Introduction

Among the plethora of impacts of climate change on flora and fauna, change in the distributional range of the species has been well reported (Gomes et al. 2018; Lamsal et al. 2018; Adhikari et al. 2019; Ahmad et al. 2019; Shrestha and Shrestha 2019). While some species may show poleward shift in their distribution, in others upward altitudinal shift in response to climate change has been reported (Pauchard et al. 2016). Of particular interest in this context are the alien (non-native/exotic) species whose fortunes may also change with the change in the climate (Bradley et al. 2010). Predicting future distribution of species over relevant spatio-temporal scales under various climate change scenarios is of particular significance in the management of invasive alien species (Porfirio et al. 2014; Fletcher et al. 2016; Chai et al. 2016). Consequently, in the last two decades several modelling tools (Elith et al. 2010; Fourcade et al. 2014; Thuiller et al. 2016; Di Cola et al. 2017) have been developed for predicting future distribution of species on the basis of current correlation between climatic variables and species occurrence records, assuming niche conservatism, no dispersal limitation or biotic interactions (Ramirez-Albores et al. 2016). Such models have become popular in predicting the distribution of invasive species (Bradley et al. 2010; Tingley et al. 2014; Kramer et al. 2017; Barbet-Massin et al. 2018), especially under changing climate conditions (Bellard et al. 2018; Ahmad et al. 2019). Changes in species distributions are likely to affect structural organization and functional integrity of communities and ecosystems with unforeseeable ecological and economic impacts and costs (Hughes 2000).

Understanding of niche dynamics in the native and introduced regions using the COUE framework of Centroid shift, Overlap, Unfilling, and Expansion (Broennimann et al. 2012; Petitpierre et al. 2012) has received lot of attention in the recent past. Results of such studies have been variable (Thuiller et al. 2005) with some species showing niche conservatism (Liu et al. 2020) and others exhibiting niche shift (Atwater et al. 2018). Based on 86 studies dealing with 434 invasive species, Liu et al. (2020) concluded that most invasive species largely conserve their climatic niche and niche conservatism not only allows for transfer of niche models to new ranges but also provides for use of ecological niche modelling in understanding the response of species to climate change.

Climate change that is driven by anthropogenic greenhouse gas emissions on account of increase in human population, economic activity, changes in life style, energy use, land use changes and other related activities is widespread and well reported (IPCC 2014). Representative concentration pathways (RCPs) of greenhouse gases including a stringent mitigation scenario (RCP 2.6), two intermediate scenarios (RCP 4.5 and RCP 6.0) and one scenario with very high GHG emissions (RCP 8.5) have been used for making future projections. Relative to 1986–2005, the increase in global mean surface temperature by the end of the twenty-first century (2081–2100) is likely to be 0.3–1.7 °C under RCP 2.6, 1.1–2.6 °C under RCP 4.5, 1.4–3.1 °C under RCP 6.0 and 2.6–4.8 °C under RCP 8.5 (IPCC 2014).

Over the last few decades, indicators of climate change in the form of increasing temperature, shrinking of glaciers and erratic precipitation pattern are easily discernible in the Himalaya (Shrestha et al. 2012; Bahuguna et al. 2014) including Kashmir Himalaya (Rashid et al. 2015; Zaz et al. 2018). Since Himalaya is a biodiversity hotspot, the impact of such changes may be more damaging on the native biodiversity (Rashid et al. 2013; Malik et al. 2015). It is very likely that new species alien to the region may gain foothold due to changing climatic conditions and P. hysterophorus is one species that is making inroads into previously uninvaded areas in Kashmir Himalaya.

Parthenium hysterophorus is an annual short-lived herbaceous species that belongs to family Asteraceae (commonly known as feverfew, carrot grass, congress grass and parthenium weed). Though native to the tropics and subtropics of Central and South America (Dhileepan and Wilmot Senaratne 2009; Adkins and Shabbir 2014), P. hysterophorus has invaded vast areas in a number of countries across five continents. Accidental or unchecked trade and transport across the borders is believed to be the primary reason for its global spread (Adkins and Shabbir 2014; Bajwa et al. 2016), in addition to its vigorous growth, high seed production and effective adaptive and dispersal mechanisms (Bajwa et al. 2016). In fact, it is one of the world’s top ten noxious weeds (Callaway and Ridenour 2004). In Asia, P. hysterophorus is believed to be unintentionally introduced with cereal and grass seed shipment from America during the 1950s (Rao 1956; Bhowmik and Sarkar 2005). However, in India it has been first reported in 1810 in Arunachal Pradesh (Gnanavel 2013) and there is mention of this weed in a book namely “Hortus Bengalensis” by Roxburgh as early as 1814. Subsequently, it became invasive in most of the Indian states but there are very few reports of occurrence of P. hysterophorus in Kashmir Himalaya (Yaqoob et al. 1988).

