Introduction
Global climate change is one of the greatest challenges (Bellard et al., Reference Bellard, Bertelsmeier, Leadley, Thuiller and Courchamp2012; Franchini & Mannucci, Reference Franchini and Mannucci2015; Malhi et al., Reference Malhi, Franklin, Seddon, Solan, Turner, Field and Knowlton2020). The annual growth rate of global greenhouse gas emissions during 2010–2019 was lower than that during 2000–2009, but overall global carbon emissions are still increasing (IPCC, Reference Lee and Romero2022). If current trends continue, we will face significant climatic, environmental and social risks (Fischer et al., Reference Fischer, Beyerle and Knutti2013). Climate change affects the survival and distribution of species (Jiang, Reference Jiang2013; Mkala et al., Reference Mkala, Mutinda, Wanga, Oulo, Oluoch and Nzei2022). Changes in global climate can cause the displacement or loss of suitable habitats for some threatened species, leading to a decrease in population numbers and the extinction of species (Cai et al., Reference Cai, Zhang, Zha, Li and Li2022; Iseli et al., Reference Iseli, Chisholm, Lenoir, Haider, Seipel and Barros2023).
Potential distribution areas of species under different scenarios can be predicted using species distribution models (Klanderud & Birks, Reference Klanderud and Birks2003; Qin et al., Reference Qin, Jin, Batsaikhan, Nyamjav, Li and Li2020; Cai et al., Reference Cai, Zhang, Zha, Li and Li2022). The most common models include the bioclimatic analysis and prediction system model (BIOCLIM; Busby, Reference Busby1991), the maximum entropy model (MaxEnt; Phillips et al., Reference Phillips, Anderson and Schapire2006), the generalized linear model (Guisan et al., Reference Guisan, Edwards and Hastie2002), the genetic algorithm for rule-set prediction model (GARP; Sanchez-Flores, Reference Sanchez–Flores2007) and the DOMAIN model (Carpenter et al., Reference Carpenter, Gillison and Winter1993). Although such models can produce species distribution maps, they are limited by the quality of data and so should be accompanied by field surveys to increase the accuracy of predictions (Abdelaal et al., Reference Abdelaal, Fois, Fenu and Bacchetta2019; Mkala et al., Reference Mkala, Mutinda, Wanga, Oulo, Oluoch and Nzei2022).
Compared to other species distribution models, MaxEnt is considered to have a relatively high predictive accuracy for species with small sample sizes, small geographical ranges and limited environmental tolerance (Phillips & Dudik, Reference Phillips and Dudik2008), and has been widely used to predict the ranges of rare and threatened species with few distribution data and narrow distributions (Hernandez et al., Reference Hernandez, Graham, Master and Albert2006; Liu et al., Reference Liu, Yang, Chen and Sun2022). Predicting the potential distributions of threatened species can help scientists and policymakers understand the effects of environmental change on species survival and can inform the planning of appropriate strategies to reduce the risk of extinction (Lawler et al., Reference Lawler, Shafer, White, Kareiva, Maurer, Blaustein and Bartlein2009; Kamilar & Beaudrot, Reference Kamilar and Beaudrot2013).
Michelia lacei W.W. Smith is an evergreen tree in the Magnoliaceae family, categorized as a Plant Species with Extremely Small Populations because of its narrow distribution, small number of individuals, the continuing impact of anthropogenic disturbance across its habitat and its high risk of extinction (Sun et al., Reference Sun, Ma and Blackmore2019a, Reference Sun, Yang and Dao2019b). The species was categorized as Critically Endangered on the China Species Red List in 2004 (Wang & Xie, Reference Wang and Xie2004), but was recategorized as Endangered in 2015 following additional field surveys and the discovery of several new populations in China (Rivers & Wheeler, Reference Rivers and Wheeler2015; Rivers et al., Reference Rivers, Beech and Murphy2016; Cai et al., Reference Cai, Dao and Sun2017). Although genomic data can contribute to the knowledge and conservation of this threatened species at the genetic level (Cai et al., Reference Cai, Liu, Yang, Zhang, Yun and Dao2024; Liu et al., Reference Liu, Cai and Sun2024), it is also crucial to understand the threats it faces and its conservation status and potential distribution under current and future climate scenarios. Here we summarize the conservation status of M. lacei in China, predict its potential distribution under current and future climate scenarios and discuss strategies for its conservation and management, especially with respect to the impacts of climate change.
