Maximizing detection probability for effective large‐scale nocturnal bird monitoring

Our specific objectives were to (a) estimate detection probabilities of nocturnal bird species, after taking into account survey‐specific covariates, and (b) investigate the influence of site‐specific covariates on owl and nightjar abundance, integrating effects of imperfect detection.


| INTRODUC TI ON
Long-term and large-scale ecological monitoring programmes of secretive or elusive species are difficult and costly (Thompson, 2013).
Datasets about distribution and abundance of a species, among others, are required to adequately develop management actions (Petitot, Manceau, Geniez, & Besnard, 2014). Managers and researchers expect to reach a profitable trade-off between available budgets (personnel, time and money) and results. Currently, ecologists and conservation practitioners hold a range of relatively "low cost" established and emerging technologies that can be used to increase the spatial and temporal scales at which they work (Marvin et al., 2016). Species distribution models (SDMs) and occupancy models are powerful tools to evaluate distribution, abundance and population trends (Kellner & Swihart, 2014). SDMs use a range of datasets collected through the time and the space with a high variety of positive records of a target species (Bradsworth, White, Isaac, & Cooke, 2017;Girini, Palacio, & Zelaya, 2017;Sarà, 2008), and they have become an important research tool to inform decision-making in conservation (Sofaer et al., 2019). Unfortunately, detection is rarely either perfect or constant due to observer error, environmental conditions and species rarity (Banks-Leite et al., 2014). It may result in bias in estimated relationships with ecological covariates and estimates of species distribution or abundance that are inaccurate or mask trends (Kellner & Swihart, 2014). Occupancy models use detection/non-detection data also collected using different methods from passive detectors such as remote cameras or acoustic recording devices, to questionnaires and specific surveys (Martínez-Martí, Jiménez-Franco, Royle, Palazón, & Calvo, 2016;Moeller, Lukacs, & Horne, 2018). However, failure to detect a species where it is present (imperfect detection) is a common source of error in the predictive ability of SDMs, which can have serious implications for the effectiveness of applications that rely on their predictions (Lahoz-Monfort, Guillera-Arroita, & Wintle, 2014). Occupancy models provide better accuracy than SDMs, although it does not always lead to a substantial improvement to predict the distribution of poorly detectable species (Comte & Grenouillet, 2013). Specific analytical methods are nevertheless required to obtain reliable benefits of these datasets (Jiménez-Franco et al., 2019).
Typically, relatively too much time and money are spent on data collection and not enough consideration is given to programme development, data management and analysis, interpretation, and reporting (Caughlan & Oakley, 2001). Lack of adequate programme development based on behavioural studies causes significant changes in the detection probability of the target species (Nijman, 2007). Many studies employ data from multiple sources, often relying on volunteer fieldwork, without a thorough understanding or critical evaluation of the influence of the background data quality, and its subsequent analytical transformation, on the research conclusions (Blanco et al., 2012). However, to date, it is widely assumed that data collection should be organized within a robust and statistically valid scheme, which optimally corrects for imperfect detection (Strebel, Schmid, Kéry, Sattler, & Knaus, 2020).
Nocturnal birds, especially owls, have drawn scientific attention to obtain a valuable method to measure their population size and trends. The conservation status and recent trends of most nocturnal species necessarily require the application of suitable monitoring programmes (BirdLife International, 2015). However, owls and nightjars are secretive species difficult to study due to their nocturnal behaviour and particular biology, which greatly differ among species (Zuberogoitia, Martínez, & Alonso, 2011). In addition, their detection probability is usually highly dependent on the skill of the observer, the period of the year, the weather and other factors (Jiguet & Williamson, 2010;Zuberogoitia et al., 2019). Although large-scale monitoring programmes are generally impractical, expensive and time-consuming, researchers have been developing specific methods to survey owls in different conditions (see, e.g., Bradsworth et al., 2017;Fröhlich & Ciach, 2017;Ibarra, Martin, Drever, & Vergara, 2014). However, large-scale censuses are still scarce and only a few cases consider the nocturnal bird community (Ibarra et al., 2014;Yahya, Puan, Azhar, Atikah, & Ghazali, 2016;Zuberogoitia & Campos, 1998).
Moreover, large-scale surveys of nocturnal species are affected by detection rates, which are typically low (Kissling, Lewis, & Pendleton, 2010;Wingert & Benson, 2018). Failure to account for detection rates in occupancy surveys can incorrectly identify occupied sites as vacant (MacKenzie et al., 2006;Stolen et al., 2019).
Ideally, one would dedicate a sufficient survey effort to ensure that detection is perfect (Lahoz-Monfort et al., 2014). Prior knowledge of the target species has a very strong positive influence on detection probability (Brubaker, Kovach, Ducey, Jakubas, & O'brien, 2014).
Knowledge of those factors affecting detection probability could also positively affect the results of models and should be considered to design the most robust survey possible (Kissling et al., 2010).
Thus, recording data in ways that allow the modelling of the detection process should be a standard practice in future surveys (Lahoz-Monfort et al., 2014). It would be desirable to calibrate our estimates of the species' distribution and population, not only for population trend analysis but also for national breeding atlas and even wildlifehabitat studies (Freeman, Balmer, & Crick, 2006;Martínez-Martí et al., 2016).
This work was conceived to develop a valuable method easy to replicate and robust enough to evaluate population trends of owls and nightjars. During the last two decades, we have developed methodological works focused on specific owl surveys in the study area (e.g. Zuberogoitia et al., 2019;. We took advantage of this knowledge to programme a largescale survey that should be conducted through one breeding season. Therefore, we designed this programme considering limiting factors, survey-specific covariates that may affect detection probability which, in turn, affect abundance models (see Lahoz-Monfort et al., 2014). Our specific objectives were to (a) estimate probabilities of detection of each species, after taking into account survey-specific covariates, and (b) investigate the influence of site-specific covariates on owl and nightjar abundance, integrating effects of imperfect detection.