To predict the distribution of P. hysterophorus at global, regional (India) and local (Jammu & Kashmir) spatial scales under various RCP scenarios in future (2050 and 2070), we used Maximum Entropy Algorithm (MaxEnt), a widely used machine-learning software (Bradley 2009; Kumar and Stohlgren 2009; Elith et al. 2010; Lamsal et al. 2018), for its easy implementation and use of presence-only data as absence data is rarely available or reliable (Phillips et al. 2006). To improve the model performance, the usually spatially biased presence records (Lake et al. 2020) were corrected for sampling and latitudinal biases. We selected occurrence records only within the specified habitat range using a particular tool in SDM Toolbox. Though several studies that have modelled the distribution of P. hysterophorus at global level (Kriticos et al. 2015; Mainali et al. 2015) and in India (Ahmad et al. 2019) are available, the present study was aimed to uncover and compare the differences in the distribution of this species at different spatial scales under various climate change scenarios in two time periods of 2050 and 2070. Also in the present study, unlike previous studies, pairwise comparison in the extent of overlap in the niche of P. hysterophorus between the native and introduced regions was undertaken following the ordination technique (PCA-env) using Schoener’s D (Broennimann et al. 2012; Datta et al. 2019). Niche dynamics was further analysed using several indices (stable, unfilling and expansion) which were statistically tested using niche equivalency and niche similarity tests. Since we suspected that P. hysterophorus may move to previously uninvaded areas in Jammu & Kashmir, the rate and direction of core distributional shift in P. hysterophorus was figured out which has implications for deploying rapid response measures for the management of this species in Kashmir Himalaya.

Materials and methods

Species occurrence data

The occurrence data of P. hysterophorus with georeferenced geographic coordinates were collected from its native and alien ranges. A total of 7185 records were compiled which were obtained from different sources, such as Global Biodiversity Information Facility (GBIF) (http://data.gbif.org/), an open access database for distribution of species across the globe and for India, CABI (Centre for Agriculture and Biosciences International) (http://www.cabi.org/isc/datasheet/45573), Indian Biodiversity Portal, published records on P. hysterophorus in journals, books, reports and other published literature, herbarium records of Botanical Survey of India, JeevSampada, Kashmir University Herbarium (KASH) and field survey records of authors.

Since being important for better predictive modelling, the occurrence records were thoroughly cleaned for spatial errors using Diva GIS (Hijmans et al. 2001) which was accessed in July 2019 and only georeferenced localities were included. Data without or with obviously erroneous coordinates (e.g. 0.0) were discarded. The selected records were first converted to decimal degrees as it is the best system for the digital notation of geographic coordinates and then converted into an MS excel format with species name, latitude, longitude, altitude, location and data source. The duplicate records were then removed. The file containing remaining occurrence points was converted to a CSV format containing only Species ID, Longitude and Latitude.

Correction of sampling bias

The input data i.e. the presence records are usually spatially biased towards more easily accessed and better surveyed areas (Phillips et al. 2009; Kramer-Schadt et al. 2013) and eventually can translate into environmental bias due to greater representation of certain features of these better surveyed regions (Kramer-Schadt et al. 2013). As most of our data were derived from opportunistic observations and herbaria, the potential effect of sampling bias in our data was acknowledged and taken into account. Firstly, to reduce the bias due to spatially auto-correlated occurrence records plus spatial clusters of localities, the Graduated Filtering method (Brown 2014) was used. This method addresses the problem by spatially filtering the data to a single point within the specified (here 10 km) Euclidian Distance (Boria et al. 2014) and thus rarefied the occurrence points to 1674. After extracting values from raster at the location of presence points using QGIS, the null values were removed, reducing the data set to 1644 points. Furthermore, the bias correction features available in MaxEnt, allows the users to include bias files for background selection. These files, besides controlling the selection of background points, also facilitate to control density of background sampling. Background points when compared to presence occurrence points aid in differentiating the environmental conditions suitable for the potential occurrence of a species (Brown 2014). We used the Correcting Latitudinal Background Selection Bias Tool in the SDM Toolbox (Cao et al. 2016) to address the issue of biasing the selection of background points (or pseudoabsence points) towards the poles. The level of bias depends on the breadth of latitudes the analysis covers (Brown 2014). It provides solution in three steps: (a) Creation of a BiasFile for Coordinate Data (BFCD) in MaxEnt (downloadable from http://sdmtoolbox.org/technical-info). This file accounts for pseudoabsence sampling biases, (b) Background Selection: Sample by Buffered MCP (buffer distance was taken as 10 km). It limits selection of background points to only feasible areas of dispersal; (c) Clip BFCD by Background Selection Bias File. It merges the two bias files. The resultant bias file was used for background selection in MaxEnt.

