Changes in Ginkgo Biloba’s Habitat Due to Climate Change in China

1. Introduction

Ginkgo (Ginkgo biloba), known as the "living fossil" of the plant kingdom, belongs to the family Ginkgoaceae and the genus Ginkgo. It is considered a relict species from the Mesozoic era [1]. Ginkgo biloba is listed as an endangered (EN) species on the IUCN Red List, a nationally protected wild plant, and holds high medicinal, timber, and nutritional value. Its distribution range is relatively wide, extending from northern to northeastern China, and from the south to Guangzhou, with a more concentrated presence in the middle and lower reaches of the Yangtze River. Research on Ginkgo has mainly focused on its physicochemical properties [2], clinical effects [3], and chemical composition [4]. However, studies on its habitat and potential habitat are scarce. In recent years, due to the rising demand and economic value of Ginkgo, over-exploitation has become evident, severely threatening wild Ginkgo resources [5]. Given the increasing threats to wild Ginkgo populations, exploring the potential geographic distribution of Ginkgo, and predicting its future range under climate change, is crucial for the conservation of wild Ginkgo resources and selecting sites for artificial planting.

Climate change plays a key role in shaping the spatial distribution patterns of plant populations. Recent global temperature increases [6], driven by human energy use and geological changes, along with frequent extreme weather events, have impacted plant growth, geographic distribution, and ecosystem stability [7,8]. Rapid climate change can lead to the contraction of spatial ranges or even the extinction of plants with low adaptability and poor dispersal abilities. Therefore, understanding the effects of climate change on plant distribution and niche areas is vital for biodiversity conservation and maintaining ecosystem stability. Ecological niche models (ENMs [9]) are widely used to study species' suitable habitats, with the Maximum Entropy Algorithm model (MaxEnt) being one of the most popular due to its ease of use, accurate predictions, and reliability [10,11,12]. MaxEnt has been employed to predict potential suitable areas for various high-value species, including relict and endangered plants such as Sophora moorcroftiana, Sophora japonica [13], and Pterocarpus marsupium [14,15]. This model evaluates the effects of climate change on plant habitats by calculating the correlation between species' spatial distribution points and environmental variables, making it a robust tool even when dealing with incomplete data or small sample sizes [16]. In addition to predicting suitable habitats, the MaxEnt model can generate spatial maps [17] of these areas and assess the contribution of environmental variables using the Jackknife test. This model has been applied in various fields, including species distribution modeling, climate change response, biodiversity conservation, and evolutionary studies.

In summary, this study employs the MaxEnt model to analyze the spatial pattern of Ginkgo distribution under present and future climatic conditions using climatic and topographic data. The goal is to identify the key environmental factors influencing Ginkgo's growth and provide theoretical references for the strategic planting and conservation of this valuable species in China. The study seeks to answer three primary questions: (1) What are the main environmental factors affecting Ginkgo distribution? (2) How will the potential habitable zone of Ginkgo shift under climate change? (3) How can Ginkgo be reasonably protected as a rare plant?

2. Materials and Methods

2.1. Species Range Point Data

Ginkgo distribution data were obtained from several sources, including the Global Biodiversity Information Facility (GBIF), the Chinese Digital Herbarium (CVH), and the Chinese National Specimen Resource Platform (NSII). The locations of Ginkgo biloba were extracted from these databases, and latitude and longitude data were collected using the Baidu coordinate system. After eliminating duplicates and invalid data, the buffer method in ArcGIS was used to retain one distribution point within a 5 km radius, resulting in 118 valid distribution data points.

2.2. Selection of Environmental Data

Plant growth and distribution depend heavily on their adaptability to environmental factors such as climate and topography. Therefore, 19 bioclimatic factors and 3 topographic factors were selected for analysis. Climatic variables were obtained from the Global Climate Database (Worldclim), and topographic data were downloaded from the Geospatial Data Cloud. Slope and slope direction were extracted using ArcGIS. Future climate data were based on the BCC-CSM2-MR model, developed by the National Meteorological Center of China, under three climate scenarios (SSPs126, SSPs245, SSPs585 [18]) for the 2030s, 2050s, and 2070s, with a resolution of 2.5 minutes.

