Species

R. crispus, also known as curled dock, is native to Europe, northern Africa (i.e., Algeria, Egypt, Libya, Morocco and Tunisia), and western Asia (i.e., Afghanistan, Iran, Iraq, Israel, Lebanon; [29]). Figure 1 shows the native range of R. crispus. This species grows in a wide variety of habitats, including disturbed soil, waste areas, roadsides, fields/meadows, shorelines, and forest edges and prefers rich, moist, and heavy soil in general. This species can be used as a wild leaf vegetable because its leaves are an excellent source of vitamin A, protein, iron, and potassium. R. crispus is propagated through the contamination of crop seeds and by sticking to clothing. It is classified as an “injurious weed” under the United Kingdom Weed Act of 1959 (http://www.defra.gov.uk/farm/wildlife/weeds/). As a widespread naturalized species throughout the temperate world, R. crispus is now present in continental Asia, Japan, North and South America, North and South Africa, Australia, and New Zealand [29],and is considered as one of the five most widely distributed plants in the world (see Figure 1 for invasive range).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Native and invasive range of R. crispus.

https://doi.org/10.1371/journal.pone.0070728.g001

T. latifolia is a perennial herbaceous plant that belongs to the genus Typha. It is a native plant species of North America (ranging from Alaska to Guatemala, as shown in Figure 2; [30]). T. latifolia grows in a variety of climates, including tropical, subtropical, southern and northern temperate, humid coastal and dry continental. This species is found at elevations ranging from 0 m.s.l. to 2300 m.s.l. As an obligate wetland species, T. latifolia is always found in or near water. It generally grows in flooded areas where the water level does not exceed 0.8 m. Traditionally, T. latifolia has been part of several native North American cultures as a source of food and medicine. Their rhizomes are edible after cooking and removing the skin. Similarly, their young flower spikes are edible. Several cultures use the roots of T. latifolia as a poultice for boils, burns, or wounds. This species often plays important roles in keeping lakes healthy by filtering runoffs. T. latifolia forms dense monocultures when a wetland disturbance occurs. It can reach up to 3 m height and can grow prolifically from thick underground rhizomes, forming dense rhizome mats and litter that may reduce the chance of survival or spreading of other plants. The invasion range of T. latifolia is shown in Figure 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Native and invasive range of T. latifolia.

https://doi.org/10.1371/journal.pone.0070728.g002

Data

Presence samples and environmental data are necessary in order to estimate the potential distribution ranges of a plant species. In the present study, presence samples of R. crispus and T. latifolia were obtained from the biodiversity data portal Global Biodiversity Information Facility (GBIF, www.gbif.org). We collected a total of 61,772 and 35,262 presence samples of R. crispus and T. latifolia, respectively. Duplicate samples may have been included (because of the resolution of environmental layers) in this data set, so we used a sample selection strategy that excluded duplicated samples (only one sample at each pixel). In addition, such a huge number of presence samples may induce overfitting. Therefore, the presence samples data obtained from GBIF must be filtered. Fortunately, the model used in this study (MaxEnt, described in the following section) fulfills this requirement. Environmental data on current climatic conditions were obtained from the WorldClim Web site (http://www.worldclim.org). The Worldclim data set includes annual time series of mean monthly data for precipitation as well as minimum and maximum temperatures recorded by more than 4,000 weather stations worldwide [31]. This data set consisted of the following 19 climatic layers: mean annual temperature, mean diurnal range, isothermality, temperature seasonality, maximum temperature of the warmest month, minimum temperature of the coldest month, temperature annual range, mean temperature of the wettest quarter, mean temperature of the driest quarter, mean temperature of the warmest quarter, mean temperature of the coldest quarter, mean annual precipitation, precipitation of the wettest quarter, precipitation of the driest quarter, precipitation seasonality, precipitation of the wettest quarter, precipitation of the driest month, precipitation of the warmest quarter, and precipitation of the coldest quarter. These layers were spatially interpolated according to the constructed relationship between recorded variables and terrain features (i.e., latitude, longitude, and elevation). Compared with other climatic data sets, the WorldClim data set exhibited the following advantages: the resolution of data layers was improved, more weather station records were used for the interpolation, and improved elevation data were used. In the present study, data layers with resolution of 2.5 arc min (approximately 5 km at the equator) were adopted. In order to predict the potential distribution of target species under future climatic conditions, future climatic layers are necessary. Future climatic conditions consisting of 19 climatic layers were downloaded from the Climate Change, Agriculture, and Food Security Web site (http://www.ccafs-climate.org). Future climatic projections included the Intergovernmental Panel on Climate Change-Special Report on Emissions Scenarios (IPCC-SRES) A1B (very rapid economic growth, global population that peaks in mid-century and declines thereafter, based on a balance across all sources), A2 (a highly heterogeneous world with continuously increasing population, economic development is primarily regionally oriented), A2A (a highly heterogeneous world with high rate of population growth, regionally oriented economies), and B1 (a convergent world with the same global population, rapid change in economic structures toward a service and information economy) for 2050 predicted by the Canadian Centre for Climate Modeling and Analysis-Third-Generation Coupled Global Climate Model (CCCMA-CGCM3), and B2A (regionally oriented economies with a general evolution towards environmental protection and social equity) for 2050 predicted by CCCMA-CGCM2 [32].