| Study area
This study was carried out in the Basque Country, northern Spain (7,234 km 2 , lying between 42° and 43°N and 1° and 3°W; Figure 1).
There are two clearly defined areas (roughly north and south). The northern area, Cantabric region, runs along the coast of the Bay of Biscay. It has an Atlantic climate and mild temperatures with a thermic oscillation of 12°C from the coldest to the hottest months and 1,200-2,000 mm of rainfall distributed throughout the year (www.euska lmet. euska di.net). The landscape is mountainous and densely populated, with extensive urban and industrial areas, mainly located in valley floors and on the gentler slopes. Forestry plantations (Pinus radiata and Eucalyptus spp.) have become widespread in the last 80 years, gradually replacing grazing land for extensively reared livestock, traditional agricultural activities, as well as a few remnants of native forest. The other area, of some 2,500 km 2 , lies to the south and is situated in a transition area to the Mediterranean climatic region. The climate is dry with a thermic oscillation of 17°C from the coldest to the hottest months, and the landscape is dominated by arable lands, vineyards, Mediterranean scrub and holm-oak woods in the sloping areas.

| Survey design and survey protocol
Survey methods were based on the methodology used in the little owl (Athene noctua) census carried out in the Basque Country (Zuberogoitia, Zabala, & Martínez, 2011), in which we conducted a large-scale detection/non-detection survey encompassing all the Basque Country. However, in the present case, we considered all the owl species and also European nightjars (Caprimulgus europaeus).
First, we randomly selected 65 5 × 5 km Universal Transverse Mercator (UTM) squares that represented all the vegetation types of the Basque Country. These UTMs were considered our sampling units (SUs). Second, we randomly selected eight survey points (SPs) in each SU. The SPs were chosen according to the main habitat types at each SU, considering a minimum distance of 1 km between two SPs. All the SPs were established before the beginning of the surveys and were kept without change until the end of the study. In all, we surveyed 521 SPs which will be considered the sampling sites for our hierarchical models ( Figure 1).
We considered seven survey periods in 2018 (one per month through January to July, most of the breeding cycle of the target species; . In each period, F I G U R E 1 Land cover and survey points of the study area, Basque Country We noted every owl species and nightjars in each survey, considering those that were detected in the first 5-min spontaneous, during the 5-min broadcast period and during the last 5-min silent period. We broadcast playback voices of only one species per survey period. The species were chosen according to the annual maximum peak of response to playback broadcasts in our study area. In this sense, we broadcast voices of eagle owls (Bubo bubo) in January , tawny owls (Strix aluco) in February (Zuberogoitia et al., 2019;Zuberogoitia & Martínez, 2000), long-eared owls (Asio otus) in March (Martínez, Zuberogoitia, Colás, & Macía, 2002), barn owls (Tyto alba) in April (Zuberogoitia & Campos, 1998), Eurasian scops owls (Otus scops) in the last weeks of April and first weeks of May  and little owls during June and the first half of July (Zuberogoitia, Zabala, et al., 2011). There were no records of boreal owls (Aegolius funereus) in the study area, although we also included survey points with broadcast voices of this species in mountain old forest areas in January (Badosa, López, Potrony, Bonada, & Gil, 2012

