Insights on biodiversity drivers to predict species richness in tropical forests at the local scale

• Published in: Ecological modelling Vol. 473; p. 110133
• Main Authors: Mateo, Rubén G., Arellano, Gabriel, Gómez-Rubio, Virgilio, Tello, J. Sebastián, Fuentes, Alfredo F., Cayola, Leslie, Loza, M. Isabel, Cala, Victoria, Macía, Manuel J.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2022
• Subjects: Altitudinal gradient; Bayesian inference; Biodiversity patterns; Madidi region; Plant species richness; Random forest; Plant species richness; Madidi region; Random forest; Altitudinal gradient; Bayesian inference; Biodiversity patterns
• ISSN: 0304-3800; 1872-7026
• DOI: 10.1016/j.ecolmodel.2022.110133

•In tropical areas, due to data scarcity, biodiversity modelling is certainly required.•Species richness is accurately predicted at a detailed resolution.•This innovative methodology could be employed in other tropical areas. Disentangling the relative importance of different biodiversity drivers (i.e., climate, edaphic, historical factors, or human impact) to predict plant species richness at the local scale is one of the most important challenges in ecology. Biodiversity modelling is a key tool for the integration of these drivers and the predictions generated are essential, for example, for climate change forecast and conservation planning. However, the reliability of biodiversity models at the local scale remains poorly understood, especially in tropical species-rich areas, where they are required. We inventoried all woody plants with stems ≥ 2.5 cm in 397 plots across the Andes-Amazon gradient. We generated and mapped 19 uncorrelated biodiversity drivers at 90 m resolution, grouped into four categories: microclimatic, microtopographic, anthropic, and edaphic. In order to evaluate the importance of the different categories, we grouped biodiversity drivers into four different clusters by categories. For each of the four clusters of biodiversity drivers, we modelled the observed species richness using two statistical techniques (random forest and Bayesian inference) and two modelling procedures (including or excluding a spatial component). All the biodiversity models produced were evaluated by cross-validation. Species richness was accurately predicted by random forest (Spearman correlation up to 0.85 and explained variance up to 67%). The results suggest that precipitation and temperature are important driving forces of species richness in the region. Nonetheless, a spatial component should be considered to properly predict biodiversity. This could reflect macroevolutionary underlying forces not considered here, such as colonization time, dispersal capacities, or speciation rates. However, the proposed biodiversity modelling approach can predict accurately species richness at the local scale and detailed resolution (90 m) in tropical areas, something that previous works had found extremely challenging. The innovative methodology presented here could be employed in other areas with conservation needs.