The WorldClim dataset provided global climate surface that had been widely used in many previous SDMs studies as bioclimatic predictors and had accurately projected species distribution range at continental regions (Abdelaal et al. 2019; Datta et al. 2020; Heikkinen et al. 2012; Mi et al. 2017; Mohapatra et al. 2019). In this study, projection of potential distribution range and quantification of climate space was available at continental scale and was evidently the advantage of the WorldClim dataset. Model projections aims to project geographical extent, identify suitable key sites, and find suitable environmental factors based on the presence/absence data and climate variables that can provide scientific information for developing conservation and management strategies at continental scale (Abdelaal et al. 2019; Hu et al. 2017; Lobo et al. 2010; Maria & Udo 2017; Mi et al. 2017; Mohapatra et al. 2019; Williams et al. 2009). Modeling efforts to asses potential distribution range had been made for suggesting conservation areas of plant species in China (Wan et al. 2017; Xu et al. 2021; Yu et al. 2017; Zhang et al. 2017). However, projection map based on the WorldClim would be the disadvantage at landscape scale. Restricted elevation ranges and isolation and fragmentation of suitable habitat were not presented by the model projection in CEA. In addition, projection map based on the WorldClim dataset presented gridded squares at the edge of the potential distribution range in CEA and that also presented gridded squares in NTWN. Gridded squares of the WroldClim dataset was evidently unavailable to reflect climate heterogeneity induced by elevation and topography and gridded distribution pattern is an unrealistic distribution pattern far from empirical distribution of the plant species. Model predictions based on the global climate dataset had evidently generated bias projection map and provided misleading results of species geographical distribution at landscape scale.
To avoid misleading result caused by the WorldClim dataset, local climate dataset was suggested to apply as bioclimatic predictors of model prediction at landscape scale. The projection maps of the plant species in NTWN based on the WorldClim and local climate datasets were verified in the filed based on expert knowledge and field examination had identified that projection map based on local climate dataset was more close to empirical distribution pattern of the plant species. Accordingly, model performance was much better when model was calibrated by local climate dataset at landscape scale. Bioclimatic predictors from local climate dataset had precisely reflected climate characteristics induced by elevation and topography (Fig. 7) and were suggested to use for model projection to guarantee the accuracy of SDMs performance in mountainous area. Model projections at landscape scale is the advantage of local climate dataset. However, local climate dataset generated huge number of gridded cells within a local geographical area and there were more than 0.4 million gridded cells in NTWN that is only 1,041 km2. Because of huge number of gridded cells, it is very difficult to expand the geographical range of SDMs study based on local climate dataset. Thus, model prediction based on local climate dataset would be the disadvantage at continental scale.
On the other hand, previous studies had proposed that RF model with low error rate (ERROOB) were considered to have more accurate performance. Our study had confirmed that low error rate (ERROOB) can be achieved even when model did not project accurate distribution range and projection map was evidently far from empirical distribution pattern of the plant species. Low error rate (ERROOB) did not guarantee an accurate projection map of species distribution.
RF model performance might be affected by bias collections and might have over-fitted to local conditions. Bias collection of presence data commonly proposed by many previous studies (Datta et al. 2020; El-Gabbas & Dormann 2018; Ferro & Flick 2015; Lannuzel et al. 2021; Tomlinson et al. 2020), since comprehensive collection is a great challenge to be conducted in field survey. Bias collection may lead to bias projection of distribution. Bias projection maps in continental scale is unavailable to be corrected in the field, whereas that at landscape scale can be corrected in the field examination by expert knowledge. Field examination based on the projection map is helpful to compensate bias collection of presence data by correcting presence and absence areas of the projection results at landscape scale. On the other hand, over-fit may lead to loss of generality of the model and may have poor performance when this method is expanded to other regions or other species. However, over-fit may not be a deficit of this method, since a previous study had accurately predicted species distribution range at landscape scale based on the same method (Liao & Chen 2021). More studies were expected to identify the generality of model prediction of this method. Despite of, this method generated precise distribution map of the plant species in mountainous area at landscape scale that is very close to its empirical distribution range and is very useful for further applications.
Marginal population of continental species had showed striking genetic differences from central populations in neighboring continent (Sexton et al. 2009; Thompson et al. 2005; Wake et al. 2009; Wu et al. 2001), while this study had further identified ecological differentiation between marginal and central populations. B. sinensis had locally adapted to climate environments in NTWN. Water availability had evidently played as a major climate factor related to the potential distribution range of the B. sinensis in NTWN and the plant species was evidently, locally adapted to extraordinary high precipitation in NTWN. Local adaptation of marginal population to novel habitats at geographical margin is akin to niche evolution (Sexton et al. 2009). Climatic characteristics of marginal population distinguished from central populations may consequently contribute to the genetic variation. Thus, marginal population on continental islands warrant a high priority in biodiversity conservation (Kier et al. 2009; Weigelt & Kreft 2013), since genetic variation (Li et al. 2016; Wang et al. 2018; Wu et al. 2001) and ecological differentiation both existed between central and marginal populations.
The growing impacts of climate change on plant species calls a request to evaluate geographical extent where species with narrow distribution ranges exist or likely exist in order to enhance their conservation and restoration (Abdelaal et al. 2019). However, designation and effective management of conservation areas is a great challenge on islands because anthropogenic disturbances caused fragmentation and isolation of natural habitats and, most importantly, the complex mosaics of natural and artificial ecosystems (Kier et al. 2009). Conservation and management planning, such as the selection of representative conservation sites, may critically depend on the detailed knowledge of empirical species distribution patterns (Allouche et al. 2006; Kier et al. 2009). Model predictions based on local climate dataset had illustrated projection map accurately showing distribution patterns of plant species along elevation and topography that is useful to delineate conservation area in fragmented habitat at landscape scale. In fact, projection map of B. sinensis in NTWN had provided useful information for delineating effective conservation area to protect isolated populations in fragmented habitats.
This study suggested a two-step procedure to identify climate characteristics of central and marginal populations of plant species at continental scale and project local distribution of marginal population at landscape scale. Climate characteristics of plant species at continental scale was available to be projected based on the WorldClim dataset. Meanwhile, local climate dataset apparently provides more reliable basis for constructing climate space and delineate effective conservation area at landscape scale. The method proposed in this study is also a powerful tool to correlate climate factors with species distribution at landscape scale that offer detailed geographical and ecological information for assessing the impact of future climate change on shifting distribution range of plant species.