Selecting a suitable nest site is critical to the survival and reproduction of birds. Prospecting allows individuals to gather information on the local quality of potential future breeding sites, which may help them make the best nest site selection decision. However, few studies have focused on the direct links between the prospecting activity of breeders and subsequent nest site selection. In this study, we investigated the prospecting pattern of Japanese tits (Parus minor) during the pre-breeding period of the first breeding attempt and whether nest site characteristics influence their nest box visiting behaviour and occupied nest site. We used radio frequency identification (RFID) to track the movements of Japanese tits visiting nest boxes and compared nest site characteristics between visited and unvisited (control) nest boxes, as well as between visited and occupied nest boxes. We found that Japanese tits started visiting nest boxes approximately 20 days before breeding, visited an average of 6 nest boxes and eventually chose the most visited nest box for breeding activities. Japanese tits were more likely to visit nest boxes that had less canopy cover and lower shrub density but a greater total number of surrounding trees and ultimately chose breeding nest boxes with a smaller entrance inclination, in nesting trees with a larger diameter at breast height (DBH) which were surrounded by trees with a larger DBH. Our results suggest that Japanese tits visit several potential breeding sites before choosing breeding nest boxes and that nest site characteristics can influence their prospecting activity and nest site selection.

Study area and study species

The study area was located in the eastern Changbai Mountains in the Zuojia Nature Reserve (44°1′–45°0′ N, 126°0′–126°8′ E), situated in Jilin Province, northeastern China. The total area of the reserve is ~5,544 ha. The area is hilly in the transition from the Changbai Mountain sub-region to the western plain, with an elevation of 200-530 m. The average temperatures in March, April, and May in the Zuojia Natural Reserve area from 2005 to 2021 were -1.01 ± 0.26℃, 8.16 ±0.24℃, 15.38 ± 0.18℃, respectively. Local temperatures were obtained from Jilin Suburban Weather Station, 44 km from the study area [Chinese National Meteorological Center of CMA (http://eng.nmc.cn/), weather station No. 54172]. The main forest type in the study area is secondary broad-leaved mixed forest, and the forest age is generally 50-60 years old. The main tree species are Quercus mongolica, Sophora japonica, Ulmus japonica, Betula davurica, Fraxinus mandshurica, Tilia mandshurica, Pinus sylvestris, Pinus koraiensis, Populus davidiana, and Pinus massoniana (Deng and Gao 2005, Fan et al. 2021). The main shrub species are Rosa dahurica, Spiraea salicifolia, Rosa d oreana, Sorbaria sorbifolia, and Lonicera maximowiczii (Deng and Gao 2005). Since 2009, we have been providing approximately 450 artificial nest boxes for secondary cavity-nesting birds, such as Japanese tits and Eurasian nuthatch (Sitta europaea), and have carried out many types of research in this area (e. g., Yu et al. 2017, Shen et al. 2021, E et al. 2021). The study was undertaken from February-July in 2020 and 2021. Nest boxes had inner dimensions of 12 cm × 12 cm × 25 cm, an entrance hole with a diameter of 5 cm with its lower end 155 mm from the floor. All nest boxes were attached to the trees and were approximately 2-3 m above the ground, with each nest box 30-50 m apart. The tree species for hanging the boxes and the orientation of them were randomly selected. We routinely cleaned out old nest materials from artificial nest boxes before the initiation of the breeding season. The Japanese tit population in our area began occupying the territory in early March, adding nesting materials to the nest box in early April, and laying eggs in mid-to-late April. We checked the nest boxes at least once a week to record breeding attempts and reproductive data (e.g., egg-laying date, clutch size, hatching date, brood size, number of fledglings) of Japanese tits (Yu et al. 2017).

Tagging and monitoring birds

In the pre-breeding period (from mid-March to early April), all nest box entrances were fixed with a circular antenna (approximately 4.2 cm in diameter), which was connected to a custom-made RFID reader (125-kHz, Guangzhou Yongjin Intelligent Technology Co., Ltd., Guangzhou, China). The reader recorded every time a bird equipped with a passive integrated transponder (PIT) tag passed through the nest box entrance. The unique 10-digit identification code and the current date and time to the seconds were recorded in a log file. PIT tags do not have their own power source and, therefore, can only be detected within the range of the antenna (~2–3 cm). The readers were powered by a lithium battery (3.7 v, 10,000 mAh) and solar panels and could also be powered via a USB connection charger.

