Dynamic generalised additive models ( <scp>DGAMs</scp> ) for forecasting discrete ecological time series

Generalised additive models (GAMs) are increasingly popular tools for estimating smooth nonlinear relationships between predictors and response variables. GAMs particularly relevant in ecology representing hierarchical functions discrete responses that encompass complex features including zero inflation, truncation uneven sampling. However, less useful producing forecasts as their provide unstable predictions outside the range of training data. We introduce dynamic generalised (DGAMs), where GAM linear predictor is jointly estimated with unobserved components to model time series evolve a function associations latent temporal processes. These especially analysing multiple series, they can estimate while learning via dimension-reduced factor implement our mvgam R package, which estimates parameters smoothing splines processes probabilistic framework. Using simulations, we illustrate how outperform competing formulations realistic ecological forecasting tasks identifying important functions. use real-world case study highlight some mvgam's key features, include calculating correlations among series' trends, performing selection using rolling window posterior predictive checks, online data augmentation recursive particle filter visualising uncertainties predictions. Dynamic (DGAMs) offer solution challenge ecologically associations. Our Bayesian approach will be exploring structures providing robust uncertainties, becoming applied ecology.