2.3. Model Construction and Accuracy Test

2.3.1. Screening of Environmental Factors

To avoid overfitting the model due to correlations between environmental factors, the distribution data and environmental variables were input into the MaxEnt model for pre-modeling. This process identified the contribution rates of each environmental factor and excluded those with a contribution rate of zero. Pearson correlation analysis was then performed on all environmental factors using SPSS 25.0 (Figure 1). Based on the correlation analysis and the table of environmental factor contribution rates, factors with |r| < 0.8 were retained, while factors with |r| > 0.8 were retained only if they had high contribution rates. This final selection yielded 11 environmental factors for modeling (Table 1). A correlation heat map of these environmental factors was plotted using the corrplot package in R.

2.3.2. Model Construction and Optimization

Given that the MaxEnt model can become overfitted due to complex operations, resulting in unreliable or difficult-to-interpret predictions, model complexity was addressed. Complexity is significantly correlated with the feature combination (FC)—LQPTH (L = linear, Q = quadratic, P = product, T = threshold, H = hinge)—and the regularization multiplier (RM). Therefore, the MaxEnt model was tuned using the "kuenm" package (R3.6.3) to ensure optimal combinations of RM and FC parameters. The best parameter combination was selected when the omission rates were less than 5% and the natural logarithm of AICc was minimized, corresponding to a delta AICc value of zero.

The other model parameters were set as follows: 75% of the species distribution data was randomly selected as the training set, and 25% as the test set. The output format was set to Logistic, and results ranged between 0-1, facilitating the delineation of fitness zones in subsequent analyses. The model was run 10 times, with a maximum background value of 10,000, and the remaining parameters were set to default.

The accuracy of the prediction results was evaluated using the area under the receiver operating characteristic curve (ROC) (AUC). Generally, the higher the AUC value, the better the model's predictive accuracy [19]. AUC values between 0.5-0.7 indicate low accuracy, 0.8-0.9 moderate accuracy, and 0.9-1.0 high accuracy. In this study, the average AUC value for Ginkgo biloba in China predicted by the MaxEnt model was 0.928, with all AUC values exceeding 0.9, indicating reliable predictions that can be used to estimate Ginkgo's potential distribution in China.

2.4. Classification of Suitable Areas

The predicted species distribution data were imported into ArcGIS, where the predicted result values (P) ranged from 0 to 1. Higher values indicate a greater likelihood of the species’ presence. Using ArcGIS's reclassification function, Ginkgo's habitat was divided into four categories: non-habitable (P < 0.09), low-habitable (0.09 ≤ P < 0.27), moderate-habitable (0.27 ≤ P < 0.48), and highly-habitable (P ≥ 0.48).

3. Results

3.1. Model Optimization and Accuracy Evaluation Results

The model optimization results involved cross-validating 1,160 FC and RM parameter combinations, all meeting the 5% omission rate threshold. Among these, 12 parameter combinations had a delta AICc of less than 2. The optimal feature combination was LPH, and the RM value was 3.2, with an omission rate of 0.03448 and a delta AICc of 0. The average AUC value of the optimized model was 0.905, indicating good predictive performance.

Figure 2. Optimal model parameter combination selection.

Figure 2. Optimal model parameter combination selection.

3.2. Main Environmental Factors Affecting the Distribution of Ginkgo Biloba

MaxEnt model outputs and regularized training gain contributions, based on the Jackknife test and one-way response curves, revealed the key environmental factors influencing Ginkgo distribution. The most significant factors were precipitation in the driest month (bio14, 35.6%), minimum temperature in the coldest month (bio6, 29.4%), precipitation in the wettest season (bio16, 16.9%), and temperature seasonality (bio4, 8.9%), with a cumulative contribution of 90.8%. Precipitation in the driest month (bio14) was the main factor, while the others were secondary. The remaining environmental factors each contributed less than 5%, which was the general environmental factor. The cumulative contribution of the precipitation factors (bio14, bio16) was higher (52.5%) than that of the temperature factor (40.4%).

