A risk assessment system for screening out invasive pest plants
from Hawai‘i and other Pacific Islands

  Curt Daehler and Julie Denslow  

report based on the following peer-reviewed publication:
Daehler, C. C., J. S. Denslow, S. Ansari, and H. Kuo. 2004. A risk assessment system for screening out
     invasive pest plants from Hawai'i and other Pacific Islands. Conservation Biology 18:360-368.

Summary of Findings

We tested the ability of a modified version of the Australia and New Zealand weed risk assessment system to identify pest plants in Hawai‘i and other Pacific Islands. We used information taken from outside Hawai‘i  to predict the behavior (“pest” or “not a pest”) for almost 200 plant species introduced to Hawai‘i and other Pacific Islands.  The screening system initially recommended further evaluation for 24% of these species, but an additional secondary screening was applied to this group, thereby reducing the rate of indecision to only 8%.    To independently test the accuracy of the screening system, we compared its decisions (pest or not a pest) to opinions of 25 expert botanists and weed scientists who had substantial field experience in Hawai‘i or other Pacific Islands. We asked the experts to rate each species as “major pest”, “minor pest” or “not a pest” in native or managed ecosystems. The screening system correctly identified 95% of major pests and correctly identified 85% of nonpests. Among minor pests identified by the experts, 33% were classified as nonpests by the screening system. Use of the screening system to assess proposed plant introductions to Hawai‘i or other Pacific Islands and to identify high-risk species used in horticulture and forestry would greatly reduce future pest plant problems and allow entry of most nonpests. The screening process is objective, rapid, and cost-efficient. With minor modifications, it is likely to be useful in many parts of the world.

Blank assessment sheet

General instructions for filling out the sheet

Testing the Screening System

1)     Between December 2001 and June 2002, approximately 200 plants were scored using the WRA system. About half of the plants were chosen from the Maui County Planting list. The remaining plants were taken from other planting lists for Hawai‘i or for other parts of the Pacific.

2)     Based on their WRA scores, species were placed into the following categories: Accept (not likely to be a pest; WRA score < 1 ), Reject (likely to be a pest; WRA score > 6), or Evaluate (requires further evaluation; WRA score = 1-6). About 27% of the species fell in the “Evaluate” category, and these species were passed through a second screening, which resulted in a quick decision (Accept or Reject) for most species originally in the “Evaluate” category.

Details for the second screening (used for species scoring between 1 and 6)

3)     In order to judge reliability of the WRA scores, the same list of 200 plant species was sent to 25 expert botanists/weed scientists who had first hand, detailed knowledge of invasive plant or weed problems in Hawai‘i and other Pacific Islands.  Some of the experts worked primarily in native ecosystems while others worked in managed ecosystems (e.g. agriculture or forestry plantations). These experts were asked to rate the 200 species as “major pest”, “minor pest” or “not a pest”, either currently or in the future.  

Definitions of major, minor and not a pest

4)     WRA decisions were compared with the responses from the 25 experts to obtain an independent assessment of how well the WRA system worked in Hawai‘i.

5)     Of the 200 species on the survey list, 172 species were used for the final analyses. Species that were rated by fewer than 3 of the expert surveyors were excluded.  We also excluded species native to Hawai‘i and species that were taxonomically uncertain. The  list of species NOT included in the analysis is provided here.


Summary of Results

Species list with WRA scores used for analysis (172 species)




Things to remember




General Discussion

Importance of objectivity, consistency and science in assessing risks

Anyone who spends time around plants develops personal opinions about whether certain plants are desirable or not. These opinions differ widely, based on personal experiences, and they have generated much disagreement, particularly with respect to assessing plant invasiveness. The WRA system that we employed in this study minimizes the role of personal opinion during the assessment process. The WRA is based on answers to about 50 questions, each relating in a logical or scientific (statistical) way to the risk of a plant becoming a pest. The answer to any one question doesn’t tell you much about whether a species is likely to become invasive, but by answering a series of independent questions, we can identify high risk species or pests, as our results have shown. Objectivity is maintained because

1)      The same set of questions was answered for each species.

2)      Consistent, pre-determined criteria were established for determining when a question should be answered “yes” or “no”.

3)      For each answer, the source (reference) was recorded, allowing anyone to evaluate the source of information used in an assessment. Anecdotal information or information appearing to be derived from personal opinion was avoided during the assessment process.  Answers to question in the WRA most commonly came from: scientific journal articles, reference books, electronic databases, and the Internet.

Furthermore, because the final assessment is based the additive contribution of each answer, changing one answer or adding an additional answer will usually not change the decision for an assessment.  Nevertheless, if new information comes to light after an assessment has been made, that new information can easily be incorporated to obtain a revised WRA score. One outcome of using the WRA system is that decisions are more decisions are more objective and consistent, compared to a survey of expert opinions (3), and this was also obvious from our study. However, high variability is not an inherent characteristic of expert committees, and it would certainly be possible to establish an objective expert committee that follows specific, transparent guidelines to make consistent and objective decisions. Such a committee could be used in conjunction with the WRA process, as explained below, to further reduce mistakes in classification that will inevitably be made within any WRA system.


