Evaluating the level of agreement between human and time-lapse camera observations of understory plant phenology at multiple scales

Abstract

The growing popularity of digital-repeat photography in field research is seeing traditional field efforts being assisted and even replaced by low-cost cameras. The efficiency of using cameras is obvious, but there is an assumption that they capture the same information as observations made by humans. This paper aims to determine the level of agreement between these two methods of interpreting understory vegetation phenology. We compared daily phenological observations made by low-cost cameras with those made by personnel during field visits every 10days. Phenophases were defined as the non-spectral, physical developmental stages of Canadian buffaloberry (Shepherdia canadensis) and alpine sweetvetch (Hedysarum alpinum). The relationship between observation methods was quantified using a weighted kappa statistic at three spatial scales ranging from individual plants to areas up to 6ha. Agreement between the camera observations and those made by field personnel was nearly perfect (Kappa>0.9) for both the vegetative and reproductive phenology of both study species at all spatial scales. The level of agreement was found to be more variable early in the season when plant growth is more rapid. Overall there was a slight bias in the image interpretations to underestimate the rate of development. Time-lapse photography was found to be an analogous replacement for field visits; however, some plant species are more suitable for observation by camera than others. Spatially, it was determined that observations of a single plant are all that is required to capture the phenology of the surrounding region in excess of 6ha. This analysis was carried out over a single growing season in the in the Rocky Mountains of western Alberta, Canada.