INTRODUCTION
Phenology, the timing of biological events within seasonal cycles, is a
vital aspect of the life history of all species (Schwartz 2013).
Knowledge of a species’ phenology is necessary to fully understand its
role within the ecosystem. In the case of endangered species,
information on phenology can help target conservation efforts during the
season when potential payoff is highest (e.g. Perkins et al. 2013).
Shifts in phenology are among the most widely documented biotic
responses to global climate change (Scheffers et al. 2016, Oliver et al.
2018), often leading to ecologically disruptive asynchronies. Phenology
is central to both understanding the ecology of ecosystems and to the
conservation of species. Reproductive phenology is a basic aspect of
birds’ life history which is well studied in temperate regions, but is a
comparatively younger field of research in the tropics (Abernethy 2018).
Phenological studies of Southeast Asian birds are especially scarce
(Sodhi 2002). The paucity of studies on tropical seasonality is
partially due to historical inequities in the focus of scientific
research, and partially due to the comparative difficulty –
phenological patterns in tropical regions are often less obvious than in
temperate ones (Stutchbury and Morton 2001).
While reproductive seasonality among birds is generally understood to be
the norm, even near the equator (Baker 1939, Snow and Snow 1964, Bell
1982), seasonal fluctuations in food supply are less pronounced in
equatorial regions, leading to greater variability among species in the
timing of the breeding season (Stutchbury and Morton 2001, Stouffer et
al. 2013). Individual species might breed outside the typical season due
to competition for nesting sites (Steward et al. 2013, Sadanandan et al.
2023), particular diets (Ralph and Fancy 1994), or taxon-specific
requirements (Serle 1981). Phenology can sometimes shift dramatically
across small geographic scales due to differences in climate (Wrege and
Emlen 1991, Thomas et al 2001, Moore et al. 2005). As anthropogenic
climate change intensifies, it is becoming increasingly important to
understand phenological cycles of biota in the tropics, where
biodiversity is greatest and under greatest threat, how those cycles are
linked to local climates, and how they might be disrupted as the climate
changes. Moreover, some avian clades may be more resilient to warming
than others (Pollock et al 2021). More detailed research is needed to
understand the phenology of tropical birds on a species-specific level
and at a detailed geographical scale.
Long term soundscape recordings can be used to determine the breeding
phenology of birds (Brumm and Zollinger 2013, Jahn et al. 2017,
Pérez-Granados and Schuchmann 2020). New bioacoustic technologies have
made it possible to explore questions which were difficult or impossible
to address with traditional methods (Pijanowski et al. 2011, Shonfield
and Bayne 2017). Autonomous recording units (ARUs), which can be
deployed outdoors to record long-term continuous soundscape data, have
become increasingly reliable, user friendly, and inexpensive in recent
years, making the widespread use of this technology more feasible for
large scale projects and more common as an aspect of long-term
ecological monitoring regimens (Hill et al. 2018, Manzano-Rubio et al.
2022, Bota et al. 2023). Long-term autonomous bioacoustic monitoring has
many advantages over traditional survey methods: it is less labor
intensive, less expensive, less prone to observer bias, can be used at
difficult-to-access locations, and creates an archival record which
enables the comparison of vocalizations and soundscapes across time
(Frommolt et al. 2008, Borker et al. 2015, Shonfield and Bayne 2017).
While acoustic monitoring is becoming more common, continuous multi-year
datasets are still rare. This study makes use of an exceptionally large
and complete soundscape dataset: two full years of continuous 24/7
soundscape recordings from four ARUs deployed in the forests of
Singapore, a small Southeast Asian country only 1˚ north of the equator.
Recent advances in machine learning technology have made it possible to
efficiently process the terabytes of data produced by each monitoring
station over the years (Stowell et al. 2019). Long term acoustic
monitoring in combination with machine learning analysis has a wide
utility in ecological research: it can be used for assessment of
population dynamics, activity patterns, and human impacts on a site over
time, for rapid site inventories, to detect and map the habitats of rare
and endangered species, and to determine the phenological cycles of
individual species (Bardeli et al. 2010, Shonfield and Bayne 2017,
Deichmann et al. 2018). Software like Kaleidoscope Pro (Wildlife
Acoustics) and BirdNET (Cornell Lab of Ornithology) are able to
automatically detect and identify species’ vocalizations with a high
degree of accuracy (Manzano-Rubio et al. 2022) and can be trained to
identify new species.
The objectives of this study were to elucidate the species-specific
breeding seasons of some of the birds of Singapore. To meet this
objective, we used two years of continuous soundscape recordings and
nine tailor-made species-specific machine learning classifiers to
determine the seasonal vocal activity of nine focal species. There were
two emergent properties in the preliminary results which warranted
further analysis. First, several species did not appear to follow any
seasonal pattern at all in their vocal activity. To quantitatively
measure this pattern, we developed a novel seasonality index which
measures the extent to which vocal activity is seasonal or aseasonal.
While a number of excellent R functions exist for detecting and
extracting the seasonal component of a trend (e.g. decompose), none
measure the strength of seasonality. Our novel seasonality index fills
this gap. Second, the least seasonal species appeared to be those which
had not historically occurred at the study site. To quantitatively
assess this apparent pattern, focal species were categorizedpost-hoc into native forest species versus parkland colonizer
species and the degree of seasonality of each species’ vocal activity
was measured.