The intra- and inter-annual seasonality of terrestrial vegetation plays a key role in regulating global carbon, water and energy cycles. This is particularly true for mid- to high-latitude temperate, boreal, and Arctic biomes where the timing and magnitude of vegetation seasonality strongly regulates land-atmosphere exchanges, while day-to-day variation is strongly controlled by diel patterns in vegetation function responding to short-term environmental conditions. In order to model daily to annual and larger time-scales, land-surface models need to represent the key components and properties of terrestrial vegetation which regulate the magnitude and timing of carbon, water and energy fluxes (e.g. stand structure, photosynthesis, leaf area, light harvesting, energy balance). However, it is well known that models show large discrepancies in the simulation of vegetation seasonality and resulting impacts on ecosystem states and fluxes, based on past single and multi-model inter-comparisons. In addition, past efforts generally haven’t included the impacts of parameter uncertainties of key processes or provided a consistent benchmarking framework to standardize simulation and evaluation.

We present a multi-model analysis of modeled vegetation seasonality across select mid- to high-latitude Fluxnet sites representing a range of plant functional types. We focus on the impacts of parameter uncertainties in key processes, including canopy radiation transfer, photosynthesis, water, and energy cycling using simulations conducted within the Predictive Ecosystem Analyzer (PEcAn). We then compare ensemble simulations across models and to novel site-level benchmarks of short (daily) to longer (seasonal) timescale fluxes derived from Fluxnet observations within the ILAMB framework. We show how overall model structure and complexity impacts predictions, but also how parameter uncertainty plays a significant role driving the divergence between models and observations. Our novel benchmarking analysis illustrates the impact of temporal scale on the comparisons, including which periods show less or more model divergence and error, as compared to observations. Analyses, such as that presented here, are critical for understanding model deficiencies and identify key areas for model improvements.