Predicting Invasion Probability from Botanic Gardens using Exotic Species Traits

Preventative management, such as framework-based assessment, considered as the best option for invasive species management. Alternatively, risk assessment can be conducted based on traits of occurred invasive species to build prediction system for invasive risk assessment. This study aimed to test whether trait-based assessment system can differentiate the escaped from non-escaped exotic collections of botanic gardens and to compare the reliability of trait-based versus framework-based risk assessment on differentiating these escaped from non-escaped exotics. In this study, Bayesian logistic regression analysis was conducted to assess the reliability of framework-based and trait-based risk assessment systems. For trait-based system, clear effect of leaf trait, height, and dispersal method to escape probability was detected. For framework-based system, clear effect of Tropical Weed Risk Assessment Protocol on escape probability was detected. Leaf trait, dispersal method and height are reliable predictors for escaped probability of botanic gardens exotic collection. The fact that the reliability of trait-based assessment systems is better than the commonly used framework-based system is the main novel finding in this study. This finding implies that trait-based is better than framework-based for invasive species risk assessment approach in Indonesian botanic gardens. Trait-based assessment also a relevant tool to support management with limited resources to conduct adequate early risk assessment.


INTRODUCTION
Apart from its important role in ex-situ plant conservation (Hidayat et al., 2017), botanic gardens are important sources of invasive plant species (Heywood, 2011).There is an urgent need to predict the risk of spread of exotic and invasive species from botanic gardens (Corlett, 2010).The ability to assess weed risks will allow botanic gardens to set priorities for the management of their exotic collections (Corlett, 2010;Heywood, 2011).Several studies provide some preliminary ideas about assessing invasive risk from tropical botanic gardens exotic collections (Daehler, 2009;Dawson et al., 2009a;Dawson et al., 2011).Framework-based invasive species risk assessments for botanic gardens were also developed and implemented for sub-tropical areas of Australia (Virtue et al., 2008) using Botanical Garden Weed Risk Assessment Protocol (BG-WRAP) and tropical Africa (Dawson et al., 2009b) using Weed Risk Assessment (WRA) system.
Preventative management is the ideal approach for invasive species management and already implemented in multiple geographical contexts, such as regional-level (Foxcroft et al., 2008;Pritekel et al., 2006), country-level (Williams & West, 2000) and continent-level (Brunel et al., 2010).Technically, preventative management may consist of early detection, screening, and implementation of designated risk assessment framework for exotic species that are potentially invasive (Leung et al., 2012).
There are many risk assessment frameworks (e.g.WRA) that were suggested as reliable systems for global application (Chong et al., 2011).However, most of these assessment frameworks were need specific data, either traits data or other supporting information.
Consequently, the application of framework-based assessment may become unrealistic for users with limited data availability.This is because we need lots of specific data in the application of these frameworks.An assessment system should be simple and easy to apply, a userfriendly system.These framework-based systems could be irrelevant if the data are limited (Sheil & Padmanaba, 2011).Thus, the application of framework-based system will be difficult if the data needed are not available.
At some situations, we can only rely on limited trait information to conduct a risk assessment and early rapid risk assessments are mostly relevant for preventative management contexts, which is considered as the best option for inva-field, not all data of height obtained from direct measurement.There were 40 percent of data of height was collected from direct measurement and the other 60 percent was collected from secondary data in available databases such as http:// www.efloras.org/and http://hear.org.The "corrected" height data from database was acquired by fitting it into a model between the 40% direct measurement data versus their median of height ranges in databases (Figure 1).
Figure 1.Scatter plot between height value from direct measurement versus height value (median) from available databases.If the minimum height value was not stated in the database, we used the median value between zero and the maximum height that was stated in the database.
Due to limited fruit availability during the survey period, secondary seed mass data for escaped and non-escaped exotics from Kew Seed Database (<http://data.kew.org/sid/>) was used for seed mass data and defined as 1000 dry seed mass (mg).Minimum residence time was obtained from botanic gardens' collection catalogues of CBG (1930, 1963, 1977and 1988) and BBG (1989, 1999, and 2006).Minimum residence time of exotic species from other two botanic gardens (KBG and BRBG) was acquired from botanic gardens' official planting date records.
The retrospective logistic regression analysis was conducted to model the probability of escape of botanic gardens exotic collection into adjacent native forests based on the prescribed traits as independent variables (model 1).P(x) = e g(x) / (1 + e g(x) ) ………..(model 1) where P(x) is the probability of escape of exotic collection of botanic gardens and g(x) = a 0 + ß 1 m i + ß 2 n i + ß 3 o i + ß 4 q i + ɛ i where m i , n i , o i , and q i were specific leaf area (SLA), dispersal method, minimum residence time, and plant height of species i respectively.The value of P(x) was equal to 1 for escaped exotics and 0 for non-escaped exotics.The Bayesian logistic regression analysis conducted in R (Team, 2013) with Bayesian framework using Just Another Gibbs Sampler (JAGS) (Plummer, 2003), called from R using package jags UI (Kellner, 2015).
Bayesian logistic regression analysis was also conducted to model the escape probability and utilized WRA, BG-WRAP, and T-WRAP score as independent variables (models 2, 3, and Mount Slamet, Central Java Baturraden Botanic Gardens S 07 0 18.096' E 109 0 13.905' 4).P(x) = e h(x) / (1 + e h(x) ) ………..(model 2) P(x) = e i(x) / (1 + e i(x) ) ………....(model 3) P(x) = e j(x) / (1 + e j(x) ) ………....(model 4) where P(x) is the probability of escape of exotic collection of botanic gardens and h(x) = a 0 + ß 1 r i + ɛ i , i(x) = a 0 + ß 1 s i + ɛ i , j(x) = a 0 + ß 1 t i + ɛ i , where r i , s i , and t i are the score of WRA, BG-WRAP, and T-WRAP of species i respectively.The analysis procedure for model 1 was also implemented for Bayesian logistic regression analysis in model 2, 3, and 4. The deviance information criterion (DIC) values from the poste-rior results and area under the curve (ROC curve) were utilized to compare and visualize the reliability of all assessed models (models 1, 2, 3, and 4).The ROC curve of the model constructed using R package pROC (Xavier et al., 2011).