Species Potential Distribution Modeling

The potential distributions of each species were predicted using the MaxEnt model, which was developed based on the principle of maximum entropy. Under this principle, a target probability distribution can be determined by finding the probability distribution of maximum entropy (i.e., the most spread out or the one closest to a uniform distribution; [21]), which is subject to a set of constraints representing incomplete information regarding the target distribution. When this principle is applied to predict the potential species distribution at each pixel across the study area, the constraint becomes the expected value of each environmental variable which matches the empirical average [21]. The MaxEnt model is increasingly being used to model potential species distribution and has been shown to outperform other modeling approaches because of the following advantages: (1) only presence samples are required; (2) it guarantees an optimal probability distribution through an efficient deterministic algorithm; (3) it generates an output with a concise definition and is, therefore, amenable to analysis; and (4) overfitting can be effectively avoided [21], [33]. In the present study, the model was applied using the default settings [34]. The presence samples were randomly partitioned with 80% assigned to the training dataset and the remaining 20% to the testing dataset. MaxEnt uses the presence data and randomly selected points, and combines these with environmental variables to predict probability values ranging from 0 (completely not suitable) to 1 (completely suitable) for each cell. Continuous probability values (ranging from 0 to 1) can be transformed into binary (0 for predicted unsuitable; 1 for predicted suitable) values by applying a threshold. The maximum sensitivity plus specificity (MSS) approach [35] was adopted for the selection of a threshold. The MSS approach originated from the confusion matrix, which is composed of four elements denoted by a(true positive, recorded present and predicted present), b(false positive, recorded absent but predicted present), c(false negative, recorded present but predicted absent), and d(true negative, recorded absent and predicted absent). The sensitivity and specificity values were determined by calculating and , respectively. The MMS approach determined the threshold by maximizing the value of [35]. When the prediction was complete, performance of the model was evaluated using the area under the receiver operating character curve (AUC; [36]).

Relationship between Potential Distribution of Species and Climatic Variables

As mentioned previously, investigating the response of invasive species to each climatic variable is essential to understand the effect of climate change on the invasiveness of species. The MaxEnt model can estimate the contribution of each climatic variable on the potential distribution of species. In the present study, the relationship between potential distribution of the target species and climatic variables was analyzed based on the following strategies. First, the current potential distribution areas of these two species were delineated, and the relatively important climatic variables that contributed more than 10% were determined. Second, the values of these important variables at current potential distribution areas were extracted from the current and future climatic layers to obtain two datasets, namely, VPC and VPF. Next, mean minimum, mean maximum, and global mean of each variable in VPC and VPF were calculated. Third, comparisons of the differences between each pair of variables in VPC and VPF were performed at three levels (i.e. mean minimum, mean maximum, and global mean). In this way, differences in the contribution of each variable in the same geographic region under different climatic conditions (current and future) could be detected. Given that the value of climatic variables was extracted from the same geographic regions (i.e., the current potential distribution area), presence of differences between current and future potential distributions would permit examination of the effect of variation of variables on species distribution. In this study, future climatic conditions were represented by five climate scenarios (A1B, A2, A2A, B1, and B2A). Therefore, the average value of each variable among these five scenarios was calculated to determine future climatic conditions.

Why do these Climatic Variables Matter?

Temperature is one of the key factors driving species survival and species distribution. Mean annual temperature has been reported to affect not only plant species assemblage [37], but their distribution as well [38]. However, mean annual temperature by itself is insufficient for species distribution modeling, particularly when considering climate change and the associated variations in temperature rise at various regions. As a result, investigating variations in temperature-related sub-variables as well as the effect of these variations on species distribution can be a means of discovering the consequences of climate change. The effect of mean temperature during the coldest quarter, which is similar to the winter minimum temperature, has been demonstrated to be an important determinant of plant species distribution [39]. The effects of these two variables are clearly demonstrated by our investigation of the impact factor of the potential distribution of T. latifolia (Table 1). Generally, increasing these two variables can be unfavorable for the geographic distribution of plant species [40], as confirmed in the present study.