We trapped the adults in a cage using song playbacks of local males in the pre-breeding period. The cage was designed to include two levels, the upper level with plenty of food, water, and a small wooden stick and the lower level with a loudspeaker (Royqueen M300, Shenzhen, China) continuously playing the song of the Japanese tits to attract conspecifics. The wooden stick held the lid partially open; when the birds were attracted to the food and stepped on the stick, they were locked in the cage. This method of capture does not harm the birds. We continuously monitored the trapping equipment, and as soon as birds were trapped, we immediately removed them, and they were sexed and tagged by unique PIT tags. Handling time averaged 2 min per individual after which they were released at the location of capture. We did not attempt to capture birds during inclement weather. The PIT tags were attached to especially made plastic blue rings with an internal diameter of 2.6 mm. The weight of the tag (0.45 g) was approximately 3% of the bird's body weight, which is well below the recommended 5%-of-body-weight criterion for flying animals (López-López 2016). The PIT tag number was read and recorded before and after the placement using a commercial handheld reader. No birds were injured or suffered long-term effects in the course of PIT tagging or RFID monitoring.

Nest site characteristics

We measured 13 nest site characteristics of the nest boxes visited (including those eventually occupied) by Japanese tits and corresponding randomly selected control nest boxes (not visited or occupied by any birds). X. L. was responsible for the measurement of nest site characteristics. In addition, two researchers, L. W. and Y. Y. assisted with the nest site characteristics after being trained by X. L. to standardize the measurements to reduce human error. The detailed measurements for all variables are described in Table 1. We located the longitude and latitude of nest boxes using handheld GPS (Garmin Intl. Inc., Olathe, KS).

Statistical analysis

To assess the influence of the 13 nest site characteristics on nest box visitation rate, we used a generalized linear mixed model (GLMM) with a logit-link and binomial error distribution (control nest box = 0, visited nest box = 1; Betts et al. (2008), Olah et al. (2014)). Since the data were collected during two years (2020 and 2021), “year” was added as a random variable. To avoid pseudo-replication, we also added the identity of the visited/control nest box as a random variable. To determine the influence of the 13 nest site characteristics on the nest box selection of Japanese tits, we used conditional logistic regression (visited but rejected nest box = 0, occupied nest box = 1), where the identity of the Japanese tit was the grouping variable (Thomas and Taylor 2006, Hartman et al. 2016, Dyson et al. 2019). Before model development, we screened explanatory variables for strong correlations (Menard 2002), and no strongly correlated explanatory variables were found (all explanatory variables displayed Pearson’s correlation coefficient < 0.65 and variance inflated factor (VIF) < 2) (O’brien 2007, Dyson et al. 2019). Among the 13 nest site characteristics, only the orientation and tree species were categorical variables, while the other 11 were continuous variables. Then, we standardized all continuous variables prior to analysis (subtracted the mean and divided by the standard deviation), resulting in the mean for each predictor equalling 0 and the standard deviation equalling 1.

In all analyses, we used Akaike’s information criterion (AIC) to perform model selection. We ranked competing models using ΔAICc and selected the top-ranked model with the ∆AICc ≤ 2 (Burnham and Anderson 2002). Then, among the selected models, we choose the model with the fewest variables as the best model. If there was a significant effect of the categorical variables (orientation and tree species), we further performed Tukey’s honest significant difference (HSD) post hoc pairwise comparisons using the ‘lsmeans’ package to determine differences between categorical variables (Lenth 2016) and used the Bonferroni correction to correct the P-values. We evaluated model performance by calculating the area under the curve (AUC) of the receiver operating characteristic (ROC) for the best model (Fielding and Bell 1997). We considered a model with an AUC score of 0.7–0.9 to be a good predictor.

We performed all analyses in the program R (version 4.2.1). The GLMMs were implemented using the ‘lme4’ package (Bates et al. 2015), conditional logistic regression was performed by the ‘survival’ package (Therneau 2015), and the multi-model inference package ‘MuMIn’ was used for model selection. The results were visualized using ggplot2 (Wickham and Chang 2022).