Leveraging foundation models and time-lapse photography for semi-automated tracking of tree apical growth

Primary growth is a key determinant of tree seedling survival, competition, forest composition, and productivity. Despite its importance, no established in-situ techniques currently exist for efficiently tracking fine-scale height growth. To address this, we developed a semi-automatic approach using time-lapse images and the Segment Anything Model v2 (SAM2; Meta AI Research) to quantify conifer apical growth. We obtained image time series of 105 apical balsam fir leaders from two sites in Québec, Canada, collected over 13 growing seasons, using two different camera types, with background varying from other trees to the sky. We used SAM2 with manual prompts to segment leaders within image time series. We measured the length of each segmentation and applied a post-processing step to address potential erroneous segmentations. Leader lengths derived from SAM2 concurred with independent manual reference measurements (R 2 = 0.984, 8309 images) which took 20 times longer to acquire (one hour) versus the SAM2 approach (three minutes one second) on average per series. Agreement between SAM2 and manual growth series for the timing of 75% growth amplitude was moderate (R 2 = 0.57) and could be augmented with mask revisions. This cost-effective fine-scale monitoring framework could be applied to track seedling, sapling, and stand-level growth. It also could augment evergreen primary growth tracking in phenocam networks and provide reference measurements for large-scale remote sensing programs. By enabling the development of intra-seasonal height growth profiles, this approach could support the identification of climate-resilient species and provenances.