TA B L E 2
Overall number of individuals of each species detected per survey period (month). Results obtained during specific broadcasting surveys in bold observer at 300 m but not as much as to produce distortion noises at close distances.

| Variables for the analyses
The number of individuals (abundance, from 0 to a maximum of 9 individuals) of each species detected per SP was considered the response variable, a zero-inflated Poisson random variable (Table 1).
We recorded two types of predictive variables, those that could affect species detectability (survey covariates) and those that could affect our ecological response variables, that is species abundance (site covariates).
We chose five survey covariates that could affect detectability.
We noted whether individuals responded to conspecific broadcast voices or produced spontaneous calls (BROAD). We also considered the experience surveying owls of each observer (EXPER), being "0" for those researchers who had no experience surveying owls and "1" for those researchers that had developed owl surveys (see, e.g., Zabala et al., 2006;Zuberogoitia, Zabala, et al., 2011). We considered linear and quadratic effects of the Julian days (1 January = 1, 31 December = 365; DATE, DATE ˄2 ) to control for seasonal effects.
Surveys should have been conducted on dry and calm nights, although it was impossible to effectively control some parameters as wind (WIND) and temperature (T), and therefore, we obtained detailed information about the wind speed (km/hr) and T (°C) of the 15-min survey period at each SP. We obtained these data from the nearest meteorological stations (n = 27) to each SP (http://www. euska lmet.euska di.eus).
We also selected 11 site-specific covariates that could affect species abundance, the regional climatic situation (REG) of the SP, being Cantabric region (1), Subcantabric region (2)

| Data analysis
We developed binomial N-mixture occupancy models, in which we estimated abundance and detectability as a function of site-specific and survey-specific covariates using the log link function (Fiske & Chandler, 2011). Our sampling design considered 521 sites (SPs) TA B L E 3 Untransformed coefficients and standard error of survey-specific covariates related to detection probability (p) estimated by the top-ranked models of owls and nightjars surveys of the Basque Country. Eagle owl, short-eared owl and boreal owl were not considered in the models due to the scarce numbers of positive records. Covariates: date (DATE), quadratic function of date (DATE ˄2 ), survey hour (HOUR), wind speed (WIND), temperature (T), broadcast voices (BROAD), observer experience (EXPER). Blank cells refer to those covariates that were not selected in the most parsimonious models were generated by a combination of (a) a state process determining abundance (i.e. counts) at each site and (b) a detection process that yields observations conditional on the state process (Kéry, Royle, & Schmid, 2005;MacKenzie et al., 2002;Royle, 2004).
Given the large number of potential candidate models to evaluate abundance and detection probabilities, model fitting was con-  & Anderson, 2002). We used the pcount function from the unmarked package (Fiske & Chandler, 2011), considering latent abundance as a Poisson distribution, and detectability as binomial distribution, as well as the identifiability problems described by Kéry (2018). We generated abundance maps of the species in R using the package raster (https://CRAN.R-proje ct.org/packa ge=raster).

| RE SULTS
Overall, we conducted 2,584 surveys in 521 SPs from January to July 2018 (Zuberogoitia et al., 2020). Tawny owl was the most abundant and widely distributed owl species in the study area, distantly followed by barn owl, Eurasian scops owl and little owl, whereas longeared owl and eagle owl were scarce and patchily distributed, and short-eared owl and boreal owl were rare species in the study area (Tables 1 and 2). European nightjar appeared in 31.1% of the SPs.