Environmental data

To model potential global distribution of P. hysterophorus, a total of 19 grid-based bioclimatic variables (Table 1) for the current and future time periods were obtained at a spatial resolution of 30 arc-seconds which corresponds to a pixel size of approximately (~ 1 sq.km) from Worldclim version 1.4 (www.worldclim.org/; Hijmans et al. 2005). We used lowest available grain size of bioclimatic variables to increase model accuracy and transferability (Manzoor et al. 2018). Since it is widely accepted that most climatic variables are highly correlated with each other, it is essential to perform multi-collinearity analysis in habitat suitability modelling as correlated environmental variables negatively affect model performance (Pearson et al. 2007). Though MaxEnt has been found to cope well with collinearity (Elith et al. 2011), Pearson correlation analysis was used to omit highest correlative predictors (Dormann et al. 2013). A commonly used  threshold of Pearson’s correlation coefficient (r > 0.7) was  used (Dormann et al. 2013). All the 19 bioclimatic raster layers were tested for collinearity by calculating pairwise correlations between them and highly correlated variables (Pearson’s correlation coefficient r > 0.75) were omitted (Syfert et al. 2013; Cao et al. 2016; Wei et al. 2017). Discarding highly correlated variables reduces autocorrelation of input environmental data and minimizes the impact of multicollinearity and overfitting of the model. Finally, only nine variables (Bio 1, Bio 2, Bio 3, Bio 4, Bio 5, Bio 12, Bio 14, Bio 15 and Bio 19) were used (Table 1) to generate the niche model under current and future climatic conditions. The datasets of selected nine bioclimatic variables were converted into ASCII format in ArcGIS 10.3 as required in MaxEnt.

Table 1 List of 19 bioclimatic variables from the WorldClim database (Hijmans et al. 2005) used for constructing the multivariate niche space for Parthenium hysterophorus
Full size table

Owing to the fact that P. hysterophorus is widespread in all the continents except Europe and Antarctica (Mainali et al. 2015), we studied its future distribution at three spatial scales under four future greenhouse gas concentration trajectories, commonly known as RCPs (Representative Concentration Pathways: RCP 2.6, RCP 4.5, RCP 6.0 and RCP 8.5) which were used by IPCC (Intergovernmental Panel on Climate Change) in its Fifth Assessment Report (AR5). We selected a GCM i.e. Global Circulation Model, HadGEM2-AO (Hadley Global Environment Model 2–AtmosphereOcean). This model is designed to run for the major scenarios as recommended by IPCC (2014) and has been used by a number of workers for species distribution modelling (Ahmad et al. 2019; Banerjee et al. 2019). The datasets were downloaded for two time periods: 2050 (average for 2041–2060) and 2070 (average for 2061–2080) from the WorldClim website (http://www.worldclim.org). In order to maintain consistency, we selected the same independent predictors (Bio 1, Bio 2, Bio 3, Bio 4, Bio 5, Bio 12, Bio 14, Bio 15 and Bio 19) as were used for mapping current distribution of P. hysterophorus. However, all the 19 bioclimatic variables were used for niche dynamic analysis (Table 1).

Modelling approach

To model the current and future distribution of P. hysterophorus under different climate change scenarios, we used MaxEnt (Maximum Entropy Niche Modelling software, version 3.4.1), a machine-learning method that employs the principle of maximum entropy to approximate the unknown probability distribution of a target species based on presence-only data (Phillips et al. 2006). We preferred it owing to its easy use, higher predictive accuracy, functionality to use presence only data and relatively better performance than other methods (Reside et al. 2019; Chen et al. 2020), and popularity (Merow et al. 2013). Also MaxEnt has successfully outperformed classical presence-only models like DOMAIN, ENFA, BIOCLIM (Hirzel et al. 2002; Elith et al. 2006; Ward 2007) and is believed to be most reliable amongst available machine learning methods (Guisan et al. 2007; Elith and Graham 2009; Fourcade et al. 2014). We ran MaxEnt mostly with default settings but the random test percentage was kept equal to 25% and the maximum number of iterations that permit the algorithm to get close to convergence was set to 5000 instead of default 500 allowing the model to have sufficient time for convergence. MaxEnt has an ability to run a model multiple times and finally averages the results from all models created. Thus, model validation was done using sub-sampling option with 10 replications and the default “logistic” format was used which is a continuous map with an estimated probability of species’ presence ranging from 0 to 1. To further improve model performance, we used the bias file obtained using Correcting Latitudinal Background Selection Bias Tool in the SDMtoolbox (Brown 2014) instead of using default randomly selected background points. Further, to calculate significant contribution of each predictor to the model, Jackknife procedure was used. This configuration has been effectively used in a wide range of niche modelling studies (Phillips and Dudík 2008; Yang et al 2013; Yi et al. 2016; Esfanjani et al. 2018).

