Climate is the main factor determining the distribution pattern of plants, and changes in this distribution pattern are a direct reflection of climate change [1
]. The Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC) indicates that the annual mean temperature of the Earth’s surface has increased by 0.85 °C over the past 130 years (1880–2012). While the mean temperature of the past 60 years (1951–2012) has increased by 0.72 °C. The mean global surface temperature is predicted to increase by an additional 0.3–4.8 °C by the end of the 21st century, according to simulations of greenhouse gas emission scenarios [6
]. Several observational studies have shown that climate change has caused or is causing significant changes to ecological processes, including distribution ranges, morphological characteristics, and phenology that are accelerating species extinction or prosperity [7
]. Because of this, quantitative analysis of climate factors and predictions of climate change impacts on potential suitable species distributions have become key to biogeography and global change research [8
Niche modelling (also known as species distribution modelling or habitat suitability modelling) uses statistical or biophysical methods to infer the environmental requirements for the survival of a species based on the environmental variables taken from the known distribution areas, and then predicts their potential or future distribution [12
]. Niche modelling has been widely used in ecology, biogeography, and conservation biology studies, such as the adaptive analysis of invasive alien species [15
], conservation of endangered species [17
], selection of natural reserves [19
], and impacts of climate change on species distribution [21
]. At present, there are many niche models used for species distribution prediction, including the bioclimate analysis and prediction system (BIOCLIM) [24
], the genetic algorithm for rule set production (GARP) [25
], the ecological niche factor analysis (ENFA) [13
], and the Maximum Entropy Model (Maxent) [26
]. These niche models, established on the basis of the statistical relationship between species distribution and environment variables (rather than the causality), only consider the environmental factors that control species distribution and do not consider the interactions between species, species evolution, extreme interference events, or species diffusion processes [12
]. The current correlation between species and the environment (climate) does not necessarily guarantee that it can be used to predict future species distribution; however, niche modelling is still the most commonly used method for assessing the impacts of climate change on species distribution at the regional scale. Niche modelling is required to predict changes in species distribution, due to the difficulty of effectively carrying out field experiments at the geographical distribution scale of an entire species [28
]. Niche models can be easily constructed and can provide rapid predictions for the changes to species distribution and the risk of habitat loss.
The Maxent model, which is based on the theory of maximum entropy, is an effective model of species distribution. The model simulates the potential geographical distribution of a species using information on its current (present day) distribution as well as various environmental data [27
]. The Maxent model is relatively simple and quick to run, has a small sample demand, provides stable operation results, and allows prediction results to be tested [27
]; thus, it is favored by many researchers [14
]. However, the model is often employed using the default parameters or those published previously without consideration of the details of the algorithm or input parameters, and all the environment variables that can be collected are included in the model indiscriminately. This not only increases the complexity and operation time of the model, but also greatly limits its scope of application and accuracy. Previous studies have shown how different environmental variables and parameters used in niche models affect the results [35
]. Complex models show low performance in identifying the most important environmental variables and in transferring habitat suitability to new environmental conditions [38
]. When using the default parameters of the Maxent model to predict the potential suitable distribution of species, overfitting can occur, leading to a decline in species-transfer ability [41
]. Therefore, the selection of appropriate environmental variables and model parameters is an important step that must be considered in the application of the Maxent model.
Cunninghamia lanceolata (Lamb.) Hook is a fast-growing and high-yielding tree that is native to the subtropical region of South China, where it is the main species used in afforestation and for timber production. It is a long-lived evergreen tree and plays an important role in regional ecological construction. Therefore, it is important to study the impact of climate change on the suitable habitat of C. lanceolata. In this study, the Maxent model is used to simulate the suitable distribution of C. lanceolata and how this might change under different climate change scenarios after parameter optimization and climate variable selection. The specific objectives of this study are as follows: (1) to explore the influence of environmental variable selection and parameter settings on the performance of the Maxent model; (2) to understand the importance of environmental variables in predicting the potential suitable distribution of C. lanceolata; and (3) to compare the differences in the predicted distribution of C. lanceolata under different climate change scenarios in different time periods. Our optimized model, based on the variables with the highest explanatory power, reveal the climate conditions required by C. lanceolate that account for its present-day distribution. The model also suggests marked changes to this distribution pattern under predicted climate change scenarios and increased fragmentation of suitable habitat by the 2070s.