Study area
Yunnan Province in south-west China covers an area of 394,100 km2, 4% of the total area of the country. Michelia lacei is mainly distributed in Yunnan Province, but is also found in northern Viet Nam and Myanmar. It is native to subtropical monsoon evergreen broad-leaved forests at altitudes of 1,000–1,800 m (Liu, Reference Liu2002; Xia et al., Reference Xia, Law and Nooteboom2008). Since 2014, we have extensively surveyed Yunnan Province every 12-months, finding a small number of M. lacei populations in south-east Yunnan (Fig. 1). In the field, Parakmeria yunnanensis Hu, Cornus hongkongensis tonkinensis (W.P. Fang) Q.Y. Xiang, Magnolia balansae (A.DC.) Dandy, Vernicia fordii (Hemsl.) Airy Shaw and Alsophila costularis Baker are the main species associated with M. lacei.
Locations of Michelia lacei individuals recorded during this study.
Methods
Investigation of conservation status and species distribution data
To gain a comprehensive understanding of the current conservation status of M. lacei in China, in September 2022 we asked local forestry bureaus and residents near its known distribution sites if they had seen any individuals of M. lacei, and, if so, how many. We subsequently surveyed all locations in which the plant had been seen or recorded in the wild. For each individual located, we measured its height and diameter at breast height. Where height was < 1.3 m or diameter at breast height was < 2 cm we recorded the plant as a seedling (Tang et al., Reference Tang, He, Gao, Zhao, Sun and Ohsawa2011; Han et al., Reference Han, Tao and Sun2019; Tao et al., Reference Tao, Han, Song and Sun2020). We documented geographical coordinates (using a GPS), fruiting status (yes/no), growth status (i.e. whether or not we observed pests and diseases, as these can destroy plant tissue, impacting development), companion species and whether the plant was in a protected area. We calculated and recorded the extent of occurrence and area of occupancy of M. lacei in China using the package ConR (Dauby et al., Reference Dauby, Stévart, Droissart, Cosiaux, Deblauwe, Simo-Droissart, ( and )2017) in R 4.0.3 (R Core Team, Reference .2020).
In addition to our field surveys, in October 2022 we examined literature and herbarium records (FRPS, 2019; CNKI, 2023; CVH, 2023; POWO, 2023). In total, we obtained 60 records for M. lacei in China (53 from our surveys and seven earlier records). We mapped these data using ArcGIS 10.8 (Esri, USA), retaining only one record per km2 to eliminate spatial bias (Rather et al., Reference Rather, Ahmad, Dar, Dar and Khuroo2021; Mkala et al., Reference Mkala, Mutinda, Wanga, Oulo, Oluoch and Nzei2022), resulting in 18 locations of M. lacei for further analysis. We also investigated the ex situ conservation of M. lacei in Chinese botanical gardens by collating information on the institutions where it is cultivated, and data such as the growth, flowering and fruiting status of cultivated plants.
