Modelling environmentally suitable areas for the potential introduction and cultivation of the emerging oil crop Paeonia ostii in China

Paeonia ostii is a traditional ornamental and medicinal species that has attracted considerable interest for its high oil value. To facilitate the effective and rational cultivation and application of P. ostii in China, it is necessary to determine its potential spatial habitat distribution and environmental requirements. Using high-resolution environmental data for current and future climate scenarios, the potential suitable area and climatic requirements of P. ostii were modelled. Among the 11 environmental variables investigated, growing degree days, precipitation of the wettest month, mean temperature of the coldest quarter, global UV-B radiation, annual precipitation, and soil pH played major roles in determining the suitability of a habitat for the cultivation of P. ostii. Under the current environmental conditions in China, a total area of 20.31 × 105 km2 is suitable for growing P. ostii, accounting for 21.16% of the country’s total land area. Under the two future climate scenario/year combinations (i.e., representative concentration pathways [RCPs], RCP2.6 and RCP8.5 in 2050), this species would increase its suitable area at high latitudes while decrease at low latitudes. These results present valuable information and a theoretical reference point for identifying the suitable cultivation areas of P. ostii.

www.nature.com/scientificreports www.nature.com/scientificreports/ disease, and exhibits wide adaptability 3,19 , it is still unclear whether all these new cultivation regions are suitable for the growth of this species. The selection of suitable cultivation areas has important implications for sustainable use, so as to avoid planting blindly and ensure good growth. Therefore, mapping the climatically suitable areas for P. ostii and identifying key environmental factors are urgently needed for the introduction and cultivation of this species in China.
Species distribution modeling (SDM) is widely used in ecology, biogeography, and evolutionary studies 20 . Whilst there are other methods available for modelling species distributions, (such as Grap, Bioclim, and Domain), MaxEnt is one of the most widely applied approaches in the field of crop science and has been used by agricultural industries for modelling crop plant distributions under both current and future environmental conditions 21,22 . MaxEnt uses presence-only data and the associated environmental variables to project the potential habitat distribution of a target species based on the principles of the maximum entropy algorithm 22,23 . This process has been widely applied to estimate the ecological requirements of medicinal plants (e.g., H. riparia, Lour) 24 , major forest tree species (e.g., Platycladus orientalis and Davidia involucrata) 25,26 , invasive plants (e.g. Mimosa diplotricha and Mikania micrantha) 27 and endangered relic species (Schisandra sphenanthera and Ammopiptanthus mongolicus) 28,29 , and then to characterize and map their suitable growing areas at a landscape scale.
To evaluate the environmental variables and properties of habitat distribution that shape habitat suitability, MaxEnt was used to model and predict distributions of P. ostii based on an extensive collection of geo-referenced occurrence records and associated current and future environmental data. This study aims to: (1) identify the main environmental variables that constrain the distribution of P. ostii, (2) map P. ostii's climatically suitable areas and estimate its climatic thresholds, (3) determine different areas degrees of suitability for cultivating P. ostii nationally, and, (4) predict the spatial patterns of changes in the suitable cultivation areas of P. ostii under two future climate change scenario/year combinations: RCP2.6-2050 (average for the years 2041-2060 under the RCP2.6 scenario) and RCP8.5-2050 (average for the years 2041-2060 under the RCP8.5 scenario). This work provides valuable information on the distribution of P. ostii and reference data to assist land managers in the development of effective plans for large-scale increases in cultivation across China.
Methods occurrence and distribution of P. ostii. The occurrence records of P. ostii were obtained from three resources: (1) published literature [30][31][32] , (2) the Chinese Virtual Herbarium (CVH, http://v5.cvh.org.cn/ 33 , a free and open access database that holds plant distribution records from all main herbaria across the country) and (3) our systematic fieldwork carried out in China in 2014. When occurrence records lacked exact geo-coordinates, Google Earth was used (http://ditu.google.cn/) to determine the latitude and longitude. In total, we collected a total of 87 geo-referenced occurrence records of the species within its known native distribution (Table S1 and Fig. 1).

Current environmental data.
