Year: 2026
Type: article
Abstract: The Maximum Entropy Theory of Ecology (METE) derives probability distributions for micro-scale variables, such as species abundances and metabolic rates, solely from macro-scale observables like total abundance and species richness. While METE performs well in systems with relatively constant macro-scale observables, its predictive accuracy declines in disturbed or rapidly changing systems due to its inherently static approach. To address evolving macro-scale observables, the theory of Dynamic Maximum Entropy across Entwined Scales (DyMES) introduced dynamical constraints on the micro-scale probability distribution, thereby linking macro- and micro-scale dynamics. Macro-scale dynamics are expressed through time derivatives of METE’s observables, while micro-scale dynamics are modeled using pre-specified mechanistic transition functions. Building on this, we introduce METimE, a dynamic maximum entropy framework that infers transition functions directly from data using sparse regression on observed variables. This data-driven approach enables application across diverse systems without requiring the explicit mechanistic functions to be pre-specified. We demonstrate METimE’s utility using both simulated and empirical time-series data by deriving transition functions, evaluating their accuracy, and incorporating them into maximum entropy inference of species abundance distributions. Results indicate that data-driven transition functions accurately capture changes in micro-scale variables when species dynamics are similar and variation is sufficient. Even in empirical datasets with complex dynamics, incorporating dynamic constraints consistently improves the accuracy of predicted rank-abundance distributions. Notably, the benefit of dynamic constraints does not depend on the exact precision of the transition functions. This framework enables new data-driven approaches for understanding and managing ecosystem recovery from a dynamic, non-equilibrium perspective.
Source: Ecological Modelling
Institution: Utrecht University
Language: English
Subfield: Nature and Landscape Conservation
Field: Environmental Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,810
OpenAlex ID: W7214438022