Table 2. The importance of each dominant environment variable in the MaxEnt model.

coding	variant	Contribution/%	Replacement contribution/%
bio14	Precipitation in the driest month	35.6	0.7
bio6	Minimum temperature in the coldest month	29.4	54.2
bio16	Wettest quarterly precipitation	16.9	19.9
bio4	Deviation from seasonal variation in temperature	8.9	12.7
DEM	altitude (e.g. above street level)	4	4.5
bio2	Annual average daily difference	1.3	0.6
bio15	Seasonal coefficient of variation of precipitation	1.1	0.3
bio5	Hottest Month Maximum Temperature	0.9	1.8
bio3	isothermal	0.9	0
Slope	elevation	0.7	3.5
bio8	Average	0.3	1.9

Table 2. The importance of each dominant environment variable in the MaxEnt model.

coding	variant	Contribution/%	Replacement contribution/%
bio14	Precipitation in the driest month	35.6	0.7
bio6	Minimum temperature in the coldest month	29.4	54.2
bio16	Wettest quarterly precipitation	16.9	19.9
bio4	Deviation from seasonal variation in temperature	8.9	12.7
DEM	altitude (e.g. above street level)	4	4.5
bio2	Annual average daily difference	1.3	0.6
bio15	Seasonal coefficient of variation of precipitation	1.1	0.3
bio5	Hottest Month Maximum Temperature	0.9	1.8
bio3	isothermal	0.9	0
Slope	elevation	0.7	3.5
bio8	Average	0.3	1.9

In summary, the response curves for the four main ecological factors influencing Ginkgo's suitable areas are shown in Figure 3. Environmental factors corresponding to an existence probability ≥0.5 are considered thresholds for suitable species distribution. Based on the curves, these thresholds were: precipitation in the driest month (20-175 mm), minimum temperature in the coldest month (-4 °C to 3 °C), precipitation in the wettest season (450-2500 mm), and temperature seasonality deviation greater than 580 mm.

3.3. Potential Distribution Areas of Ginkgo Under Current Climatic Conditions

Under current climatic conditions (Figure 5), Ginkgo’s potential habitat is primarily concentrated in the Yangtze River Basin, particularly in the lower reaches. The total suitable area accounts for 24.2% of the total land area. The highly suitable area (P ≥ 0.48) covers 508,800 km², or 21.9% of the total suitable area, and is concentrated in the junction of Hubei, Jiangxi, Henan, and Anhui provinces, as well as central Sichuan and northern Guangdong. The moderately suitable area (0.48 > P ≥ 0.27) covers 912,000 km² (39.2% of the suitable area) and is scattered across central Hunan, north-central Anhui, and most of Sichuan. Low-suitability zones (0.27 > P ≥ 0.09) occupy 892,800 km² (38.4%), primarily surrounding the medium and high-suitability zones, with concentrations in Yunnan, Guizhou, and Guangxi. Non-habitable areas are predominantly located in the northwest.

The variation in precipitation and temperature likely contributes to this distribution pattern. In northern regions, low annual precipitation and harsh winter temperatures (below -20 °C) hinder Ginkgo's long-term survival and root development.

Figure 4. Potential geographic distribution of Ginkgo under historical climatic conditions. Note: The map is produced based on the standard map with review number GS (2019)1822 downloaded from the standard map service website of the State Administration of Surveying, Mapping and Geoinformation, and there is no modification to the base map.

Figure 4. Potential geographic distribution of Ginkgo under historical climatic conditions. Note: The map is produced based on the standard map with review number GS (2019)1822 downloaded from the standard map service website of the State Administration of Surveying, Mapping and Geoinformation, and there is no modification to the base map.

Figure 5. Distribution range of potential suitable areas for Ginkgo in China under future climate scenarios.

Figure 5. Distribution range of potential suitable areas for Ginkgo in China under future climate scenarios.