Niche modelling is a powerful tool for the prediction of potential species distribution that can help inform applied ecology. However, there are several limitations to niche modelling, such as low transfer ability and discrepancies between the simulation results and the actual species distribution, that can result in an inappropriate conclusion being drawn from the models. Improving the transfer ability of niche models not only helps to accurately simulate the potential species distribution, but can also generate important reference data for theoretical issues such as niche conservation.
Maxent has been widely used in the prediction of potential distribution of species in recent years [34
]. The model assumes that species will be present in all areas with suitable climate conditions and will not present in any area with unsuitable conditions; it is believed that the larger the entropy of a species under known conditions, the closer it is to the reality [27
]. The default parameters of Maxent were originally set through the testing of data from 266 species in six different geographical regions by early model developers [30
]. They used the massive distribution data of the test species and a variety of experimental schemes to define optimal model parameters as the default to promote and simplify the application of the model; however, the purpose of application of these models was to predict the present-day distribution of a species, and the model did not have to be transferable after construction [30
]. Adjustments to the simulation scheme are necessary to meet different simulation requirements depending on the research question, although the species distribution simulated by the Maxent model is between the potential distribution and the real distribution [86
]. To simulate the potential distribution of a species, the default parameters of the Maxent model may not be applicable. For example, the parameters of our optimal model differ from the default parameters (Figure 3
, Table 2
). If only the default parameters are used, the transfer of the model will be reduced because of overfitting, although it can fit the current distribution of the species for the study area well.
The complexity of a Maxent model has a significant influence on its transfer ability. Both simple and complex models have advantages and disadvantages. Simple models may show a poor fit to the species distribution data of the study area, and while complex models can produce a better fit, the response relationship between species and environmental variables can deviate from the reality. The adverse effect of Maxent model complexity on prediction results when simulating potential distributions is more serious because of the need to transfer the model to other geographical areas after model construction [38
]. The prediction ability of complex models will be reduced, and the simulation results may not be reliable or can deviate from the actual situation, because of overfitting when they are transferred to other geographical areas.
In this study, steps were taken to optimize the two main factors known to influence the prediction results of the Maxent model. First, as much species distribution data as possible were collected to reduce error caused by incomplete data collection; a total of 1877 sample points were collected. Generally, the species distribution points should not only adequately cover the geographical distribution and ecological space of a species, but also reduce the inaccuracy of sampling points. It has been shown that the sample size of species distribution data significantly affects the accuracy of model simulation results [87
]. In this study, the modeling data covered the southern provinces of China and only one point was reserved in the same pixel. Second, we optimized the parameters of the model. We constrained the complexity of the model by reducing the number of variables through variable selection and using AICc values to select the optimal combination of feature types and RM. Along with decreasing the complexity of the model, reducing the number of variables can reduce the operation time and improve the interpretability. The Maxent model uses the default criterion (i.e., the AUC) to evaluate the prediction ability of the model, but the AUC may overestimate this ability and directly affect the species distribution results when test data and training data are affected by sampling error [40
]. Nevertheless, using the AICc to select the feature type and the built-in RM can constrain the complexity of model [38
Five factors (i.e., climate change, habitat change, overexploitation, pollution, and invasive alien species) are viewed as the principal threat to global biodiversity in the 21st century [92
]. As an important aspect of global change, climate change mainly manifests as temperature rises, changes to precipitation patterns, and increasing extreme climate events; in addition, changing climate factors will also affect the growth and reproduction, phenology, and geographical distribution of species at the regional scale [1
]. Current biodiversity conservation measures, such as the establishment of nature reserves, are usually based on the current distribution of species, but are facing increasing challenging as the effects of climate change increase [20
]. It is of key importance to understand changes in suitable areas of distribution under the future climate change scenarios, and implement targeted conservation measures as early as possible to improve the effectiveness of biodiversity conservation [29
]. With the exponential growth of global greenhouse gas emissions, the trend of climate warming in China will be further intensified in the future. The results of this study show that climate change will have a significant impact on the suitable distribution area of C. lanceolata
, and the suitable habitat area will likely gradually decrease. In addition, we found that the prediction results using RCP4.5 data were different from other climate scenarios. Previous studies have shown that the climate scenario of RCP4.5 is the closest to the actual predicted climate situation in China [55
]; thus, the suitable distribution area of C. lanceolata
has not been significantly reduced over a short time period under this climate scenario. However, the suitable distribution area of C. lanceolata
will still likely see a large reduction with the likely effects of climate change caused by continuing increases in greenhouse gas emissions.