Aside from temperature-related variables, variations in precipitation-related variables and the impact of these variations on species distribution cannot be neglected [41]. It was reported that decreased summer precipitation results in expansion in land areas suitable for invasive plant species (and conversely, increased precipitation leads to habitat reduction) [12]. By contrast, our results demonstrate a different response by invasive species to variations in precipitation. As shown in Table 1, although precipitation of the coldest quarter, mean annual precipitation, and precipitation during the driest month increased under future climatic conditions compared with those under current conditions, this increase does not seem to benefit the geographic distribution of invasive plant species. Contributions of variation in precipitation seasonality (from a narrower range to a broader one) may account for this unfavorable prediction–despite an increase in precipitation, widening of seasonal distribution of precipitation may prove unfavorable for plant species invasion.

Source of Uncertainties

The future is, by definition, uncertain [42]. As a consequence, our predictions comprise inherited uncertainties from climate scenarios, data, and the manner in which species cope with climate change.

Emission scenario, global climate model (GCM), and initialization of GCM are clearly main sources of uncertainty in global climate projections [43]. First, different emission scenarios reflect different assumptions about development, which are translated into different greenhouse gas emission levels. Second, different GCMs describe climate processes and corresponding feedbacks in various ways. Third, different initialization states make various climate projections to be more or less in phase with actual low-frequency climate oscillations, thus reflecting natural climate variability. Any combination of these sources will generate a future climate that is more or less different from other combinations [44]. A possible way to reduce the uncertainty of climate scenarios might be to adopt the average prediction of future climatic conditions when modeling potential distribution of species.

Uncertainty of data originates from the knowledge that any modeling practice is sensitive to the quality and quantity of employed data, thus species distribution modeling is no exception [42]. Uncertainty of data is related to spatial and temporal resolutions of weather records, reliability and selection of species presence and absence observations, and selection of climatic variables. On one hand, spatial and temporal resolutions of weather records might have an influence on downscaling predictions of GCMs [31]. On the other hand, the finer the spatial resolution of climatic variables is, then the more the micro-climatic features of the climatic conditions could be defined; the coarser the spatial resolution of the climatic variables is, then the more the macro-climatic conditions could be depicted. Second, the reliability of species presence and absence may have an effect on the quantified relationship between species and climatic variables [45]. The selection of presence samples represents the sampling bias of the modeling, and as a result, may influence the prediction of the potential distribution range of subject species. As shown in Figures 3 and 4, the presence records we selected for the modeling procedure are relatively concentrated in Europe and North America. Although such a selection is reasonable for guaranteeing “intact and unbiased sampling”, it may underestimate potential distribution of invasive species at successfully invaded areas because samples in invaded regions are not efficiently used for constructing the relationship between species and environmental variables [46]. Finally, different variables depict environmental conditions in different ways. Climate (such as mean mean annual precipitation, mean annual surface temperature), topography (such as altitude and slope), water availability (such as mean relative air humidity and topographic wetness index), productivity (such as mean annual actual evapotranspiration), human activities (such as distance to the nearest town; [47]), surface condition (such as land use and land cover; [48]), and soil feature (such as fertility, texture and pH; [49]) all have effects on the distribution of invasive species. When selecting variables, the auto-correlation among them needs to be considered. Aside from these abiotic factors, biotic variables also need to be considered [50].

The manner in which a species copes with climate change may be synthesized as evolutionary adaption, dispersal, and extinction [51]. Studies that focused on the response of species to projected climate change assume that climate change outpaces micro-evolutionary processes, and therefore, species have no time to adapt [52]. As a result, the only way for these species to avoid extinction is to move to suitable areas. This finding means that geographic dispersal can be an alternative strategy with which species can maintain reproduction of their populations. In fact, dispersal has been identified as an important response of species to climate change, usually via range shifting to suitable geographic regions. Unfortunately, in the present study, we found that no projected areas are suitable for our target species (Rumex crispus and Typha latifolia) to disperse to. As such, if these species could not adapt to future climatic conditions, their only destiny would be extinction. By contrast, recent studies have pointed out that evolutionary changes can rapidly take place in several species, especially for species that have invaded new areas [53], [54]. This result indicates that evolutionary adaptation can be an important technique for natural populations to counter rapid climate changes [55].

Conclusions

In this study, we predicted the potential distribution of two invasive species (Rumex crispus and Typha latifolia) under current and future climatic conditions by comparing predicted potential distributions. We found that compared with the relatively suitable current climatic conditions, these two species will undergo harsh climatic challenges in the future. In the worst-case scenario, these species will not only lose areas to invade, but will also face possible extinction. We then analyzed possible reasons for such results by extracting important climatic variables and found that warmer and wetter conditions during the coldest season (or month) mainly determine harsh conditions for these invasive plant species. We finally discussed existing uncertainties during our modeling approach. We suggest that climate scenarios, data (climatic variables and species distribution records), and the manner in which species cope with climate change may influence results related to invasion of species, and need to be further investigated in order to reduce the alien species invasion risk under climate change.