| Factors affecting detection probability
Detection probability was strongly influenced by playback broadcast for all owl species modelled (Table 3; Appendix Table S1.1). In fact, surveys using broadcast voices considerably improved detection probability (Figure 2). Observer experience was another covariate affecting detection probability for all studied species except long-eared owl (Table 3), slightly increasing detectability, especially in little owl (Figure 3).
Date negatively affected tawny owl detectability, and positively, though weakly, affected barn owls, Eurasian scops owl and nightjars ( Figure 4). These two last species are migratory species that arrive in spring for breeding. The survey hour also affected detectability in all species but Eurasian scops owls (Table 3). Vocal activity diminished proportionally to the hour after sunset, the first hour after dusk being the best for surveying owls and nightjars ( Figure 5).
Finally, wind negatively affected detectability of tawny owls and little owls but had no effects for the other species ( Figure 6).
Temperature slightly affected European nightjars.

| Factors affecting abundance
Tawny owls were more abundant in mountain forest areas (mainly pine timber forests) of the northern area (Cantabric region) of the Basque Country (Figure 7), whereas large urban and agricultural areas, mainly those located in the south (Mediterranean region), negatively affected this species abundance (Table 4;  Barn owls avoided forest areas, but their abundance was also affected by altitude, large urban and agricultural areas and scrub areas TA B L E 4 Untransformed coefficients of site-specific covariates related to abundance estimated by the top-ranked models of owls and nightjars surveys of Basque Country. Scarce species were not considered in the analysis. Covariates: regional climatic situation (REG), altitude (ALT) grass-fields (FIELD), agriculture area (AGR), urban area (URB), quadratic function of urban area (URB ˄2 ), pine plantations (PIN), scrub and heather areas (SCRUB), deciduous forest (DEC), eucalyptus plantations (EUC) and forest surface (FOR). For open areas species, barn owl and little owl, we joined forest types (DEC, HOLM, PINE and EUC) in a unique covariate (FOR), whereas for the rest of species, we considered the forest types in the analysis. NA refers to those covariates that were not included in the analysis. Blank cells refer to those covariates that were not selected in the most parsimonious models   Figure S2.1.c). Little owls avoided mountainous areas, but in contrast to barn owl, their abundance was favoured by agricultural and scrub areas and showed a negative quadratic effect of urban areas (villages and medium-sized towns, Appendix S2, Figure S2.

1.d). Eurasian scops owls selected open areas, from
Atlantic fields to agricultural areas and scrub areas, rejecting deciduous forests. Moreover, Eurasian scops owls were located in small rural areas and parks in the middle of some cities, large urban areas, explaining a weak quadratic effect of urban areas (Table 4, Appendix S2, Figure S2.1.e).
European nightjar abundance was highest in scrub areas and affected positively but weakly by eucalyptus patches, whereas it was negatively affected by large forest areas (pine and deciduous forests, Table 4, Appendix S2, Figure S2.1.f).

| D ISCUSS I ON
Twenty years ago, we developed an intensive large-scale census of owls in Biscay, included in the current study area (Zuberogoitia & Campos, 1998), which was one of the first large surveys of the complete owl community in large areas. To date, we increased the study area to the whole Basque Country (more than threefold of the previous extent), but we could not apply the same intensive methodology, because of its excessive cost. Instead, we developed a survey protocol to obtain valuable information of nocturnal birds in a relatively short period (seven months) at a lower cost. The difference nevertheless is that Zuberogoitia and Campos (1998) obtained the population size of every owl species, and now, our results are expressed as detectability and abundance. These parameters, however, allow us to apply for monitoring programmes in a cost-effectively way, accounting for imperfect detection (Martínez-Martí et al., 2016), and to establish trends of populations using the same methods in future surveys (Jiménez-Franco et al., 2019;MacKenzie et al., 2006).

