Attributing interannual variability of net ecosystem exchange to modeled ecological processes in forested wetlands of contrasting stand age

The drivers of interannual variability (IAV) of net ecosystem exchange (NEE) in forested wetlands are poorly understood, making it difficult to predict changes in atmospheric fluxes in response to land use and climate change. Similarly, these ecosystems demonstrate dynamic physiological and phenological responses to climate over time yet are typically modeled using static parameters that represent unchanging ecological conditions. Though static first-order ecosystem models are informative, they fundamentally lack the ability to represent dynamic annual changes in ecological processes that may drive IAV of NEE through time. We aimed to improve understanding of how forested wetlands dynamically respond to climate and which key ecological processes may contribute to IAV of NEE. Simultaneously, we aimed to develop tools to evaluate dynamically parameterized process based first-order ecosystem models. To achieve these objectives, long-term ecological data were fused with the Total Ecosystem (TECO) model in three loblolly pine plantations and a bottomland hardwood forest of contrasting stand age in wetland areas of the lower coastal plain of North Carolina. Variance decomposition was used to assess changes in large-scale ecosystem drivers. To investigate individual processes, both static and dynamic data-assimilation were conducted to simulate time–invariant and time–varying ecological response. Anomalies in dynamic ecosystem process response were correlated with NEE anomalies to attribute IAV of NEE to underlying process-based mechanisms that may drive annual changes in NEE across stand age and sites. Assessment of large-scale drivers of IAV of NEE across sites demonstrated that maximum carbon uptake (MCU) dominated IAV of NEE in the mature pine plantation. These large-scale NEE signals were further parsed into ecological processes in the TECO model, where process anomaly correlation showed that slight variations in root maintenance respiration and woody biomass turnover rates may be underlying drivers of IAV of MCU and subsequently NEE. However, in the young pine plantations and bottomland hardwood forest IAV of NEE was not dominated by MCU. In contrast, IAV of NEE in young plantations was influenced most by annual changes in maximum carbon release (MCR) and carbon uptake period (CUP), while IAV of NEE in the bottomland hardwood forest was dominated by CUP. These results demonstrate that dynamic data assimilation (DA), variance decomposition, and process anomaly correlation are investigative and diagnostic tools for process-based models, though maximum GPP was systematically underestimated by models across sites. Despite problems with peak GPP representation, anomaly correlation between ecological processes and IAV of NEE allowed investigation of the specific ecological drivers of annual variability in ecosystem-level carbon exchange. As ecosystems show dynamic physiological and phenological properties through time, it may be important to allow models to have dynamic/time–varying ecological responses, especially if the root causes of IAV of NEE are to be attributed to ecological processes in process-based models.