Model evaluation

The main outputs of MaxEnt included ROC curve, response curves, Jackknife tests of variable importance and probability maps. The capacity of the model to differentiate between presence and absence states was determined by using the Area Under the Curve (AUC) of the Receiver Operating Characteristics (ROC) plot test statistics. The strength and accuracy of the model is commonly determined by AUC values ranging between 0 and 1 whereby AUC values of 0.5–0.7 are considered low, 0.7–0.8 is ‘acceptable prediction', 0.8–0.9 is ‘excellent’ and > 0.9 is ‘outstanding prediction’ (Pittman and Brown 2011). MaxEnt output is continuous data with values ranging from 0 (representing lowest) to 1 (representing highest) probability of potential distribution. In addition, the Boyce Index commonly used for presence only models was calculated using ‘Ecospat’ package in R (version 3.5.3) for model evaluation. This index is considered to be adequate especially in case of MaxEnt as it uses background data instead of true absences (Di Cola et al. 2017). The Boyce index, a threshold independent metric, measures Spearman’s Rank correlation coefficient which ranges from -1 to 1. While negative values signify false or erroneous predictions, values close to zero imply no better than random results and positive values close to 1 indicate good or correct model predictions (Boyce et al. 2002).

For area change analysis, we imported our MaxEnt output data and reclassified it into five categories of habitat suitability, namely unsuitable (0–0.2), less suitable (0.2–0.4), moderately suitable (0.4–0.6), suitable (0.6–0.8) and highly suitable (0.8–1.0). The suitable area predicted for P. hysterophorus under present and future climate scenarios was calculated by multiplying the number of presence grid cell to their spatial resolution. Data for India and Jammu & Kashmir were extracted from world data file using ‘Extract by Mask’ in Spatial Analyst tools of ArcGIS. Further, the data were given proper projection for geometrical analysis. All GIS operations were performed in ArcGIS 10.3 (ESRI).

Core distributional shifts

To further examine the trend of suitable area change in Jammu & Kashmir, we calculated and compared the centroids of current and future suitable areas by using the python-based GIS toolkit available in the SDMtoolbox V2.4. The SDMs (current and future) were converted to binary maps [presence (1) and absences (0)] using MaxEnt produced threshold. MaxEnt software generated 10 percentile training presence threshold with a value of 0.404. The models were reclassified 0 (absence) values < 0.404 and 1 (presence) > 0.404. The core distributional shifts of the P. hysterophorus were summarised using the binary maps. In this analysis, the species distribution was reduced to a single central point known as centroid and a vector file was created depicting the direction of shift in species ranges over time. The centroid was calculated by averaging the latitude and longitude of all MaxEnt predicted suitable/presence input pixels. The distance in centroid shift was calculated using distance measure tools in GIS software and the average annual shift in the distance of P. hysterophorus was estimated.

Niche dynamics

Niche dynamics of P. hysterophorus was studied using the analytical framework proposed by Broennimann et al. (2012) and modified by Datta et al. (2019). Climatic data related to all the 19 bioclimatic variables (Table 1) of all the occurrence points of P. hysterophorus in both native (Americas) and introduced regions (Africa, Asia, Australia and Oceania) were pooled to calibrate a principal component analysis on the entire environmental space of the species (PCA-env). The first two components of the PCA-env were taken to define the “global PCA space” which was subsequently divided into 100 × 100 bins of equal size with each bin representing a unique set of climatic conditions. Gaussian Kernel density function was used for smoothing the density of occurrences and density of environmental grid cells in each bin of the global PCA space. In the native as well as each of the introduced continents, 10,000 random background points were generated separately to account for the available (background) environments using Research tools function in QGIS software. Bias due to geographic differences in the range size of P. hysterophorus was also corrected as suggested by Datta et al. (2019). This allowed us to compare and visualize niche dynamic analysis between the native and each of the introduced regions in the same ‘global PCA’ space.

