Assessing Ecosystem State Space Models: Identifiability and Estimation

Abstract Hierarchical probability models are being used more often than non-hierarchical deterministic process in environmental prediction and forecasting, Bayesian approaches to fitting such becoming increasingly popular. In particular, describing ecosystem dynamics with multiple states that autoregressive at each step time can be treated as statistical state space (SSMs). this paper, we examine subset of models, embed a process-based model into an SSM, give closed form Gibbs sampling updates for latent precision parameters when observation errors normally distributed. Here, use simulated data from example (DALECev) study the effects changing temporal resolution observations on (observation gaps), (model step), level aggregation fluxes (measurements transfer rates process). We show parameter estimates become unreliable gaps between observed increase. To improve estimates, introduce method tuning while still using higher-frequency driver information helps estimates. Further, cloning is suitable assessing identifiability class models. Overall, our inform application ecological forecasting applications where (1) not available all transfers operational (2) uncertainty estimation desired.