Bioclimatic variables
To predict the potential present and future distribution of M. lacei, we used the 19 bioclimatic variables, with a resolution of 2.5 arcminutes, provided by WorldClim 2.1 (Fick & Hijmans, Reference Fick and Hijmans2017). We used mean climate data for 1970–2000 to predict current distribution and for 2021–2040, 2041–2060, 2061–2080 and 2081–2100 to predict future distribution. Shared Socioeconomic Pathways can be used to generate different climate change scenarios to assess the impacts of policies and actions on future climate change (Weng et al., Reference Weng, Cai and Wang2020). We modelled each time period using four Pathways (SSP126, SSP245, SSP370, SSP585) under the second-generation Earth system model of the Centre National de Recherches Météorologiques (CNRM-ESM 2-1; Fick & Hijmans, Reference Fick and Hijmans2017). We chose this model, which was developed for the sixth phase of the Coupled Model Intercomparison Project, because in a number of experiments it has shown a more significant response to the external environment than other models as it includes interactive Earth system components such as the carbon cycle, aerosols and atmospheric chemistry (Seferian et al., Reference Seferian, Nabat, Michou, Saint-Martin, Voldoire and Colin2019). The four Shared Socioeconomic Pathways represent the net radiative forcings at the end of 2100 (2.6, 4.5, 7.0 and 8.5 W/m2) and simulate global warming trends in the absence of climate policy intervention. As the forcing value increases, so does the radiative capacity, representing higher levels of greenhouse gas emissions that therefore have a stronger impact on environmental changes (Fick & Hijmans, Reference Fick and Hijmans2017; O'Neill et al., Reference O'Neill, Kriegler, Ebi, Kemp-Benedict, Riahi and Rothman2017; Riahi et al., Reference Riahi, Van Vuuren, Kriegler, Edmonds, O'Neill and Fujimori2017).
To reduce analytical interference from strong correlations between the bioclimatic variables, we extracted the 19 bioclimatic variables for each of the 18 locations of M. lacei. We ran correlation analyses of these data in R, and we analysed the contributions of the 19 bioclimatic variables to the species distribution using MaxEnt 3.4.3 (Phillips et al., Reference Phillips, Anderson and Schapire2006). When screening bioclimatic variables we first retained that with the largest contribution, and then removed variables that had a correlation of r > 0.8 with the variable retained, and then retained the remaining variables. We repeated this step until no more variables could be removed. Finally, we used the remaining variables to predict the potential distribution area of M. lacei (Dormann et al., Reference Dormann, Elith, Bacher, Buchmann, Carl and Carre2013; Yi et al., Reference Yi, Cheng, Yang and Zhang2016; Wang et al., Reference Wang, Li, He and Liu2018).
Model description
We divided the species location information into two subsets and used 75% of the location information as training data to build the distribution model. We used the remaining 25% to validate the model, with 10 repetitions and the replicate run type set to bootstrap (Efron, Reference Efron1979; Khanal et al., Reference Khanal, Timilsina, Behroozian, Peterson, Poudel and Alwar2022; Soilhi et al., Reference Soilhi, Sayari, Benalouache and Mekki2022). We set the output format to cloglog, which estimates the probability of presence between 0 and 1, as recent studies have suggested that cloglog has greater theoretical support than one of the alternative output formats, logistic (Phillips et al., Reference Phillips, Anderson, Dudik, Schapire and Blair2017). We left the remaining parameters as the MaxEnt 3.4.3 defaults (Phillips & Dudik, Reference Phillips and Dudik2008). To demonstrate the reliability of the generated model, we calculated the area under the curve (AUC) of the receiver operating characteristic curve (ROC). The AUC value is related to the accuracy of the model, and the closer the AUC value is to 1, the better the test data fit the model (Swets, Reference Swets1988; Coban et al., Reference Coban, Orucu and Arslan2020).
Suitable habitat classification and distribution changes
We used SDM_Toolbox 2.5 (Brown et al., Reference Brown, Bennett and French2017) in ArcGIS to convert the resulting MaxEnt ASCII format files to raster files, and created a map of the potential distribution of M. lacei. Cloglog values range from 0 to 1, with 1 being areas most suitable for the species. We then used the Reclassify (Spatial Analyst) tool in ArcGIS (Brown et al., Reference Brown, Bennett and French2017) to categorize habitat into four grades based on previous studies of native Chinese species: unsuitable areas (0–0.05), areas of low suitability (0.06–0.33), moderately suitable areas (0.34–0.66) and highly suitable areas (0.67–1.00; He & Zhou, Reference He and Zhou2011; Gong et al., Reference Gong, Li, Wu and Jiang2022). Finally, we calculated the area and proportion of each habitat suitability class using ArcGIS.
