Ensemble runs of a simple forest gap model produce very uncertain forecasts of the abundance of eastern hemlock, a species that plays an important role in the terrestrial carbon cycle.
From rising sea levels to melting permafrost, society faces environmental challenges unprecedented in human history. Predictive forecasts of environmental change require a mechanistic process model informed by data with a full accounting of uncertainty.
Process models of ecosystem change range from simple box and flux models to mechanistic Earth system models with hundreds of thousands of lines of code. Processes of interest occur at temporal scales ranging from seconds (photosynthetic response to changing light) to centuries (slow pools of soil carbon, evolutionary changes in plant traits) and at spatial scales from leaf to globe. Even simple models include nonlinear dynamics and feedbacks across temporal and spatial scales.
Data to empirically constrain modeled environmental state and model parameters are often heterogenous, sparse, noisy, and indirect. Bayesian hierarchical models allow us to estimate model parameters and state variables across space and time with a full accounting of uncertainty.
Because estimates of ecosystem change account for uncertainty, researchers can assimilate them into ensemble runs of ecosystem models using a variety of parameter and state variable data assimilation schemes. Forecasts of ecosystem dynamics are thus informed by the mechanisms of the process models and constrained by real world empirical data.