RESULTS AND DISCUSSION
There were 913 escaped exotic individuals and 23 species detected from all four study sites.These escaped exotics were mostly shrubs and herbs with only several small tree species (Table 2).There were also 996 non-escaped individuals   The lower reliability of framework-based assessment system relative to trait-based might be explained by several weaknesses of the application of these framework-based systems.First, most of the framework-based assessment systems are using scoring system in the quantification processes and these scoring processes may be a subject to the epistemic and linguistic uncertainties.Second, a priori knowledge can also become a source of bias in the scoring procedures of risk assessment framework such as WRA (Onderdonk et al., 2010).For instance, Matthews et al. (2017) showed that risk assessment may suggest different risk result due to the variation of the risk assessment systems and contexts.Finally, the source of uncertainty during the scoring processes may arise due to the level of details in the data and the assessor knowledge capacity.Thus, the implementation of practical guidance of the risk assessment framework application is a must to minimize bias and uncertainty in the scoring processes, the devil is in the detail (Onderdonk et al., 2010).
The finding in this study indicates several promising aspects regarding invasive species risk assessment implementation.First, the fact that trait-based assessment produced more robust prediction result than framework-based indicate that trait-based assessment approach is reliable and realistic.Traits used in this study are relatively easy to measure or collect but are reliable predictors.For instance, SLA is a relatively simple trait but a good predictor for invasion likelihood because SLA indicates ecological and physiological characteristic of plants such as shade tolerance (Lusk & Warton, 2007;Poorter, 2009) and growth rate (Gibert et al., 2016).Second, well established framework-based assessment system may not necessarily give reliable result.As stated before, the weakness of framework-based assessment is data demanding and this approach will become unrealistic and not reliable when those data are not available or incomplete.Lastly, simple and reliable traits for risk assessment can support managers and risk assessment user to conduct robust invasive species risk assessment under limited resources condition.Traits used in this study are relatively simple and easy to measure but produce good prediction in the risk assessment.Simple and easy to measure data will become a useful proxy to support the managers that have limited resources to conduct adequate risk assessment.

CONCLUSION
SLA, dispersal method and height are reliable predictor for escaped probability of botanic gardens exotic collection into adjacent native forest.There is no effect of residence time detected.Among all framework-based system tested in this study, only T-WRAP can reliably differentiate the escaped from non-escaped exotic collections.SLA performance is better than framework-based system on this risk assessment study, indicated by lower DIC value and visually by ROC curve.Due to its simplicity of measurement method and good predictive ability, trait-based risk assessment system can be implemented for exotic invasive rapid assessment.This rapid assessment is essential part of preventative invasive species management in Indonesian botanic gardens.

Table 1 .
Study sites and locations of line transect distance sampling conducted to detect escaped botanic gardens' exotic collections.