Nuanced understanding of seasonal movements of partially migratory birds is paramount to species and habitat conservation. Using nascent statistical methods, we identified migratory strategies of birds outfitted with radio-frequency identification (RFID) tags detected at RFID feeders in two sites in California, USA. We quantified proportions of migrants and residents and the seasonal phenology for each movement strategy in Allen’s and Anna’s hummingbirds; we also validated our methodology by fitting our model to obligate migratory black-chinned hummingbirds. Allen’s and Anna’s hummingbirds exhibited characteristics of facultative migratory behaviour. We also quantified apparent annual survival for each migratory strategy and found that residents had significantly higher probabilities of apparent survival. Low survival estimates for migrants suggest that a high proportion of birds in the migrant group permanently emigrated from our study sites. Considered together, our analyses suggest that hummingbirds in both northern and southern California sites partake in diverse and highly plastic migratory behaviours. Our assessment elucidates the dynamics underlying idiosyncratic migratory behaviours of two species of hummingbirds, in addition to describing a framework for similar assessments of migratory behaviours using the multi-state open robust design with state uncertainty (MSORD-SU) model and single-site dynamics.

The dataset "EnglishEtAl2022_CaptureHistories.csv" includes individual monthly detection histories for each tagged bird in the study. Each detection history comprises 36 characters corresponding to each month from February 2018 to January 2021.

Detections were encoded with one of three characters: `0` denotes no detection within the monthly period, `u` denotes that the bird was detected, but detection took place outside of the 3 winter months (November to January) and its state is therefore unknown, and lastly `R` denotes that a bird was detected during winter and is therefore classified as a resident.

These data were analysed using Program MARK via RMark version 2.2.7.