Schoener’s D, a measure of the degree of niche overlap, was calculated as a bootstrapped estimate with 95% confidence interval using 100 resampling iterations of uncorrected occurrence density of P. hysterophorus as proposed by Datta et al. (2019). Schoener’s D ranges between 0 and 1 with value of 1 indicating complete overlap and no niche shift, and 0 indicating no overlap. The effect of the exclusion of 25% of rare climatic conditions on the niche overlap index (D) was also investigated. Single-tailed niche similarity and equivalency tests were carried out following the methodology described by Broennimann et al. (2012) using uncorrected occurrence densities (Datta et al. 2019). The null hypothesis for the niche similarity test assumes that the observed niche overlap is due to similarity in the climatic conditions of the two ranges for which the niche overlap is compared. Using an iterative process with 100 iterations, the niche in an exotic range was randomly shifted within the available environmental space while keeping the niche constant in the native range. Since niche overlap index was calculated for each iteration, a simulated distribution of niche overlap values was obtained. When the observed niche overlap is significantly lower than the simulated niche overlap values (P < 0.05), niche shift is indicated. For the equivalency test, the null hypothesis posits that niche overlap between the two ranges that are compared remains consistent even if the occurrence data in the two ranges are pooled and reallocated randomly between the ranges. A randomization procedure with 100 permutations was used wherein the species occurrences of the two ranges were pooled and then randomly reallocated between the two ranges and the simulated overlap index was calculated for each such iteration. The observed value of niche overlap when significantly lower (P < 0.05) than the simulated niche overlap value indicates niche shift. Niche unfilling, niche expansion and niche stability for both analogous and entire climatic space were also calculated using Ecospat package (Di Cola et al. 2017) implemented in R (version 3.5.3) to provide a more complete depiction of niche dynamics. The effect of progressive exclusion of rare climatic conditions (up to 25 percentile) on various niche dynamic indices was also studied.

Results

Current distribution

Current occurrence of Parthenium hysterophorus is presented in Fig. 1. It is apparent that in addition to its native regions of Central and South America, Parthenium is quite common in Indian sub-continent, South-eastern Asia, tropical/subtropical Australia, Southern and Eastern Africa, Madagascar and many Oceanic islands with warm climates.

Fig. 1
figure 1

Current distribution of P. hysterophorus in its native and non-native ranges

Full size image

Model performance and variable importance

Maxent was run with a dataset of 1644 occurrence points of P. hysterophorus using only the current climatic data. In this run, 25% of the occurrence records were withheld for model substantiation and average of ten replications was used to estimate the performance of the model on the basis of AUC (area under cover) of the ROC (receiver operating characteristic). Model calibration with a mean training AUC of 0.873 and a test AUC of 0.871 (Table 2) indicated high accuracy of our model in discriminating suitable and unsuitable habitats (Fig. 2). In order to identify contributions of each of the bioclimatic variables employed for modelling habitat suitability, we used percent contribution and Jackknife procedure. Out of the nine bioclimatic predictors’ used for the present study, Isothermality (bio 3) and Annual Mean Temperature (bio 1) were seen to be the best predicting variables contributing together about 56.8% to the predictive model, followed by Annual Precipitation (bio 12) and Temperature Seasonality (bio 4) contributing a total of 31.1% to the predictive model. However, Maximum Temperature of Warmest Month (bio 5) was the least contributing variable with percent contribution of 0.3% (Table 3).

Table  2 Area under curve (AUC) values for Current Scenario for training and testing shown in the table are averages of the ten replicated runs
Full size table
Fig. 2
figure 2

Receiver operating characteristic (ROC) curve for P. hysterophorus (Current scenario)

Full size image
Table 3 Contribution of various variables to the MaxEnt Model
Full size table

The Maxent internal jackknife test which is yet another method of testing the importance of each predictor showed that the predictor with highest gain when used in isolation was bio 4, which, therefore, appears to contain the most useful information by itself. The explanatory variable that decreases the gain the most when it is omitted is bio 12, which, therefore, appears to have the most information that is not present in the other bioclimatic predictors (Fig. 3). The results were also validated on the basis of the analysis of omission/commission and predicted area was utilized to determine whether presence data occurred in suitable or unsuitable habitat in relation to the given threshold by MaxEnt. The omission/commission analysis (Fig. 4) showed the omission on test samples (green line) are in harmony with the predicted omission rate (black line) from the MaxEnt distribution indicating suitable habitat exists above the threshold. The omission rate and predicted area as a function of the cumulative threshold for P. hysterophorus Area Under Curve (AUC) of the Receiver Operating Characteristic (ROC) curve (Fig. 4) is a parameter that was used to evaluate the predictive ability of the generated model. It is apparent that MaxEnt showed better discrimination of suitable and unsuitable areas of the species in the analysis of AUC. Moreover, positive Boyce Index (Spearman rank correlation coefficient = 0.986) validated that our results were consistent with the significant model.

