Comparing detectability patterns of bird species in small ponds using multi-method hierarchical modelling

Robust knowledge of biodiversity distribution is essential for designing and developing effective conservation actions. However, monitoring programmes have historically assumed all species are detected equally with no spatial or temporal differences in their detection rates. However, recently, interest in accounting for imperfect detection has greatly increased in studies on animal diversity. In this respect, birds are the most widely used group for hierarchical occupancy-detection modelling, mainly due to the relative ease of sampling and the large number of bird datasets that are available. Nevertheless, there are no studies that have tried to evaluate the effectiveness of different bird sampling methods based on a hierarchical modelling approach. In an attempt to remedy this situation, we conducted point transects (PT), point transects plus video monitoring (PV) and mist netting (MN) in 19 small ponds located in the province of Murcia, southeastern Spain, one of the most arid regions of Europe. Multi-method hierarchical models were fitted to the detection histories of 36 common bird species with three main objectives: to compare the effectiveness of the three sampling methods for detecting the bird species using ponds, to assess the effect of sampling date on species detectability, and to establish the influence of body size and diet on species detectability. The results showed PV to be the most effective sampling method for detecting species occupancy, although detection rates ranged widely among bird groups, and some large species were weakly detected by all the methods. Average detectability increased during the breeding period, a pattern shown similarly by all sampling methods. Our approach is particularly applicable to both single- and multi-species bird monitoring programmes. We recommend evaluating the cost-effectiveness of available methods for sampling design in order to reduce costs and improve effectiveness.

method hierarchical models were fitted to the detection histories of 36 common bird 25 species with three main objectives: to compare the effectiveness of the three sampling 26 methods for detecting the bird species using ponds, to assess the effect of sampling 27 date on species detectability, and to establish the influence of body size and diet on 28 species detectability. The results showed PV to be the most effective sampling method 29 for detecting species occupancy, although detection rates ranged widely among bird 30 groups, and some large species were weakly detected by all the methods. Average 31 detectability increased during the breeding period, a pattern shown similarly by all 32 sampling methods. Our approach is particularly applicable to both single-and multi-

38
Monitoring biodiversity is key to providing measures of status and trends of wildlife as well as 39 for understanding its responses to threats derived from human activities. Abundance and 40 distribution are the most widely used biological measurements in ecological studies and are 41 frequently provided by large-scale monitoring programmes [1]. However, despite their historically been expressed by ecologists [4], but these concerns have only begun to be explored from an analytic point of view in the two last decades. Recently, some studies have 49 reported strongly inaccurate abundance estimates as a result of not taking into account 50 possible imperfect detection, masking trends and providing misinformation that can affect 51 conservation actions [5,6]. Hence, setting an accurate study design based on effective 52 sampling methods that maximize species detectability is key to outlining any biological 53 monitoring programmes.

54
The detection rate or detectability (p) is defined as the probability of detecting at least one 55 individual of a given species in a particular site, given that individuals of that species are 56 present in the study area during the survey [7,8]. Traditionally, the vast majority of studies 57 have assumed all species composing a biological community are similarly detected [4],and 58 detectability is constant over space, time, different methods or weather conditions. However, 59 interest in incorporating imperfect detection (p < 1) into biodiversity studies has largely 60 increased in the last two decades due to the development of novel hierarchical modelling 61 techniques [9,10]. Several approaches have been suggested in this modelling framework in 62 order to estimate species richness, abundance and distribution corrected for incomplete 63 detection. For example, species richness models enable unbiased estimates of site-specific 64 species richness to be calculated while accounting for imperfect detection [11], thus enhancing climate with colder winters and higher mean annual precipitation than the coastal zones. Detection histories 218 We generated method-and survey-specific detection histories for all the breeding species 219 recorded during the study period. Species with less than five records were removed from the models in order to avoid bias related to small sample size, since estimates may be unreliable when information collected on species presence/absence is limited [45,46]. Additionally, 222 migratory non-breeding species in the study area were also removed from the models to meet 223 the closure assumption. Separate detection histories were generated for each sampling 224 method and each survey visit. Therefore, as sampling event refers to one survey per method, a 225 maximum of nine detection events were possible for each species because three surveys were 226 conducted for each of the three sampling methods. A value of 1 was attributed when a species 227 was detected in a specific survey using a single method, a value of 0 being given otherwise.  were similar for all surveys so were not expected to affect detectability. We also assumed that 279 occupancy was independent among study sites because the minimum distance between ponds 280 was always greater than 1.5 km, which is a reasonable distance to consider sites as 281 independent when the survey period covers the breeding season of birds.  (<30g); 4, small seed-eaters (<30g): 5, medium-sized and large 332 seed-eaters (≥30g); and 6, medium-sized and large generalists (≥30g). Model-averaging estimates of detection probability for study species showed clear differences 343 in method-specific detectability (Fig 2). Occupancy detection increased slightly during breeding 344 season but differences among the three sampling methods remained constant for the three 345 surveys. PT and PV provided similar detectability estimates but slightly higher in the second 346 case. PV obtained detectability estimates significantly higher than MN, but the other pairwise 347 comparisons did not provide significant differences. Nevertheless, detection estimates of some 358 methods was found (Fig 3A). The occupancy estimates ranged widely from = 0.14 in Eurasian

368
A lineal relation was clearly seen for the estimated detectability using each method when 369 pairwise comparisons of sampling methods were carried out (Fig 3B-3D  and Sylvidae families (finches and warblers, respectively). However, the estimated 395 detectability of Emberizidae (buntings) and Paridae (tits) families was very similar for the three 396 sampling methods (Fig 4). Detectability with the PT and PV methods was very similar for all the 397 studied families, except Muscicapidae family which were significantly better detected by PV.

398
On the other hand, visual methods were also more effective than MN at detecting species at 399 group-level (Fig 5). Only small seed-eaters were equally detected by the three sampling 400 methods. Detectability for the other groups was always higher with PT and PV than with MN.  28,58,59], such as warblers. Only small seed-eaters were detected with similar effectiveness by the three target methods. Importantly, MN showed the higher variability in the detection estimates and also led to large differences in species detectability even within families and 488 groups (Fig 4 and Fig 5). Aknowledgments Murcia for permission to access to protected areas.