Mechanistic persistence as predictor or constraint? Evaluating alternative pathways for integrating biological knowledge into species distribution models

Publication Type

Journal Article

Journal Name

Ecological Modelling

Publication Date

12-1-2026

Abstract

Hybrid species distribution models (SDMs) aim to combine the ecological realism of mechanistic approaches (process-based) with the predictive power of correlative models. However, it remains unclear whether mechanistic information should be incorporated as an explanatory predictor or used to constrain model calibration itself. This study evaluated these alternative integration strategies using a temperature-dependent mechanistic persistence index integrating stage-specific development, mortality, and reproduction into a single measure of population persistence. This biologically derived persistence index was incorporated into generalized linear models (GLMs) and generalized additive models (GAMs) as (i) an additional predictor, (ii) an observation-level weighting factor, or (iii) both simultaneously. Three orthopteran species spanning contrasting ecological strategies, from the highly migratory, arid-adapted Schistocerca gregaria to the temperate Calliptamus italicus and the sedentary, humid-environment Atractomorpha sinensis, were used as case studies. Predictive performance differed little among formulations, with predictor-based integration producing classification metrics nearly identical to baseline SDMs (ΔAUC ≤ 0.003 across species). In contrast, weighting-based integration substantially altered suitability distributions and spatial patterns despite only modest changes in discrimination performance. Across species, predictor-only models remained structurally almost identical to baseline SDMs (SSIM = 0.988–0.9999), whereas weighting approaches reduced structural similarity to as low as 0.388 and substantially modified suitability gradients, entropy, and the spatial organization of highly suitable areas. These effects were strongest for A. sinensis, where weighting-based integration produced marked redistribution of predicted suitability. The study results show that the influence of mechanistic information in hybrid SDMs depends more on the mode of integration than on predictive accuracy alone. Mechanistic weighting acts as a soft ecological constraint capable of reshaping spatial suitability patterns, providing an alternative pathway for integrating biological knowledge into correlative distribution models.

Keywords

Ecological constraints, Grasshopper, Hybrid species distribution modelling, Mechanistic persistence, Weighted likelihood

Disciplines

Data Science | Ecology and Evolutionary Biology

Link to Full Text

Share

COinS