3.4. Modeling of Potential Future Ginkgo Habitat Areas

Simulations of future climate scenarios under three shared socio-economic pathways yielded nine potential distribution models for three time periods. The average AUC values of all models exceeded 0.9, confirming the model’s validity. As shown in Figure 6, the area of suitable zones fluctuates under different scenarios (Table 3), with moderate-suitability zones changing little. However, highly suitable areas are predicted to increase dramatically, becoming 2-3 times larger than they are under current conditions. The distribution of highly suitable areas is expected to concentrate and spread northward toward the lower Yellow River. From 2021 to 2040, under scenarios SSP126, SSP245, and SSP585, highly suitable areas are predicted to decrease by 105,600 km². The largest total suitable area (3.312 million km²) is predicted for SSP126 from 2021 to 2041, while the smallest increase (2.9664 million km²) is expected for SSP585 from 2041 to 2060. Over the next 60 years, the total suitable area is projected to follow the trend SSP126 > SSP245 > SSP585.

3.5. Center-of-Mass Migration in Potential Habitat Areas

The centers of the aptitude zones for different periods are represented by their centers of mass. By comparing the positions of these centers in future climatic conditions with their positions in the current climate, the overall trend of the aptitude zones can be demonstrated more intuitively. While the migration distances of the Ginkgo habitat vary, the overall trend remains the same (Figure 5), showing a clear northward migration. The center of mass of the current Ginkgo habitat is located in Yichang City, Hubei Province (111°92′E, 30°57′N). In the SSP126 and SSP245 scenarios for the 2070s, the center of mass moves to Xiangfan City, Hubei Province (111°96′E, 32°22′N and 111°42′E, 31°84′N, respectively). Under the SSP585 scenario for the 2070s, it shifts to Shiyan City, Hubei Province (110°89′E, 32°54′N), which will become the most suitable area for Ginkgo in future climatic conditions. This northward migration of the center of mass indicates that Ginkgo biloba strongly responds to temperature changes. Although the migration paths in different scenarios vary, the general trend during 2021-2040 shows a significant northward shift, and by 2041-2060, the SSP245 scenario shows the farthest northward migration.

Table 4. Longitude and latitude of the center of mass of Ginkgo biloba distribution at different stages.

Period	current	2030s			2050s			2070s
		SSP126	SSP245	SSP585	SSP126	SSP245	SSP585	SSP126	SSP245	SSP585
Longitud	111.92	109.14	109.85	110.96	109.56	111.12	111.23	111.89	11.42	110.89
Latitude	30.57	31.81	31.94	31.98	31.66	32.71	31.37	32.22	31.84	32.54

Table 4. Longitude and latitude of the center of mass of Ginkgo biloba distribution at different stages.

Period	current	2030s			2050s			2070s
		SSP126	SSP245	SSP585	SSP126	SSP245	SSP585	SSP126	SSP245	SSP585
Longitud	111.92	109.14	109.85	110.96	109.56	111.12	111.23	111.89	11.42	110.89
Latitude	30.57	31.81	31.94	31.98	31.66	32.71	31.37	32.22	31.84	32.54

4. Discussion

4.1. Key Environmental Variables Affecting Ginkgo Distribution

The geographic distribution of plants on a regional scale is influenced by various environmental factors, with hydrothermal conditions being the primary determinants of plant growth [20]. Precipitation and temperature effects vary greatly between species [21]. Through the analysis of each ecological factor’s contribution rate and the results of the Jackknife test, it was found that precipitation plays a greater role than temperature in determining Ginkgo’s distribution, with precipitation-related environmental variables contributing 52.5%. The precipitation in the driest season is the most critical factor [22]. For subtropical species like Ginkgo, low temperatures typically limit northern distribution [23]. Therefore, temperature conditions must be fully considered in the conservation and planting of Ginkgo biloba. Xiaoxia et al. [24], using the MaxEnt model, also found that precipitation had a greater influence than temperature in the potential distribution of yellow jackets in China. Topography also affects Ginkgo’s distribution, with topography-related environmental variables contributing 8.6%. Lower altitudes are more favorable for Ginkgo growth [25], likely due to higher oxygen levels and other resources available at lower elevations. Generally, slopes less than 15° offer fertile, well-drained soil conducive to plant growth. Northern regions, which experience lower annual precipitation and winter temperatures often below -20 °C, are less suitable for Ginkgo’s long-term survival and root growth.