In the current study, 29 environmental variables (Table S2) that may influence the distribution of P. ostii were initially selected to model the current species distribution patterns. These included 19 climatic variables (Bio1-Bio19) at spatial resolutions of 1 km obtained from the WorldClim website, which uses observed monthly temperature and precipitation data sourced from globally distributed weather stations, averaged over the period 1950-2000 (www.worldclim.org) 34 . The other ten environmental variables were obtained as follows: soil organic carbon (SC), soil pH (SpH), and growing degree days (GDD) from the Center for Sustainability and the Global Environment (http://www.sage.wisc.edu/atlas/index.php) 35 , ground-frost frequency (FRS), wet-day frequency (WET) and vapor pressure (VAP) from the IPCC database (http://www.ipccdata.org/ obs/cru_ts2_1.html), and global UV-B radiation (UVB1, UVB2, UVB3, UVB4) from the gIUV database (http:// www.ufz.de/gluv/) 36 . These data were preprocessed to a general spatial resolution of 30 s latitude/longitude (ca. 1 km 2 at ground level).
Strong collinearity between the environmental variables could artifically inflate the accuracy of the model 37 . To avoid the potential problem of multi-collinearity in the meteorological and environmental factors, all the variables were subjected to a principal component analysis (PCA) to reduce the dimensionality of the models and avoid multi-collinearity (see the summary of variable correlations in Table S3). After evaluation, we retained 11 environmental variables (Bio3, Bio6, Bio11, Bio12, Bio13, Alt_China, GDD, UVB1, SpH, SC, WET) with pairwise Pearson correlation coefficients of r ≤ 0.70, for subsequent analyses.
Future climate scenarios and data. To predict potential changes of the geographic distribution of P. ostii, we used the BCC-CSM1.1 GCM, one of the most-used models currently available for simulating the global climate's response to increasing greenhouse gas concentrations 38 , and simulated two of the four RCPs (i.e., RCP2.6, RCP4.5, RCP6.0 and RCP8.5) that were used in the 5 th IPCC Assessment Report 39,40 . RCP 2.6 represents the most "benign" scenario (i.e., a likely average global temperature increase of 0.3-1.7 °C for ca. 2081-2100), whereas RCP 8.5 is the most extreme scenario (a likely global temperature increase of 2.6-4.8 °C for ca. 2081-2100). For modeling purposes, climate variables were selected for simulation in two climate change scenario/year combinations: RCP2.6-2050 (average for the years 2041-2060 under the RCP2.6 scenario) and RCP8.5-2050 (average for the years 2061-2080 under the RCP8.5 scenario). Climate variables for future climate scenarios were also downloaded from the WorldClim website. The ten environmental parameters (SC, SpH, GDD, FRS, WET, VAP and four UV-B radiation parameters) remained unchanged for the following analyses of SDM projections under future climate conditions. evaluation of current and future potential habitat. MaxEnt v3.3.3 22,23 was used as the algorithm to model the current and future habitat distribution of P. ostii in China. Model prediction parameters for the baseline model (representing current environmental conditions) were: convergence threshold = 10 −5 , maximum iterations = 1000, regularization value β = 10 −4 , and use of linear, quadratic, product and binary features 23 .75% of the location point data were used for training the model, while the remaining 25% were used to validate the MaxEnt model. The algorithm was set to run either 1000 iterations of these processes or to continue until convergence (threshold 0.00001). The logistic output of each MaxEnt run was then interpreted as a probability map. In addition to visual inspection of the output maps, model performance was evaluated using the area under the receiver operating characteristic curve (AUC), where an AUC = 0.5 indicated that the model predicted outcomes no better than random chance, an AUC ≥ 0.7 indicated strong predictive power, and an AUC = 1.0 indicated a perfectly fitted model 41 . The Jackknife test was used to assess the relative importance of the variables. The distribution of P. ostii in terms of each environmental variable was displayed in kernel density plots.
