Soil factors improve predictions of plant species distribution in a mountain environment

Abstract

Explanatory studies suggest that using very high resolution (VHR, 1–5 m resolution) topo-climatic predictors may improve the predictive power of plant species distribution models (SDMs). However, the use of VHR topo-climatic data alone was recently shown not to significantly improve SDM predictions. This suggests that new ecologically-meaningful VHR variables based on more direct field measurements are needed, especially since non topo-climatic variables, such as soil parameters, are important for plants. In this study, we investigated the effects of adding mapped VHR predictors at a 5 m resolution, including field measurements of temperature, carbon isotope composition of soil organic matter (δ13CSOM values) and soil pH, to topo-climatic predictors in SDMs for the Swiss Alps. We used data from field temperature loggers to construct temperature maps, and we modelled the geographic variation in δ13CSOM and soil pH values. We then tested the effect of adding these VHR mapped variables as predictors into 154 plant SDMs and assessed the improvement in spatial predictions across the study area. Our results demonstrate that the use of VHR predictors based on more proximal field measurements, particularly soil parameters, can significantly increase the predictive power of models. Predicted soil pH was the second most important predictor after temperature, and predicted δ13CSOM was fourth. The greatest increase in model performance was for species found at high elevation (i.e. 1500–2000 m a.s.l.). Addition of predicted soil factors thus allowed better capturing of plant requirements in our models, showing that these can explain species distributions in ways complementary to topo-climatic variables. Modelling techniques to generalize edaphic information in space and then predict plant species distributions revealed a great potential in complex landscapes such as the mountain region considered in this study.

Identifying multidisciplinary research gaps across Arctic terrestrial gradients

“…Data are split internally multiple times into training and evaluation sets, and trees are built recursively using the information from previous trees (Elith et al 2008). GBMs have been widely used in environmental science research (Marmion et al 2009, Buri et al 2017, Nussbaum et al 2018, because they consider interaction effects between predictors and can model non-linear relationships (Elith et al 2008). We used the 'Bernoulli' error distribution of the response variable as we were working with a binomial presence-absence data (1=sampling location exists, 0=sampling location is missing), and soil (SOC, pH, MAGT), vegetation (NDVI) and topography (DEM, TWI) as explanatory variables.…”

Learning from model errors: Can land use, edaphic and very high‐resolution topo‐climatic factors improve macroecological models of mountain grasslands?

“…Cornelissen et al., ), and the inclusion of edaphic factors has been demonstrated to improve the quality of predictions of SLA (Dubuis et al., ). In a previous study, two soil chemical properties, pH and carbon isotopic ratios, were predicted across the geographic area (Buri et al, In press), and additional maps are currently being developed for other soil properties. If the C/N ratio could be similarly mapped, C/N ratio and pH would provide high potential for model improvement, especially for SLA.…”

Site fertility drives temporal turnover of vegetation at high latitudes

“…However, it is not known whether plant strategy-based approaches provide any useful predictions for plant community changes under climate warming across broader spatial scales. Despite advances in theory, modeling, and experimentation (Buri et al, 2017;Díaz et al, 2016;Pellissier et al, 2018;Reich, 2014), this remains an unresolved issue, recently been stressed also by Parmesan and Hanley (2015).…”