To analyse changes in potential distribution over time, we examined only those areas with suitable habitat (i.e. with a cloglog value > 0.05). We used SDM_Toolbox to compare potential distribution in each future period with the current potential area of distribution (Brown et al., Reference Brown, Bennett and French2017), identifying areas that will become more suitable, less suitable or will be unchanged. We viewed this output in ArcGIS and graded habitat as: −1 (the future habitat is more suitable than it is at present), 0 (the habitat is suitable neither in the future nor at present), 1 (the habitat is suitable during both periods) or 2 (the habitat is suitable at present but will not be suitable in the future; Mkala et al., Reference Mkala, Mutinda, Wanga, Oulo, Oluoch and Nzei2022).
Results
Population status survey
During our surveys we recorded 10 populations and 53 individuals of M. lacei, comprising 50 mature individuals and three seedlings in five counties. Seven individuals were found in Daweishan National Nature Reserve, the remainder were outside protected areas, in villages, along roadsides, on farmland and elsewhere. The extent of occurrence and area of occupancy of M. lacei in China are 3,881 km2 and 52 km2, respectively.
The height of 90% of the M. lacei individuals was 6–30 m (Fig. 2a); 85% had a diameter at breast height of 11–90 cm (Fig. 2b). One tree had a diameter at breast height of 192 cm, which is the largest M. lacei individual on record (Plate 1a).
The (a) height and (b) diameter at breast height of all 53 individuals of M. lacei found during our surveys in China.
(a) The largest Michelia lacei individual on record, found during our surveys in Yunnan Province, China, in 2022, and (b) flower bud, (c) flower and (d) fruit.
Five institutions run ex situ conservation programmes for M. lacei in China: Kunming Botanical Garden, South China Botanical Garden, Wuhan Botanic Garden, Guilin Botanical Garden and Fairy Lake Botanical Garden (Xian Hu; Sun et al., Reference Sun, Ma and Blackmore2019a, Reference Sun, Yang and Dao2019b; Linsky & Sun, Reference Linsky, Sun, Linsky, Crowley, Beckman Bruns and Coffey2022). In 1987, Kunming Botanical Garden planted 12 M. lacei individuals, all of which have grown into trees > 10 m tall and are in good condition. Four have flowered, and two have also fruited.
Model performance and the importance of environmental variables
The AUC values of the ROC curve were > 0.9 in all predicted future periods, indicating that the simulation of the area potentially suitable for M. lacei is credible (Phillips et al., Reference Phillips, Anderson and Schapire2006). Following examination of the correlation coefficients and the contributions of the 19 bioclimatic variables, we selected four to predict the potential distribution of the species (Fig. 3): temperature seasonality (Bio_4), minimum temperature of coldest month (Bio_6), temperature annual range (Bio_7) and precipitation of wettest month (Bio_13). In the MaxEnt simulation the two highest-contributing factors were Bio_4 (57.1%) and Bio_7 (19.1%), which together accounted for 76.2% of the total contribution (Table 1).
Pearson's correlation analysis of the 19 bioclimatic variables from WorldClim 2.1 (Fick & Hijmans, Reference Fick and Hijmans2017). Black outlined boxes indicate the correlation coefficients between the four bioclimatic variables used for modelling (Bio_4, Bio_6, Bio_7 and Bio_13; Table 1), after screening. Correlation coefficients range between −1 and 1. (Readers of the printed journal are referred to the online article for a colour version of this figure.)
Bioclimatic variables downloaded from WorldClim 2.1 (Fick & Hijmans, Reference Fick and Hijmans2017) used for modelling and their per cent contributions to the assessment of the distribution of Michelia lacei (Fig. 3).
Potential distribution under current and future climate conditions
In addition to the known current distribution in Yunnan, our model suggests areas of suitable habitat in Xizang, Sichuan, Guizhou, Guangxi, Guangdong, Fujian, Taiwan and Hainan (Fig. 4a). However, during our literature searches, we found no record of M. lacei in China beyond Yunnan Province. Within Yunnan, predicted suitable habitat under current conditions is mainly in the west and south-east (Fig. 4b). In western Yunnan, the moderately and highly suitable areas are predominantly in Dehong, and in south-eastern Yunnan these areas are predominantly in Wenshan and Honghe. Of the potentially suitable habitat in Yunnan, 59% (19.38 × 105 km2) had only low suitability, 33% (10.87 × 105 km2) was moderately suitable and 8% (2.70 × 105 km2) was highly suitable. Of the total habitat in Yunnan Province, 16% is unsuitable for M. lacei (6.46 × 105 km2; Table 2).