Area suitability mapping and output divisions. The outputs were transformed into raster format using the ArcMap tool in ArcGIS v10.0 software 42 for further analysis. To define habitat versus non-habitat for P. ostii, we used the "maximum training sensitivity plus specificity" threshold that produced highly accurate predictions 43 , and classified the appropriate level of division by using the "reclassify" function of the spatial analysis tools in ArcGIS. According to the weighted average method, spatial analysis in ArcGIS was applied to calculate the rasters, producing maps that can be used to assess habitat suitability at resolutions of a 1 km × 1 km grid cell 28 . The calculated raster data were then exported into a spreadsheet.

Results
Model performance and important environmental variables. Model performance for P. ostii was classified as being better than random, with a training AUC of 0.983 and test AUC of 0.979 (Fig. S1). The test AUC was close to 1, which indicated that the model performed better than random and therefore showed the high accuracy of the model.  Table 2. Portions of different classes of potential distribution area of P. ostii under current climatic and geological conditions. Notes: PR = province (region); NC = number of cities/counties; CA = coverage area; AA = accounted area for area of cities/countries. www.nature.com/scientificreports www.nature.com/scientificreports/ The estimates of relative contributions of the environmental variables to the MaxEnt model and the ranges of the 11 environmental variables across the four habitat classes is shown in Table 1. Among the 11 environmental variables, growing degree days (GDD, 26.4%), precipitation of the wettest month (Bio13, 17.7%), mean temperature of the coldest quarter (Bio11, 16.9), annual mean UV-B radiation (UV-B, 12.2%), annual precipitation (Bio12, 9.7%), and soil pH (SpH, 6.1%) played major roles in determining the suitability of a habitat for the cultivation of P. ostii, with cumulative contributions as high as 89% ( Table 1). The response curves of these six predictor variables showed unimodal relationships, with the response peak representing the optimum environmental conditions for P. ostii: GDD at 3700 °C, Bio13 at 200 mm, Bio11 at 2 °C, UVB1 at 2900 J.m −2 .day, Bio12 at 800 mm and SpH at 6.5-6.9 (Fig. 1). The frequency distribution of P. ostii for each environmental variable is presented in the kernel density plots in Fig. 2. the current and potential distribution of P. ostii. Based on known occurrences of P. ostii and current climate data, a habitat map of suitable climates was produced using the MaxEnt model with 11 environmental variables. The suitable habitat areas were divided into four classes with the "maximum training sensitivity plus specificity" threshold value of 0.166, in which they were defined as: (i) highly suitable areas (0.6-1.0), (ii)  (Tables 2  and S4). Among the provinces, Henan, Shandong, and Hebei Provinces had the most highly, moderate, and marginally suitable habitat areas for P. ostii, respectively. The current total suitable area in China for this species was predicted to encompass ca. 20.31 × 10 5 km 2 .

Discussion
This study performed a detailed analysis on the suitable habitat of P. ostii under current and future climate scenarios, which can serve as an important step in developing strategies and policies for the effective management and utilization of this emerging oil crop.
P. ostii has a wide distribution range in China, ranging across Bozhou and Tongling city in Anhui Province, the southwest region of Hunan Province, and sparsely across in Shanxi, Shandong Henan, Hubei, Shaanxi provinces, and Sichuan 3 . Based on large-scale environmental data, our modeling study projected a spatial distribution in China of ca. 20.31 × 10 5 km 2 for the species under current climate conditions, in which the center areas are in Henan, Anhui, the southern sector of Shaanxi and Shanxi, the northwest of Hubei, and the northeast of Sichuan Province. This model was in agreement with the precious occurrence records 30-32 that found this species' range was from 26 to 42°N and 100 to 124°E.   www.nature.com/scientificreports www.nature.com/scientificreports/ Our simulation results showed that temperature, precipitation, and light were among the most important factors affecting the distribution of P. ostii. Previous studies on the growth characteristics of other plant species showed that GDD as a representation of accumulated temperature were closely related to plant growth and development, including emergence, flowering, and maturity dates [44][45][46][47][48][49] . Likewise, we found that GDD were the most important factor (26.4%) for P. ostii and that its base value was 2000-5000 °C (≥10 °C) for normal growth. Temperature was reported to affect the seed dormancy and germination of P. ostii 18 . Dormancy in peony seeds typically is complex, requiring cold stratification (0-10 °C) to break the dormancy of the epicotyl 50 , while temperatures >20 °C were found to be unfavorable for seed germination. This suggests that the emergence and the death of P. ostii seedlings be directly affected by temperature, which played a main role in shaping the ecological adaptation of this species.
