Opportunistically collected photographs can be used to estimate large-scale phenological trends

ABSTRACTPremiseResearch on large-scale patterns of phenology have utilized multiple sources of data to analyze the timing of events such as flowering, fruiting, and leaf out. In-situ observations from standardized surveys are ideal, but remain spatially sparse. Herbarium records and phenology-focused citizen science programs provide a source of historic data and spatial replication, but the sample sizes for any one season are still relatively low. A novel and rapidly growing source of broad-scale phenology data are photographs from the iNaturalist platform, but methods utilizing these data must generalize to a range of different species with varying season lengths and occurring across heterogenous areas. They must also be robust to different sample sizes and potential biases toward well travelled areas such as roads and towns.Methods/ResultsWe developed a spatially explicit model, the Weibull Grid, to estimate flowering onset across large-scales, and utilized a simulation framework to test the approach using different phenology and sampling scenarios. We found that the model is ideal when the underlying phenology is non-linear across space. We then use the Weibull Grid model to estimate flowering onset of two species using iNaturalist photographs, and compare those estimates with independent observations of greenup from the Phenocam network. The Weibull Grid model estimate consistently aligned with Phenocam greenup across four years and broad latitudes.ConclusioniNaturalist observations can considerably increase the amount of phenology observations and also provide needed spatial coverage. We showed here they can accurately describe large-scale trends as long as phenological and sampling processes are considered.