The current potential distribution of suitable habitat for M. lacei (a) in China and (b) in Yunnan Province.
Predicted areas of M. lacei habitat in Yunnan Province, China, for future time periods under four Shared Socioeconomic Pathway global warming scenarios (SSP126, SSP245, SSP370 and SSP585), compared with the current potential distribution of the species.
For each future period we predicted habitat suitability in Yunnan for the four Shared Socioeconomic Pathways (Fig. 5; Table 2). The largest areas of unsuitable and low suitability habitat were both predicted under the worst-case scenario (SSP585), with unsuitable habitat of 8.66 × 105 km2 predicted during 2061–2080, and low suitability habitat of 28.06 × 105 km2 predicted during 2081–2100. We predicted areas of moderately and highly suitable habitat to be greatest under the SSP126 scenario during 2041-2060, at 18.05 × 105 and 2.81 × 105 km2, respectively.
Predicted future areas of M. lacei habitat in Yunnan Province, China, with different levels of suitability under different Shared Socioeconomic Pathways (SSP) climate scenarios.
Figure 6 shows the changes in the future potential distribution of M. lacei compared to that under current conditions, showing areas of expansion, areas with no change in suitability (the areas are suitable neither in the future nor at present), areas of stability (the areas are suitable during both periods) and areas of contraction. The highest increase in area of suitable habitat is 1.58 × 105 km2 under SSP585 during 2081–2100. Under the SSP585 scenario during 2061–2080 the area suitable for M. lacei decreased by 2.62 × 105 km2 compared to the present. Based on current and future climate data, we predict that as climate conditions change in the future an increasingly large area will be transformed into habitat with low suitability for M. lacei.
Changes in future habitat suitability for Michelia lacei relative to current habitat suitability in Yunnan Province, China, under different Shared Socioeconomic Pathways. Expansion indicates that the future habitat is more suitable than it is at present, no change (unsuitable) indicates that the habitat is suitable neither in the future nor at present, stability (suitable) indicates that the habitat is suitable both now and in the future, and contraction indicates that the habitat is suitable at present but will not be suitable in the future. (Readers of the printed journal are referred to the online article for a colour version of this figure.)
In summary, our modelling predicts that areas of moderately and highly suitable habitat for M. lacei will gradually shrink in west and south-east Yunnan. Areas of suitable habitat will also decline in central Yunnan.
Discussion
Modelling performance and the influence of the variables
Reliable distribution data are crucial for accurately predicting species’ potential distribution (Phillips & Dudik, Reference Phillips and Dudik2008; Chen et al., Reference Chen, Lei, Zhang and Jia2012). Modelling location data for 60 individuals of M. lacei, the AUC values calculated with MaxEnt were all > 0.9, indicating that the simulation prediction is accurate (Swets, Reference Swets1988). Previous studies have found precipitation and temperature to be important factors influencing the distribution of species (Shi et al., Reference Shi, Preisler, Quinn, Zhao, Huang and Roll2020; Rather et al., Reference Rather, Ahmad, Dar and Khuroo2022). In this study, we found that temperature had a greater influence on predicting the potential range of M. lacei than precipitation. In addition to the significant influence of climatic and topographic variables on the distribution of species, soil and hydrogeological variables also play important roles, and we plan to investigate the roles of these variables on the distribution of M. lacei in future studies. The current distribution of the species is relatively small compared with the potential area of suitable habitat, and these findings should be taken into account in future field surveys and potential reintroductions of this species.