Water availability is correlated with many environmental factors that influence the physiological and biochemical processes of plants 51 . The number of female flowers and the root growth of P. ostii are significantly associated with precipitation amount. Abnormally-grown or even rotted roots can occure in Areas with high precipitation; while those areas with less precipitation, such as Xinjiang Province (the annual precipitation was no more than 100 mm, typically below 50 mm) 52 , will be not beneficial to the growth and development. In our modeling exercise, precipitation including Bio12 (9.7%) and Bio13 (17.7%) greatly impacted the habitat suitability of P. ostii, with base values of 40-350 mm and 500-2000 mm, respectively. This suggests that not only severe drought but heavy and concentrated precipitation, especially during flowering, would greatly impede the growth www.nature.com/scientificreports www.nature.com/scientificreports/ and development of P. ostii. These environmental factors obviously impacting the distribution of P. ostii would be helpful in formulating a suitable region-wide recommendation for the introduction and cultivation of it.
Notably, the intensity of UV-B radiation was identified as one of the most biologically relevant ultraviolet radiation, and it significantly affected the growth and distribution of P. ostii. In agricultural systems, UV-B radiation affects the yield of many crops [53][54][55][56][57] and increased UV-B radiation can alter the photosynthetic capacity and reduce biomass production in plants 58,59 . However, our knowledge of the effects of UV-B radiation on P. ostii is very limited, and more studies on this issue are necessary. The SpH (6.1), Alt_china (2.7), SC (1.4) and WET (1.4) variables played relatively smaller roles in the growth of P. ostii. Of these factors, only SpH seemed to significantly affect the habitat suitability of P. ostii. The overall pH range of suitable areas was from 5.5 (slightly acid) to 8.5 (alkaline); however, the pH range considered highly ecologically suitable was comparatively narrow (6.5-6.9), which demonstrated that both extremely acid and saline-alkali soils could considerably diminish the suitability of P. ostii in some places.
Climate change is driving latitudinal and altitudinal shifts in species distribution worldwide 60 , although most species are generally migrating to higher latitudes and elevations as climates warm 26,61 . Consistently, we found that under both the low (RCP2.6) and the higher (RCP8.5) scenarios of concentrations of greenhouse gas emissions, the suitable habitat range of P. ostii will increase as global warming intensity proceeds. Low-latitude regions, such as Hunan, Sichuan, Jiangxi, and Zhejiang Provinces would see a decrease in distribution. Therefore, when new areas were chosen to cultivate this species, areas should be reserved for marginal populations in low-latitudes, so as to increase survival of P. ostii plants to cope with climate change in the future.
Assessment of habitat suitability is important in understanding what defines suitable ecological conditions for cultivating P. ostii, and can provide valuable information and a theoretical reference for the identification of future suitable areas. In this study, our results show that the suitable environmental areas of P. ostii accurately predicted at the county scale can be applied to the evaluation of land use patterns at a large scale. For management efficiency, the distribution of highly and moderate suitable areas for cultivating P. ostii should be focused on. In addition, China's national policy could be utilized to encourage farmers to cultivate P. ostia for oilseed production in more regions. In agricultural production, such environment variables as temperature, precipitation, and soil features can be artificially regulated (e.g., irrigation, and soil improvement using greenhouses and shade) 62,63 . According to the key variables identified in this study, Xinjiang Province was unsuitable for P. ostii cultivation due to water-deficiencies. Howere, when irrigation facilities are used, some of the unsuitable areas in this region would become available for growing P. ostii, which would increase the land's value for potential oilseed production.
Due to a long-term medicinal cultivation history, the modern oil use breeding of P. ostii is rather short. The modelling of suitable environmental cultivation areas in this study was based on the occurrence of this species where it was traditionally cultivated for medicinal use. Although the areas are suitable for the growth of P. ostii, it is still uncertain if these regions will produce high seed yield and quality. Therefore, additional research should focus on determining the ecological quality and crop regionalization of P. ostii.