Fig. 3
figure 3

Jackknife of regularized training gain {Dark blue bars indicate how well a model performs using only that feature compared to the maximal model (red bar), and light blue bars indicate how well a model performs excluding that feature. Thus, important variables can either have (1) large dark blue bars, indicating strong (but perhaps non-unique) contribution to presences; (2) short light blue bars, indicating no other variable contains equivalent information; or (3) both, indicating the variable is independently predictive}

Full size image
Fig. 4
figure 4

MaxEnt testing and training omission analysis and predicted area for P. hysterophorus

Full size image

Predicted change in suitable habitat area

A slight increase in habitat suitability from 15.67% under current scenario to 16.02% under RCP 8.5 was predicted in 2050 and to 16.19% under RCP 8.5 in 2070 (Tables 4 and 5). In India, however, a reverse trend was seen i.e., reduction in suitable area in both the future climate scenarios as compared to present scenario. The approximate suitable area under current scenario is believed to be 65.92% and decline to 52.19% was observed under RCP 8.5 in 2070. Though shrinkage of suitable areas is predicted in India, interestingly in Jammu & Kashmir, expansion of suitable area of P. hysterophorus was observed under 2050 climate scenario followed by slight increase of the same by the year 2070 (Fig. 5, 6, 7, 8, and 9). In Jammu & Kashmir, the current estimated potential suitable area for P. hysterophorus is approximately 8.91% which is predicted to increase up to 11.15% under RCP 8.5 in 2050 and 12.16% under RCP 8.5 in 2070 (Tables 4 and 5).

Table 4 Suitable area (sq. km) for occurrence of P. hysterophorus at present (current) and in future (2050) (Percent suitable area is given in parenthesis)
Full size table
Table 5 Suitable area (sq. km) for occurrence of P. hysterophorus at present (current) and in future (2070). (Percent suitable area is given in parenthesis)
Full size table
Fig. 5
figure 5

Modelled current distribution of P. hysterophorus in the world, India and Jammu & Kashmir

Full size image
Fig. 6
figure 6

Modelled global distribution of P. hysterophorus under various RCP scenarios in 2050

Full size image
Fig. 7
figure 7

Modelled global distribution of P. hysterophorus under various RCP scenarios in 2070

Full size image
Fig. 8
figure 8

Modelled distribution of P. hysterophorus in India under various RCP scenarios in 2050 and 2070

Full size image
Fig. 9
figure 9

Modelled distribution of P. hysterophorus in Jammu & Kashmir under various RCP scenarios in 2050 and 2070

Full size image

Categorization of suitable area into moderately suitable, suitable and highly suitable revealed that of the total area suitable for P. hysterophorus under current climate conditions (2,11,42,336 sq km, approx. 15.67% of total global land cover excluding Antarctica) 1,07,61,810 sq. km were moderately suitable, 85,53,960 sq. km were suitable and 18,26,566 sq km. were highly suitable. Globally, the regions predicted suitable for P. hysterophorus include Indian subcontinent, Eastern parts of Africa, parts of South and North America and Australia and South-eastern parts of Asia including China. Under the current scenario, about 65.92% of India seems favourable for P. hysterophorus. The climatic simulations in the present study predicted decrease in suitable area for P. hysterophorus in India (Fig. 8) with Eastern Ghats, Western Ghats, Central India and a significant portion of Himalaya remaining suitable for P. hysterophorus while as moderately suitable areas in future would include, Karnataka, Tamil Nadu, Telangana, Maharashtra, Madhya Pradesh, Andhra Pradesh, Chhattisgarh, West Bengal, Uttar Pradesh, Uttrakhand, Himachal Pradesh, Punjab, Haryana and Delhi. Although our model does not predict highly suitable areas in Jammu & Kashmir, but all districts of Jammu and Kashmir are predicted to offer moderate as well as suitable habitats for this weed.

Core distributional shifts

The current centroid of P. hysterophorus is in south of Jammu and Kashmir within the geographic coordinates of 74.50 E longitude and 33.39 N latitude. The centroid of P. hysterophorus is predicted to shift by an average of 20.48 km in the north-west direction by 2050 and by 36.83 km in 2070. The future centroids (2050 and 2070) are mostly located around areas surrounding the Hirapora Wildlife Sanctuary (Table 6; Fig. 10) which is situated about 70 km south of Srinagar, in Shopian district of Jammu & Kashmir. This sanctuary has recently been placed in the eco-sensitive list of the Union Ministry of Environment, Forests and Climate Change. It is an abode to many rare and threatened plants and animals. Further, the estimated change in centroid was at an average of 0.68 km/year by 2050 and shift was noticed at an average of 0.73 km/year by 2070 (see Fig. 10).