Climate change, especially global warming, not only causes temperature changes in different regions, but also changes the distribution pattern of precipitation. When the change of these climatic factors is close to or beyond the adaptive threshold of current plant growth, it will lead to the migration of their distribution [7
]. Water availability has a significant impact on plant height, leaf area, branch number, photosynthesis and growth [96
], thus, precipitation can constrain species distribution and influence distribution in various ways [97
]. The increase of precipitation during the driest month results in a longer growth season and helps species migrate to more suitable habitats within their distribution [98
]. In addition, extreme high and low temperatures also have a significant influence on plant growth. The decrease of the minimum temperature of the coldest month results in premature freezing injuries to plants, and long-term low temperatures will lead to the death of plants at the distribution limit [97
]; while the increase of the maximum temperature of the warmest month will destroy the water balance of plants, promote the coagulation of proteins and the internal mechanism of harmful metabolites, thus hindering plants’ growth [99
]. Climate extremes have been shown to have important impacts on species’ distribution and diversity, and adding extreme climate variables in niche modelling can improve the predicted accuracy and limit of species distributions under future climate change scenarios [100
Because of changes in environmental factors, particularly those of the climate, the suitable habitat of tree species has changed in China, which makes it difficult to correctly formulate a long-term forest management plan. Sustainable forest management planning at the regional scale includes predicting the spatial distribution of different types of forests and forest management division; it is therefore a spatial decision-making process. The impact of forest management on ecosystems and the evaluation of forest management effectiveness are long-term processes; therefore, a long-time scale must be adopted to measure the sustainability of forest management. When developing forest management plans, optimized model simulations and decision analysis of the management process should be carried out over at least one forest management cycle. During the process of long-term (>30-year) forest management planning, climate change will lead to changes in the environment, and thus suitable habitats, for tree species. Therefore, it is necessary to consider the impact of climate change in the process of long-term forest management planning. The Maxent model is a key tool for simulating the distribution of suitable habitats of tree species, and therefore predicting suitable distribution areas, under future climate change scenarios. Our optimized model can provide vital reference data for the selection of C. lanceolata afforestation site in the formulation of long-term forest management planning.
Accurate simulation of species distribution areas is of key importance for species protection and restoration. However, many biological and non-biological factors must be considered to achieve this goal. In practice, important factors are often ignored to simplify the data needed for the model for the feasibility of the research, which affects the operation results to a certain extent, leading to errors between the predicted results and reality. Different simulation results of a species may be generated because of different sample counts and parameters applied to the model by different researchers, even for the same species. Liu et al. used the data from the counties in the subtropical area in China as distribution sample sites and the WorldClim data with 2.5’ (approximately 16 km2
) spatial resolution in combination with BIOCLIM model to predict the potential distribution area under the present and future climate conditions for C. lanceolata
. The results indicated that the suitable distribution area would shrink along with distinct fragmentation when the greenhouse gas emission was doubled [105
]. While Lu et al. used the process-based growth model and the WorldClim data with 2.5´ spatial resolution as well as the digital version of Vegetation Map of China (1:1,000,000) to map the distribution and productivity of C. lanceolata
. The results showed that a species is likely to experience a northward shift with minor changes in the south, and the central regions of China are likely to become more suitable for C. lanceolata
under future climate conditions [106
]. By contrast, with the optimized Maxent model, we not only obtain the potential distribution results with high accuracy and high spatial resolution, but also improve the transfer ability of the Maxent to predict the potential distribution of C. lanceolate
in the future climate scenarios. Besides climate factors, there are many variables that affect the distribution of C. lanceolata
, including land use, soil conditions, and topography. Additionally, human activity is an important factor affecting species distribution and the key factor determining the actual niche of a species. While human activity directly changes the habitat characteristics of species, we can also implement measures that can also affect how species distribution is influenced by climate change. We did not consider these factors because there are no suitable data on soil and land use under different greenhouse gas emission scenarios. The impact of climate change on soil and land use change is currently poorly understood, and the interaction between human activity and climate change is complex. However, as more data on variables affecting species distributions become available, niche models can better predict the likely effects of climate change.