Abstract Trees provide ecosystem services that improve the environment and human health. The magnitude of these improvements may be related to tree diversity within green spaces, yet spatially explicit diversity data necessary to investigate such associations are often missing. Here, we evaluate two methods to model tree diversity at genus level based on environmental covariates and presence point data. We aimed to identify the drivers and suitable methods for urban and rural tree diversity models in the heterogeneous region of Flanders, Belgium. We stratified our research area into dominantly rural and dominantly urban areas and developed distribution models for 13 tree genera for both strata as well as for the area as a whole. Occurrence data were obtained from an open-access presence-only database of validated observations of vascular plants. These occurrence data were combined with environmental covariates in MaxEnt models. Tree diversity was modelled by adding up the individual species distribution models. Models in the dominantly rural areas were driven by soil characteristics (soil texture and drainage class). Models in the dominantly urban areas were driven by environmental covariates explaining urban heterogeneity. Nevertheless, the stratification into urban and rural did not contribute to a higher model quality. Generic tree diversity estimates were better when presences derived from distribution models were simply added up (binary stacking, True Positive Rate of 0.903). The application of macro-ecological constraints resulted in an underestimation of generic tree diversity (probability stacking, True Positive Rate of 0.533). We conclude that summing presences derived from species distribution models (binary stacking) is a suitable approach to increase knowledge on regional diversity.

中文翻译：

---

物种分布模型评估在异质城乡景观中在属水平上估计树木多样性

摘要 树木提供改善环境和人类健康的生态系统服务。这些改进的幅度可能与绿色空间内的树木多样性有关，但通常缺少调查此类关联所需的空间明确的多样性数据。在这里，我们评估了两种基于环境协变量和存在点数据在属级对树木多样性进行建模的方法。我们旨在确定比利时佛兰德异质地区城市和农村树木多样性模型的驱动因素和合适的方法。我们将我们的研究区域划分为以农村为主和以城市为主的区域，并为两个层以及整个区域开发了 13 个树属的分布模型。发生数据是从维管植物验证观察的开放访问数据库中获得的。这些发生数据与 MaxEnt 模型中的环境协变量相结合。通过将单个物种分布模型相加来模拟树木多样性。主要农村地区的模型由土壤特征（土壤质地和排水等级）驱动。主要城市地区的模型由解释城市异质性的环境协变量驱动。然而，城市和农村的分层并没有提高模型质量。当来自分布模型的存在被简单地相加时，通用树多样性估计会更好（二进制堆叠，真阳性率为 0.903）。宏观生态约束的应用导致对一般树木多样性的低估（概率叠加，真阳性率为 0.533）。