| Detection probability
Among the covariates affecting detection probability, the use of playback broadcast voices was the most obvious, improving results for all species as expected according to our previous experience  and also other works with these and other owl species (Braga & Motta-Junior, 2009;Cooke et al., 2017;Kissling et al., 2010;Mori, Menchetti, & Ferretti, 2014;Wingert & Benson, 2018 (Barata, Griffiths, & Ridout, 2017).
Knowledge of the biological cycle or vocal behaviour of every species is needed to adjust the date of the survey programmes (Flesch & Steidl, 2007;Olsen, Trost, & Hayes, 2002). Our results confirmed differences in detection probability for most species through the seasons that were related to specific breeding cycles, from eagle and tawny owls, the earliest breeders, until little and Eurasian scops owls, the latest ones (León-Ortega, Jiménez-Franco, Martínez, & Calvo, 2017;Zuberogoitia, 2002). In some cases, for example little owls, vocal activity reaches a maximum peak in spring , but the best period to detect breeding territories is in June and July (Zuberogoitia, Zabala, et al., 2011).
Most of the owl and nightjar surveys are conducted during the first hours after sunset (Kissling et al., 2010;Raymond et al., 2019).

| Abundance
More than half of the study area is primarily covered by forest, both reforested for timber production (24.6% of pine and 2.6% of eucalyptus) and native deciduous forest (29.6%), whose extent has increased during the last decades, mainly in the medium and north of the study area, in detriment to open lands (http://www.nasdap. net/inven tario forestal). This is an ideal habitat for a generalist forest species, the tawny owl, that reaches maximum densities in pine plantations and oak forests fragmented with small grass-fields (Michel et al., 2016;Zuberogoitia, 2002), but reduces its densities in large homogeneous forest areas (Burgos & Zuberogoitia, 2018) and avoids young forests (Rumbutis et al., 2017). Tawny owl abundance showed a negative relation with open landscapes, mainly those agricultural landscapes located in the Mediterranean region and large urban areas. Tawny owls were also found in these types of habitats but at low abundance values. In fact, tawny owls show high flexibility to adapt to semi-arid landscapes, at the limits of its distribution range (Sánchez-Zapata & Calvo, 1999) and novel habitats too (Fröhlich & Ciach, 2019;Solonen, 2014).
Except for the long-eared owl, which is usually linked to agro-forestry systems and forest edges (Martínez & Zuberogoitia, 2004a), the rest of species we studied are not forest-dwelling ones. However, those habitat variables that favoured tawny owls negatively affected long-eared owls due to both differential habitat requirements and the effect of the intra-guild competition of tawny owls on longeared owls (Zuberogoitia, 2002;Zuberogoitia, Martínez, Zabala, & Martínez, 2005). Therefore, the abundance of this species increased in the Mediterranean region, mostly associated with mixed habitat conditions (Emin et al., 2018), whereas its abundance was low or even null in some favourable habitats (grass-fields and heathlands) in the Cantabric and Subcantabric regions. In addition, the potential breeding area of long-eared owls in the north half of the study area um-aged and older forest (Petty, 1996). Successional changes during a forest rotation provide different habitats and food resources for raptors to exploit, which favour some species, for example tawny owls, which obtain most of their food from clear-cuts, but also need older forest for roosting and breeding (Petty, 2011). However, intensive logging activities (i.e. clear-cuttings) alter successional states of vegetation and reduce open-land dwelling raptors (Tapia et al., 2017). In our study area, clear-cuts and pre-thicket sites do not constitute a suitable habitat for long-eared owls; on the contrary, the increase in availability and extension of these habitats seems to favour European nightjars. Our results, in accordance with previous work (Evens et al., 2017), show a negative effect of large forests on the abundance of nightjars. However, its abundance was high in scrubland areas, close to forests, where nightjars nest and forage (Sharps, Henderson, Conway, Armour-Chelu, & Dolman, 2015).
Therefore, the increase of early-seral stage and afforested areas showed a positive effect on the species in the study area, with most of the populations in the northern, forested area (Appendix S2. Figure S2.1). Nightjar abundance was slightly related to eucalyptus plantations as they are harvested on a rotation of 12-18 years, whereas pine rotation is close to 35 years; therefore, clear-cuts are available sooner in eucalyptus plantations than in pine forests. In general, these temporal scrublands are low-quality habitats for birds (Goded et al., 2019), but they are positively selected by nightjars as breeding sites, moving to open lands for foraging (Evens et al., 2017).
