0.5 ± 3.6 (0.0 - 35.0)
% farm/grassland
%
12.5 ± 25.4 (0.0 - 100.0)
5.8 ± 14.6 (0.0 - 79.6)
27.9 ± 32.4 (0.0 - 100.0)
% hilltop shrub
%
7.3 ± 21.2 (0.0 - 100.0)
4.7 ± 17.6 (0.0 - 97.0)
6.8 ± 19.2 (0.0 - 89.0
% stunted forest
%
56.1 ± 38.6 (0.0 - 100.0)
63.3 ± 33.1 (0.0 - 100.0)
52.7 ± 34.6 (0.0 - 100.0)

10
% tall forest
%
9.6 ± 20.7 (0.0 - 100.0)
25.6 ± 31.1 (0.0 - 100.0)
4.5 ± 15.6 (0.0 - 99.7)
% other types
%
0.5 ± 3.4 (0.0 - 85.8)
0.4 ± 1.6 (0.0 - 11.9)
2.1 ± 9.4 (0.0 - 82.6)

11
Table A1.3. Model set for modeling the ecological niche of the Japanese Paradise-Flycatcher on Lanyu
Island, Taiwan, containing different combinations of the number of background points (BPs), coefficient
features (L: linear; Q: quadratic; L + Q: linear + quadratic) and regularization constants (0.25, 0.50,
0.75, 1.00). AICc, ΔAICc and Wi are the Akaike’s Information Criterion corrected for small sample
sizes, the delta value, the Akaike weight and evidence ratio as defined by Burnham and Anderson
(2002). Only models with ΔAICc < 2 marked with an asterisk (*) were included to build our ensemble
model (see Methods).
Model
AICc
ΔAICc
Wi
Evidence ratio
500 BPs
L (0.25)
399.79
8.73
0.00207
78.78
L (0.50)*
392.30
1.24
0.08768
1.86
L (0.75)
395.69
4.63
0.01610
10.13
L (1.00)*
392.12
1.06
0.09605
1.70
Q (0.25)
397.78
6.72
0.00566
28.83
Q (0.50)
395.69
4.63
0.01613
10.11
Q (0.75)
414.51
23.45
0.000001
123382
Q (1.00)
403.64
12.58
0.00030
540.23
L+Q (0.25)
396.27
5.21
0.01203
13.55
L+Q (0.50)
397.43
6.37
0.00674
24.21
L+Q (0.75)
400.51
9.45
0.00145
112.64
L+Q (1.00)
396.48
5.42
0.01083
15.06
1000 BPs
L (0.25)
401.11
10.05
0.00107
151.98

12
L (0.50)*
391.63
0.57
0.12246
1.33
L (0.75)
399.63
8.57
0.00224
72.67
L (1.00)
402.13
11.07
0.00064
254.04
Q (0.25)
418.43
27.37
0.0000002
875465.30
Q (0.50)
407.24
16.18
0.00005
3259.27
Q (0.75)
409.94
18.88
0.00001
12598.76
Q (1.00)
393.56
2.50
0.04663
3.50
L+Q (0.25)
400.86
9.80
0.00122
134.14
L+Q (0.50)
393.20
2.14
0.05600
2.91
L+Q (0.75)
395.01
3.95
0.02262
7.21
L+Q (1.00)
393.39
2.33
0.05080
3.21
2500 BPs
L (0.25)
405.62
14.56
0.00011
1451.64
L (0.50)
399.76
8.70
0.00210
77.58
L (0.75)
395.70
4.64
0.01603
10.17
L (1.00)
398.52
7.46
0.00392
41.60
Q (0.25)
394.44
3.38
0.03014
5.41
Q (0.50)
403.48
12.42
0.00033
498.29
Q (0.75)
396.17
5.11
0.01265
12.89
Q (1.00)
393.11
2.05
0.05840
2.79
L+Q (0.25)
400.24
9.18
0.00165
98.69
L+Q (0.50)
399.45
8.39
0.00245
66.51
L+Q (0.75)
411.87
20.81
0.000005
33086.25

13
L+Q (1.00)*
391.06
0.00
0.16309
1.00
5000 BPs
L (0.25)
401.80
10.74
0.00076
215.29
L (0.50)
414.37
23.31
0.000001
115380.70
L (0.75)
393.94
2.88
0.03864
4.22
L (1.00)
424.84
33.78
0.00000
21681906.00
Q (0.25)
395.02
3.96
0.02254
7.24
Q (0.50)
403.16
12.10
0.00039
423.26
Q (0.75)
408.89
17.83
0.00002
7454.00
Q (1.00)*
392.89
1.83
0.06540
2.49
L+Q (0.25)
401.25
10.19
0.00100
163.45
L+Q (0.50)
395.51
4.45
0.01765
9.24
L+Q (0.75)
398.50
7.44
0.00395
41.24
L+Q (1.00)
416.76
25.70
0.0000004
380024.60

14
Supplementary figures
Fig. A1.1. Gridded maps of the important environmental data layers of Lanyu used in the distribution
modelling of the Japanese Paradise-Flycatcher. This map depicts elevation mean; the values (cf. Table
A1.2) are depicted in two colours from lowest (blue) to highest (red) in 10 equal intervals, and the 100m
x 100m grid cells are bounded by grey lines.

15
Fig. A1.2. Gridded maps of the important environmental data layers of Lanyu used in the distribution
modelling of the Japanese Paradise-Flycatcher. This map depicts NDVI mean; the values (cf. Table
A1.2) are depicted in two colours from lowest (blue) to highest (red) in 10 equal intervals, and the 100m
x 100m grid cells are bounded by grey lines.

16
Fig. A1.3. Gridded maps of the important environmental data layers of Lanyu used in the distribution
modelling of the Japanese Paradise-Flycatcher. This map depicts % tall forest; the values (cf. Table
A1.2) are depicted in two colours from lowest (blue) to highest (red) in 10 equal intervals, and the 100m
x 100m grid cells are bounded by grey lines.

17
Fig. A1.4. Gridded maps of the important environmental data layers of Lanyu used in the distribution
modelling of the Japanese Paradise-Flycatcher. This map depicts dominant land cover type: the colours
refer to roads (black), bare coastal lands (greenish brown), other built-over land (yellow), farmland or
grassland (grey), hilltop shrub (greenish yellow), stunted forest (orange beige), and tall forest (light
beige). The category ‘other types’ is not depicted because it covered a very small area (Table A1.2).

18
Fig. A1.5. Frequency distribution depicting the variation in percentage of stunted forest associated with
the 120 presence grid cells for the Japanese Paradise-Flycatcher. Note that the right peak in this Figure
largely corresponds to the left peak in Fig. 3B.
0
5
10
15
20
25
30
35
40
45
Frequency
010 20 30 40 50 60 70 80 90 100
% stunted forest

19
Fig. A1.6. (A) Forest present in 2006 (dark green) and forest lost between 1948 and 2006 (yellow).
Grey areas are non-forest areas (i.e., not categorized as either hilltop shrub, stunted forest, or tall forest).

20
Fig. A1.7. Forest present in 1948 (dark green) and forest gained between 1948 and 2006 (yellow). Grey
areas are non-forest areas (i.e., not categorized as either hilltop shrub, stunted forest, or tall forest).

21
Fig. A1.8. Binary distribution model of Japanese Paradise-Flycatcher as depicted in Fig. 5 of main text,
but here, the predicted presence grid cells are overlaid with the forest areas gained between 1948 to 2006
(yellow areas in Fig. A1.7).