4.2. Changes in Potential Suitable Areas for Ginkgo in China Under Climate Change

Regional climate models project that temperature bands will shift northward in the future, with the most significant movements occurring along the northern boundaries of subtropical, warm, and highland temperate zones [26]. Under climate change, many species are expected to migrate to higher latitudes or altitudes [27]. For instance, species like Chinese cherry (Prunus pseudocerasus [28]) and Gymnadenia conopsea [29] are predicted to experience range contractions and northward migrations. However, a few species, such as Podocarpus neriifolius [30] and Quercus alpina, may adapt to these changes and expand their habitats. In this study, MaxEnt predicted that Ginkgo biloba’s suitable habitat will expand under future climate scenarios, with habitats shifting to higher latitudes and altitudes [31]. This expansion, particularly under the SSP585 scenario, may be driven by increased precipitation. Chen et al. [32] also demonstrated that Ginkgo’s suitable areas will move to higher latitudes under future climate conditions, aligning with this study's predictions.

4.3. Accuracy of Simulation Results

The MaxEnt model typically retains a random subset of data for modeling and uses the AUC value to evaluate predictive performance. While this method can effectively test prediction accuracy, model results may be affected by sampling bias and parameter settings [33], which can cause overfitting. First, AUC values can overestimate model performance if training and test data are influenced by sampling bias. Second, the complexity of the MaxEnt model, as a machine learning model, can result in overfitting during species distribution modeling, affecting the accuracy of results. Although many studies have successfully used the MaxEnt model for species distribution [34] predictions, some have not adequately addressed overfitting issues, limiting their models’ generalization abilities [35]. Model complexity can be managed by adjusting the RM and selecting appropriate feature combinations using AICc. In this study, the spatial sparsity tool, Pearson correlation coefficient, and kuenm package were used to select the best MaxEnt model, minimizing overfitting. The results showed that RM=3.2 and FC=LHP had the lowest overfitting and improved model confidence. After reconstructing the model using these optimized parameters [36], predictions of Ginkgo habitat for different periods were obtained, yielding better prediction accuracy than using default parameters [37].

5. Conclusion

This study modeled the spatial distribution of suitable areas for Ginkgo biloba in China under current climate conditions and three future emission scenarios (SSP126, SSP245, and SSP585), with the following conclusions:

(1) Precipitation and temperature are key environmental factors influencing Ginkgo distribution. Precipitation in the driest month, minimum temperature in the coldest month, precipitation in the hottest quarter, and seasonal temperature variation are the dominant factors. The thresholds for suitable growth conditions are: 20-175 mm precipitation, -4 °C to 3 °C minimum temperature, 450-2500 mm precipitation in the wettest quarter, and >580 mm temperature variation. Ginkgo's suitable areas are primarily located in southern China, particularly along the middle and lower reaches of the Yangtze River.

(2) Under future climate scenarios, Ginkgo’s suitable distribution areas will expand. By the 2070s, the suitable area under the green development model (SSP126) will be larger than under the medium-emission scenario (SSP245) and high-emission scenario (SSP585). As temperatures rise, Ginkgo’s suitable distribution will shift to higher latitudes.

(3) Limitations of this study: First, the selection of environmental factors—Ginkgo's distribution and suitability are influenced by many complex factors, such as land use, human interference, and species interactions, which were not fully accounted for in this study. Only 22 environmental factors were considered. Second, the precision and stability of the MaxEnt model are affected by sample size. While increasing sample size can stabilize AUC values and reduce standard deviations, it may also cause overfitting. Selecting an appropriate sample size can improve model accuracy.