Towards an automated approach for monitoring vegetation green-up dynamics using vehicle dashcams in urban environments

Phenological events are highly sensitive to climatic variation, and temporal phenological shifts have significant impact on ecosystem function. Vegetation in urban environments holds significant value in providing ecosystem services, of which will become increasingly important as urban populations grow. Insights into vegetation phenological transitions have typically long been monitored through satellite imaging analysis and ground-based field measurements, but these methods are limited by financial costs and coarse resolutions, both spatially and temporally. Despite an increase in the growth of fixed digital camera networks for monitoring vegetation phenology, there still exists a data gap in urban settings. Findings of this study showcased that time series imagery of street level trees in urban environments is obtainable from vehicle dashcams. The YOLOv3 deep learning algorithm demonstrated suitability for automating stages of processing towards deriving a greenness metric. However, further work is required to determine an optimum sized detector training dataset, which also proportionally represents trees across the phenological cycle. Questions remain as to how error caused by scene illuminance variation can be mitigated and as to how full automation from raw data to the final green-up metric can be reached