Recent developments in empirical dynamic modelling

Ecosystems are complex and sparsely observed making inference prediction challenging. Empirical dynamic modelling (EDM) circumvents the need for a parametric model complete observations of all system variables. Classical univariate approaches, which require time-series only single focal variable, can produce verifiable out-of-sample forecasts; however, they sometimes long time series that may be difficult to obtain. More importantly, classical approaches limit depth mechanistic understanding gained generalizability forecasts non-analogue futures. We review main ideas EDM more recent extensions expand their capabilities improving mechanism. Algorithms now available allow missing data, unequal sampling intervals combining short series, increase number datasets used. Recent multivariate substantially range applications questions addressed, including detecting causal coupling, tracking changing interactions in real time, leveraging from information shared coupled variables, dynamically stability, scenario exploration, management involving optimal control.