Data from: Predicting spatial patterns of plant species richness: a comparison of direct macroecological and species stacking modelling approaches

PLEASE NOTE, THESE DATA ARE ALSO REFERRED TO IN TWO OTHER PUBLICATIONS. PLEASE SEE http://dx.doi.org/10.1111/j.1365-2486.2008.01766.x AND http://dx.doi.org/10.1111/2041-210X.12222 FOR MORE INFORMATION. Aim: This study compares the direct, macroecological approach (MEM) for modelling species richness (SR) with the more recent approach of stacking predictions from individual species distributions (S-SDM). We implemented both approaches on the same dataset and discuss their respective theoretical assumptions, strengths and drawbacks. We also tested how both approaches performed in reproducing observed patterns of SR along an elevational gradient.
Location: Two study areas in the Alps of Switzerland. Methods: We implemented MEM by relating the species counts to environmental predictors with statistical models, assuming a Poisson distribution. S-SDM was implemented by modelling each species distribution individually and then stacking the obtained prediction maps in three different ways – summing binary predictions, summing random draws of binomial trials and summing predicted probabilities – to obtain a final species count. Results: The direct MEM approach yields nearly unbiased predictions centred around the observed mean values, but with a lower correlation between predictions and observations, than that achieved by the S-SDM approaches. This method also cannot provide any information on species identity and, thus, community composition. It does, however, accurately reproduce the hump-shaped pattern of SR observed along the elevational gradient. The S-SDM approach summing binary maps can predict individual species and thus communities, but tends to overpredict SR. The two other S-SDM approaches – the summed binomial trials based on predicted probabilities and summed predicted probabilities – do not overpredict richness, but they predict many competing end points of assembly or they lose the individual species predictions, respectively. Furthermore, all S-SDM approaches fail to appropriately reproduce the observed hump-shaped patterns of SR along the elevational gradient. Main conclusions: Macroecological approach and S-SDM have complementary strengths. We suggest that both could be used in combination to obtain better SR predictions by following the suggestion of constraining S-SDM by MEM predictions.

请注意，本数据集另有两篇相关出版物提及。更多信息可通过以下链接查阅：http://dx.doi.org/10.1111/j.1365-2486.2008.01766.x 与 http://dx.doi.org/10.1111/2041-210X.12222。

研究目标：本研究将用于物种丰富度（Species Richness, SR）建模的直接式宏观生态学建模方法（MEM），与近年提出的堆叠单物种分布预测方法（S-SDM）进行对比。我们基于同一数据集对两种方法均进行了实现，并探讨了各自的理论假设、优势与局限性。此外，我们还检验了两种方法在重现沿海拔梯度的物种丰富度观测模式方面的表现。

研究区域：瑞士阿尔卑斯山区的两处研究样地。

研究方法：我们以泊松分布（Poisson distribution）为假设前提，通过统计模型将物种计数与环境预测因子相关联，以此实现MEM方法。S-SDM方法则通过逐一建模每个物种的分布，随后以三种不同方式对所得预测图谱进行堆叠：对二分类预测结果求和、对二项分布试验的随机抽样结果求和，以及对预测概率求和，最终得到物种丰富度的估计值。

研究结果：直接式MEM方法可生成近乎无偏的预测结果，其预测值围绕观测均值分布，但相较于S-SDM方法，该方法的预测值与观测值之间的相关性更低。此外，该方法无法提供具体物种身份相关信息，因此也无法揭示群落组成。但该方法能够准确重现沿海拔梯度观测到的物种丰富度驼峰状分布模式。采用二分类预测图堆叠的S-SDM方法可预测单个物种及其群落组成，但往往会高估物种丰富度。另外两种S-SDM方法——基于预测概率的二项分布试验求和法，以及预测概率求和法——均不会高估物种丰富度，但前者会输出大量相互竞争的群落组装终点，后者则无法保留单个物种的预测结果。进一步而言，所有S-SDM方法均无法恰当重现沿海拔梯度观测到的物种丰富度驼峰状分布模式。

主要结论：宏观生态学建模方法与S-SDM方法具备互补性优势。我们建议可将二者结合使用，参考通过MEM预测结果约束S-SDM模型的思路，以获得更精准的物种丰富度预测结果。