From biodiversity response to ecological inference: context- and scale-dependent relationships between land use and biodiversity change in tropical forest landscapes

Abstract

Land-use change does more than remove habitat for species. It reorganises vegetation structure, climatic exposure, spatial configuration and the composition of the communities that remain. Biodiversity change is nonetheless commonly summarised through average responses, and through model coefficients whose ecological meaning is assumed rather than demonstrated. This dissertation asks when environmental context changes the biological consequences of land use, and when the structure of observational landscapes changes what those consequences can be inferred to represent. Across 2,986 georeferenced, effort-reported assemblages spanning seven dataset-defined animal groups across five broad taxonomic groups (mammals, birds, reptiles, amphibians and butterflies) and broad climatic, topographic, and productivity gradients in the Atlantic Forest of South America, land-use effects on species richness varied strongly with ecological context. Thermal variability intensified negative land-use effects, most consistently among forest specialists. That contrast is consistent with the loss of climatic buffering when forest is removed, although canopy microclimate was not measured directly. Heterogeneity carried no universal biodiversity value either. Persistent geophysical variation supported higher baseline richness, whereas remotely sensed vegetation greenness variability compounded cropland effects, showing that different heterogeneity axes can carry contrasting ecological meanings. Guild responses opposed one another along the heterogeneity and productivity axes, so aggregation obscured moderation that was present within guilds. The same land-use transformation that alters climatic and vegetational context jointly changes how much habitat remains and how it is spatially arranged. Fragmentation therefore provides a particularly demanding case for asking whether correlated landscape changes can be separated strongly enough to support the ecological interpretation assigned to a model coefficient. The problem appears in two contrasting fragmentation cases, a global multi-taxa forest dataset and an Atlantic Forest tree beta-diversity study. In the global case, habitat amount and patch number remained nonlinearly coupled in the realised predictor space, pushing the patch-number coefficient towards zero while conventional collinearity screening gave little warning. In the Atlantic Forest case, a supported negative patch-number coefficient documented a real subdivision association without identifying the mechanism invoked. Contrasting coefficient outcomes led to the same inferential lesson. What a model estimates is not automatically what an ecological claim requires. Competing fragmentation frameworks imply different response surfaces along the habitat-amount gradient. The habitat amount hypothesis predicts invariance once amount is controlled. Scarcity and threshold dynamics predict stronger costs where habitat is limited. Coexistence can generate sign reversal, whereas extinction filtering predicts stronger detectable effects where sensitive taxa remain. Across Atlantic Forest vertebrate analyses spanning multiple forest definitions, configuration metrics, spatial scales and predictor formulations, forest amount remained the dominant landscape component. Configuration associations were smaller but became more negative towards higher forest amount, most consistently among forest specialists. The overall geometry was most consistent with extinction filtering, although some supported endpoint patterns departed from its prospective prediction, and cross-sectional data and reduced recoverability at low cover prevent unique identification of the mechanism. Biodiversity responses to land-use change are therefore conditional twice over. Ecological context governs which responses are expressed, and inferential context governs which of those responses can be recovered and interpreted. Assemblage composition determines which signals remain in the response and whether aggregation preserves them. Spatial scale determines which processes enter the analysed grain and extent, and therefore which contrasts an observational analysis can recover. Aggregate richness can preserve shared land-use damage while missing the contexts and assemblage components in which that damage intensifies. Forest retention and restoration remain the clearest conservation priorities supported by the Atlantic Forest data, and nothing here justifies trading forest amount for configuration. Even so, configuration was consistently associated with an additional biodiversity cost after accounting for forest amount, particularly for forest specialists and in better-forested landscapes. Interpreting land-use effects therefore requires knowing which species remain, the spatial scale at which responses are observed, and whether the available landscape contrasts identify the ecological process of interest before a fitted coefficient can be interpreted as evidence for the process it is meant to represent.