We expected the two open area specialist species (i.e. the barn owl and little owl), to be positively related to agro-pastoral areas, but our results did not support this anticipated relationship. On the one hand, the abundance of these two species decreased with altitude, as has been previously reported (Zuberogoitia, 2002). There were few records of these species in grasslands and heathers of the highlands.
This could be related to weather conditions in the Basque Mountains, where the maximum precipitation values for the whole study area (close to 2,000 L/m 2 ; www.Euska lmet.euska di.eus) are obtained.
Likewise, scrublands do not favour the foraging behaviour of barn owls (Arlettaz, Krähenbühl, Almasi, Roulin, & Schaub, 2010), and the species was scarce, or even absent, in heathlands and early-seral stages of deciduous forest and afforested areas, whereas little owl abundance was positively related to this type of habitat, although avoiding afforested areas. On the other hand, previous works in the study area showed a clear relationship between the two species and grasslands and agriculture areas (Aierbe, Olano, & Vázquez, 2001;Zabala et al., 2006;Zuberogoitia, 2002) as it occurs all through their range (Andersen, Sunde, Pellegrino, Loeschecke, & Pertoldi, 2017;Hindmarch, Krebs, Elliott, & Green, 2012;Taylor, 1994;Van Niewenhuyse, Génot, & Jonson, 2008). However, these habitats also suffered severe transformations during the last decades, being the most affected habitats for the urban increase and also partially affected by logging activities. Fragmentation and reduction of grasslands drove the extinction of isolated populations of both species (see Alonso, Caballero, Orejas, Sáez, & Yánez, 1999;Zabala et al., 2006;). An increase of road network and traffic along them can increase rates of barn owl-vehicle collisions (Regan et al., 2018), which also negatively affect population abundance, distribution and persistence (Borda-de-Água, Grilo, & Pereira, 2014;Grilo et al., 2012;Hindmarch et al., 2012;Silva et al., 2012).
Moreover, the negative response of barn owls to agricultural landscapes is a new problem that has also been detected through the species range in Spain (Escandell, 2012), and it is also affecting many species in Europe (Chrenková, Dobrý, & Sálek, 2017 et al., 2009). In fact, barn owls largely occupied and prospered in these habitats until recently, and population declines of the species were related to foraging habitat loss, an increase of road network and shortage of suitable breeding sites (Arlettaz et al., 2010;Askew, Searle, & Moore, 2007;Hindmarch et al., 2012;Martínez & Zuberogoitia, 2004b). Nowadays, the effect of intensive farming plus the abuse of agro-chemical biocides accelerated the habitat homogenization and biodiversity loss and it is related to a reduction of barn owl population viability (Bruce, Christie, & Kirwan, 2014;König & Weick, 2008;Schmid, 2002 However, as it occurs in other European areas, eagle owls started to successfully breed in urban areas preying on alternative species (e.g. rats and pigeons; Penteriani & Delgado, 2019). Short-eared owls bred for the first time in our study area, with only one secure and three possible breeding events during the study period, all in extensive grasslands. We also registered the first record of boreal owl in a stand of mature mixed forest of beeches and pines with pastoral grasslands, located at 1,000 m a.s.l. The species could have been unnoticed in these habitats, similar to those found in other regions of its global range (Brambilla et al., 2013;Domahidi, Shonfield, Nielsen, Spence, & Bayne, 2019;Korpimäki & Hakkarainen, 2012;López et al., 2010). In fact, the south-westernmost European population of the species is located in the Pyrenees, 160 km from our record (Mariné, Lorente, Dalmau, & Bonada, 2005), and this distance is included within the breeding dispersion range of the species in the Pyrenees (Badosa et al., 2012). Castro, Muñoz, and Real (2008) included the Basque Mountains in the distribution projections modelled for the species.

| CON CLUS ION
Large-scale surveys are needed to obtain data to apply towards species conservation. However, we have shown that previous efforts focused on the knowledge of biology and behaviour of the target species are needed to adequately develop survey programmes and to correct the effects of imperfect detection on the results. Likewise, to reduce the effect of imperfect detection it is fundamental to consider the effects of survey-specific covariates in the methodology design and the analytical development, mainly those that we can a priori manage as the use of broadcast voices, observer experience or the survey time. Our results also indicate some ecological adaptations and population changes in the nocturnal bird community following an increase in urbanization and the extent of timber plantations, and also the simplification of natural habitats. This information is crucial to design future monitoring programmes across our study area, as well as other large-scale areas, and to adopt management actions for conservation purposes.