Predicted habitat suitability for M. lacei under current climate conditions
Current predicted suitable habitat for M. lacei in China is primarily in the southern provinces. We only surveyed in Yunnan, but we did not find any literature records of the species in China beyond this province. Suitable area predicted using species distribution models is often wider than the actual distribution range of a species (Cai et al., Reference Cai, Zhang, Zha, Li and Li2022; Yan & Zhang, Reference Yan and Zhang2022). When determining the distribution of threatened plant species, additional factors such as pollinator numbers, self-breeding, seed dispersal, community competition, habitat destruction, reduced population and habitat fragmentation need to be taken into account. These, amongst other factors, all have a greater impact on the distribution of highly threatened species than on those less threatened (Chen, Reference Chen2017).
Changes in habitat suitability under future climate change scenarios
Based on our modelling, we predict the distribution of M. lacei will gradually decrease in the future, with the highly suitable habitat shrinking in size as the climate warms. This is consistent with previous research on the distribution patterns of threatened species in south-eastern Yunnan (Cai, Reference Cai2020). In the four future periods we modelled, highly and moderately suitable areas were predicted to become areas of low suitability as temperatures change. In particular, the moderately suitable areas in Xishuangbanna were predicted to decline to almost zero. Under the worst-case scenario (pathway SSP585 during 2061–2080), the moderately and highly suitable habitats were predicted to be restricted to Dehong in western Yunnan, with a few small patches in Honghe and Wenshan in south-eastern Yunnan. In general, rising temperatures will promote the migration of plant populations into boreal forests (Boisvert-Marsh & de Blois, Reference Boisvert-Marsh and De Blois2021; Behera et al., Reference Behera, Kellner, Kendrick and Sax2023). Such increases in temperature are likely to be accompanied by the movement of mountain plants to higher elevations. As a Plant Species with Extremely Small Populations, the migratory ability of M. lacei is limited compared to other, more widespread species, and its suitable habitat is gradually being reduced.
Protection and management of M. lacei
Some of the M. lacei individuals we recorded are growing in fengshui forests, preserved by local communities from nearby villages who believe the forests can bring prosperity and good fortune (Coggins & Minor, Reference Coggins and Minor2018). Two large individuals growing near a village have been listed as ancient trees by the local forestry bureau, and some individuals are located in a nature reserve, providing some level of protection. However, we found most individuals growing near roadsides, fields or riverbanks, with no protective measures. These individuals are at high risk of being cut down or otherwise destroyed. We found only three seedlings during our field studies, and can infer from this that wild populations are struggling to reproduce naturally. We found these seedlings in a vegetable field, by a roadside and in farmland where black cardomom Amomum tsaoko was being cultivated. During our 2022 surveys of wild M. lacei populations we found only four adult plants that had fruited.
Conclusion
We predicted suitable habitat for M. lacei under current and future climate conditions and found that areas suitable for this species will decline. Furthermore, M. lacei populations face significant threats from anthropogenic interference, among other factors. In addition to existing local protection, further appropriate conservation measures are needed for M. lacei. In situ conservation should focus on areas where the habitat of this species is predicted to remain stable in the future, by establishing conservation sites. We have achieved germ-free germination and seed propagation of M. lacei by collecting fruits from wild habitats and botanic gardens, but these methods are currently still being developed. Collection of fruits from the various M. lacei populations should continue and their seedlings should be grown for ex situ or near situ (i.e. establishment of populations at sites near wild populations; Sun et al., Reference Sun, Liu and Zhang2021; Reference Sun, Ma and Corlett2024) conservation.
Author contributions
Field surveys: all authors; data analysis: YL, LC; writing, revision: all authors.
Acknowledgements
We thank Zhiling Dao, Liewen Lin, Bo Xiao, Zhiyong Yu and Zongli Liang for their help in the field; Yuhang Liu for her guidance on the use of MaxEnt; and Detuan Liu for providing valuable comments. This research was supported by the Science and Technology Basic Resources Investigation Program of China (Grant No. 2017FY100100), the NSFC (National Natural Science Foundation of China; Grant No. 32101407), the NSFC–Yunnan Joint Fund (Grant No. U1302262) and Yunnan Provincial Science and Technology Talent and Platform Plan (202305AM070005).
Conflicts of interest
None.
Ethical standards
This research abided by the Oryx guidelines on ethical standards.
Data availability
All species distribution data and bioclimatic variables have been described in the Supplementary Material.
Open access
