Networked web-cameras monitor congruent seasonal development of birches with phenological field observations

Abstract

Ecosystems’ potential to provide services, e.g. to sequester carbon, is largely driven by the phenological cycle of vegetation. Timing of phenological events is required for understanding and predicting the influence of climate change on ecosystems and to support analyses of ecosystem functioning. Analyses of conventional camera time series mounted near vegetation has been suggested as a means of monitoring phenological events and supporting wider monitoring of phenological cycle of biomes that is frequently done with satellite earth observation (EO). Especially in the boreal biome, sparsely scattered deciduous trees amongst conifer-dominant forests pose a problem for EO techniques as species phenological signal mix, and render EO data difficult to interpret. Therefore, deriving phenological information from on the ground measurements would provide valuable reference data for earth observed phenology products in a larger scale. Keeping this in mind, we established a network of digital cameras for automated monitoring of phenological activity of vegetation in the boreal ecosystems of Finland. Cameras were mounted at 14 sites, each site having 1–3 cameras. In this study, we used data from 12 sites to investigate how well networked cameras can detect the phenological development of birches (Betula spp.) along a latitudinal gradient. Birches typically appear in small quantities within the dominant species. We tested whether the small, scattered birch image elements allow a reliable extraction of colour indices and the temporal changes therein. We compared automatically derived phenological dates from these birch image elements both to visually determined dates from the same image time series and to independent observations recorded in the phenological monitoring network covering the same region. Automatically extracted season start dates, which were based on the change of green colour fraction in spring, corresponded well with the visually interpreted start of the season, and also to the budburst dates observed in the field. Red colour fraction turned out to be superior to the green colour-based indices in predicting leaf yellowing and fall. The latitudinal gradients derived using automated phenological date extraction corresponded well with the gradients estimated from the phenological field observations. We conclude that small and scattered birch image elements allow reliable extraction of key phenological dates for the season start and end of deciduous species studied here, thus providing important species-specific data for model validation and for explaining the temporal variation in EO phenology products.