Table 6 Centroid shift of P. hysterophorus under various RCP scenarios between current and future (2050 and 2070)
Full size table
Fig. 10
figure 10

Predicted distributional centroid shift in P. hysterophorus under various RCP scenarios in Jammu & Kashmir. (Black colour arrow) represents centroid shift from current to future)

Full size image

Niche dynamics

The  PCA-env  based on all the 19 bioclimatic variables  revealed that the axis 1 retained about 48.3% variation while as axis 2 retained about 32.5% variation (Fig. 11). Furthermore, niche overlap (Schoener’s D) between the native and introduced regions varied across all pairwise comparisons. The niche overlap was very low for Oceania (D = 0.24), moderate for Asia (D = 0.59) and high for Africa (D = 0.69) and Australia (D = 0.70) (Fig. 12; Table 7). Except Oceania, the observed values for equivalency test were significantly lower than the random niche overlap indicating non-equivalency of the climatic niches (Table 7). As is evident from the data (Table 7), the niche similarity test did not yield any significant result. Perusal of the data related to niche dynamic indices (Table 8) revealed very low niche expansion but considerable stability irrespective of whether the niche dynamic indices were computed using analogous or entire climates between native and non-native regions. Niche unfilling was usually low except Oceania (Table 8). Even in the case of Oceania, niche unfilling was more (0.748) when computed on the basis of entire climate as against a value of 0.370 when calculated using the analogous climate. Perusal of data in Table 9 reveals that exclusion of 25% of rare climatic conditions did not alter the niche overlap index (D). It is more or less true for other niche dynamic indices where progressive exclusion of rare climatic conditions did not affect niche dynamic indices with the exception of niche unfilling in the Oceania which decreased from 0.37 to 0.186 with the exclusion of rare climatic conditions (Table 10).

Fig. 11
figure 11

PCA plot based on 19 bioclimatic variables used to determine niche overlap dynamics of P. hysterophorus

Full size image
Fig. 12
figure 12figure 12

Multivariate climatic niche space of P. hysterophorus in native region of Americas (a) and pairwise comparison of niche dynamics in the introduced regions of Africa (b), Asia (c), Australia (d) and Oceania (e). Unfilled, stable and expanded niches are represented by green, blue and red shades, respectively. The grey shading shows the smoothed occurrence density in the native niche space (a) and in the introduced niche space (b–e). The bold lines mark the available environment in each range (green native, red introduced). Histograms show the observed niche overlap (d) between the native and introduced regions (bars with a diamond) and simulated niche overlap (grey bars). P values for similarity and equivalency tests are also given

Full size image
Table 7 Niche overlap (Schoener’s D) and 95% confidence intervals of P. hysterophorus between the native and non-native ranges
Full size table
Table 8 Niche dynamic indices calculated for Parthenium hysterophorus between analogous and entire available climates of native and non-native ranges
Full size table
Table 9 Effect of excluding 25% of rare climatic conditions on niche overlap
Full size table
Table 10 Effect of progressively excluding rare climatic conditions (up to 25 percentile) on different components of niche dynamics indices
Full size table

Discussion

It is a well known fact that P. hysterophorus is one of the widespread invasive alien species and is known to occur in about 50 tropical and sub-tropical countries across the world (Cowie et al. 2020) which indicates that it is not only able to overcome various environmental constraints but also adapts to a wide range of environmental conditions (Kohli et al. 2006; Cowie et al. 2018). Among the climatic variables, isothermality (bio 3), annual mean temperature (bio 1), annual precipitation (bio 12) and temperature seasonality (bio 4) were the best predictors of the distribution of P. hysterophorus. Importance of temperature and precipitation in predicting the distribution of P. hysterophorus has also been reported by Ahmad et al. (2019). In fact, climatic variables, particularly temperature, rainfall and their interaction (seasonality), in conjunction with the species’ ecology, physiology and adaptability often predict the potential distribution of a species (Sakai et al. 2001). The importance of isothermality, the quotient of the difference between the daily and annual temperature ranges, in predicting the distribution is quite obvious because P. hysterophorus grows in different regions across the world where significant variation in daily and seasonal temperatures exist. Aguirre-Gutierrez et al. (2015) also reported similar relationship between isothermality and widespread distribution of Mexican white pines. The role of temperature in predicting the distribution of P. hysterophorus is because of favourable influence of spring and summer season temperature on the germination of its seeds and seedling growth in the areas of its occurrence (Friedman and Rubin 2015) but availability of water in soil, determined by the amount and frequency of precipitation, may decline with increase in temperature under the warming climate. Though this situation may apparently limit the growth and distribution of a species (McConnachie et al. 2010) but P. hysterophorus being a C3-C4 intermediate species (McConnachie et al. 2010) is known to tide over water stress through drought resistant seeds and persistence of young plants as rosettes (Annapurna and Singh 2003). Interestingly, decreasing soil moisture is reported to promote early flowering and seed set in adult plants of P. hysterophorus (Bajwa et al. 2017) which clearly establishes the role of temperature, rainfall and their interaction in the distribution of this species. Such a conclusion draws support from many other studies that have also emphasized the importance of temperature and precipitation in the geographic distribution of invasive species (Thuiller et al. 2007; Bradley et al. 2010; Priyanka and Joshi 2013; Li et al. 2019; Yan et al. 2019).

Since models built solely on the basis of occurrence data from native range are known to be poor predictors of invasive ranges (Broennimann and Guisan 2008), presence data from both native and non-native ranges was used in the present study because it improves model prediction (Mainali et al. 2015; Bocsi et al. 2016). AUC values averaging 0.871 for the current distribution also indicated a reliable performance of the predictive model with good ability to differentiate between presence and absence areas for the target species (Barik and Adhikari 2011). The highly positive value for Boyce index (Spearman’s Rank Correlation Coefficient = 0.986) further validated the predictive ability of the model. However, uncertainty associated with the accuracy of the model predictions when transferred in space and time remains a major concern and we acknowledge the fact that the predictions from SDM’s are confounded by sampling biases, traits of the target species and omission of the important predictors, such as biotic interactions, population pressure, habitat structure (Yates et al. 2018).

The present study predicted a modest expansion in the distribution range of P. hysterophorus on a global scale under future climate scenarios, especially under RCP scenario 8.5 in 2070 but decrease in the suitable area for P. hysterophorus was predicted at the regional scale (India). In contrast, increase in the suitable area was predicted at a local scale (Jammu and Kashmir). Variation in the spatial distribution under different spatial scales has also been reported by Nielsen et al. (2008). These results bring out the need for predictive modelling of species distribution at different spatial scales which is not the usual practice.

The results obtained during the present study are consistent with numerous studies that have also predicted change in the range expansion of invasive species at the global level due to climate change (Walther et al. 2009; Barbet-Massin et al. 2013; Bellard et al. 2013; Adhikari et al. 2019). Decrease in the suitable area of P. hysterophorus at the regional level has also been reported by Bezeng et al. (2017) for South Africa, Ahmad et al. (2019) for India and Lamsal et al. (2018) for the Himalayan region. Increase in the suitable area at the local level (Jammu & Kashmir) unlike India, for P. hysterophorus, was further studied to predict the trend in the suitable habitat area changes by comparing the centroids of present and future models (Brown 2014). The centroids shift at an average of 0.68 km/year around 2050 and 0.73 km/year by 2070 is quite significant and the centroids were found to be shifting within and towards surrounding areas of the Hirapora Wildlife Sanctuary which is a cause of concern keeping in view the spreading nature of the species.

Present study also revealed that P. hysterophorus shows moderate to high niche overlap (Schoener’s D) between the native Americas and the introduced continents except Oceania. Further analysis (Table 11) pointed out that higher niche overlap between Americas and Asia is due to similarity in their climatic conditions but same does not hold true for higher overlap between native Americas and introduced regions of Africa and Australia. Thus, it clearly brings out that niche overlap is not always a manifestation of climatic overlap between the native and introduced regions. Such inferences are consistent with the study of Datta et al. (2019). Furthermore, insignificant similarity test though points towards a more similar climatic niche between native and introduced regions but the equivalency test revealed that native niche is not equivalent to any of the climatic niches of this species in the introduced regions except Oceania. Such an observation is consistent with the results of Aguirre-Gutierrez et al. (2015). Thus, it can be inferred from these results that P. hysterophorus will continue to invade the areas with climate similar but not necessarily equivalent to the climate prevailing in the native Americas.

Table 11 Pairwise comparison of overlap (Schoener’s D) of available climatic conditions and overlap of niche space between native and invasive ranges
Full size table

Niche dynamic analysis revealed high degree of niche stability between the native and introduced regions irrespective of whether analogous or entire climates were considered. It indicates that the species occupies more or less similar climates though not equivalent in the native and introduced regions. Low unfilling indicates reduced climatic space that could be colonized by this species in future except Oceania where less residence time or constraints to dispersal could not have allowed the species to occupy all the available climatic niches. Low expansion values also reflect that P. hysterophorus is not showing any niche shift in the introduced regions. The observation of high niche stability and little niche expansion and low unfilling is consistent with many previous studies on various invasive species (Petitpierre et al. 2012; Callen and Miller 2015).

Conclusions

Based on the detailed analysis, it can be safely concluded that the P. hysterophorus exhibits niche conservatism which may restrict its spread under climate change in the regions where it has already spread and occupied most of the climatic niche space but may spread to new areas where it finds suitable climate niche. Thus, the present study provides a strong rationale for studying niche dynamics to better predict the distribution of alien species under climate change as modelled distribution P. hysterophorus differed at different spatiotemporal scales under changing climate and more importantly in Jammu & Kashmir. Since the species is expected to spread to previously uninvaded areas in Kashmir with debilitating ecological and economic consequences, the study would serve the purpose of raising the concern with land managers about the impending threat from invasive alien species.