Does the WRA system reject too many non-invasive species?

            If the WRA system “rejects” too many species that are not really pests, then attempting to follow the recommendations of the WRA system could potentially lead to unnecessary economic hardships. We found that most plant species were “Accepted” using the WRA system. The “Accepted” plants were extremely diverse and include popular landscape plants. Although we have only assessed a sample of 200 species, most of the “staple” species for the landscape industry that we have assessed so far received an “accept” rating. Among our sample of 200 species, 61% had an “accept” rating, but about 70 species on our list were forestry species already suspected of being invasive by US Forest Service personnel. The “accept” rate among common landscaping plants on our list was around 90%.


Let’s examine the list of species “rejected” by the WRA system but not generally found to be pests according to the expert surveys (Table 3). There are two interpretations of any disagreement between the WRA results and the expert surveys. First, the WRA may be wrong. Although information was found that makes these species statistically likely to be invasive (thus generating a high WRA score), there may be unknown factors or variables in Hawaii’s environment that will always prevent these species from becoming pests here. For example, Tamarix aphylla is not thought to produce seeds in Hawai‘i.  Another interpretation is that some or many of these species with high WRA scores  (Table 3) will become invasive in the future but are not yet recognized as pests by the expert surveys. 

By far, the most economically important species on this list are the two grasses, Paspalum vaginatum (seashore Paspalum) and Eremochloa ophiuroides (centipede grass).  Other species are minor forestry trees in Hawai‘i (Acacia auriculaformis, Bischofia javanica), fast growing agricultural legumes (Centrosema pubescens, Stylosanthes guianensis), or relatively rare horticultural plants (Muntingia calabura, Passiflora rubra, Pittosporum undulatum, Elaeagnus umbellata).

Usually, the WRA assessments are consistent with expert opinion (as our results have shown), but there are bound to be some cases where “experts” disagree with a WRA assessment.  In those cases, it can be enlightening to examine why as species had a high WRA score. Let’s examine the case of Paspalum vaginatum (seashore Paspalum).  Paspalum vaginatum scored high on the WRA for several reasons. For example, it is a grass; grasses are statistically more likely to become pests than other plant families. It has become a serious pest of wetlands in New Zealand; it forms dense stands in wetlands there, eliminating native species. It spreads rapidly by vegetative reproduction. All these traits contribute to the species’ potential to become a pest in Hawaii.  Nevertheless, its WRA score was 7, only one point above the cut-off of 6 for a rating of “Evaluate”. Note that our WRA assessment was based on plant traits for generic (wild) Paspalum vaginatum, and its is possible that breeds selected for turf in Hawai‘i have specific plant traits that will alter the answers for a number of questions on the WRA assessment, thereby changing the score and bringing the WRA decision in line with the expert survey results.   

If the WRA score is still high, but the species is clearly not a pest now, probably the best approach to resolve these cases is to have a small group of objective experts with broad field experience consider the WRA results together with their field observations, and recommend a change to the WRA recommendation if they are reasonably confident that the species will not be a serious pest. This same general procedure could be applied for any species that has been “rejected” by the WRA (Table 4).


Does the WRA “accept” too many pests?

At the opposite end of the spectrum is the question of whether the WRA is effective at “rejecting” most pest plants. Only 1 out of the 21 “major pests” identified from the expert surveys was “accepted” by the WRA system:   Fraxinus uhdei (tropical ash).  For Fraxinus uhdei, we encountered conflicting information that affected answers to some questions in the WRA. For example, a “personal observation” found on a webpage suggested that this tree forms dense monocultures, excluding native species. However, an article published in a peer-reviewed scientific journal concluded that more native plant species were found within Fraxinus uhdei plantations than in other plantation types, suggesting that its effect on native species was relatively low.  Similarly, an anecdotal and vague report of the species’ spreading was found on a webpage, but a published book indicated that planted stands “reseed themselves” but are “apparently only spreading in a few locations”.  In both, cases, the published information was favored over unconfirmed “observations” encountered on webpages. Of course, this policy increases the risk that very recent, accurate information will be excluded from the WRA.  As with high scoring species that do not seem to be pests, a small committee of objective field experts could examine the WRA results while also making use of their direct field-experience to determine if a species’ WRA classification should be changed. 

In addition to the 1 “major pest”, the WRA “accepted” 15 species that could be considered “minor pests” according to the expert surveys (Table 5). We are not aware of control programs specifically targeting any of these species at this time, suggesting that these “accepted” species are not considered high priority pests.


Pests of native forest versus pests of managed lands

The pests of native forests often have different ecological characteristics from the pests of managed lands (e.g. agricultural fields or plantations). The Australia/New Zealand WRA system was originally designed to identify pest plant in both types of environment, so it is most reasonable to evaluate the overall effectiveness of the system at identifying both pests of native forest and pests of managed lands, as was presented in the Summary Results. However, since some of our expert surveyists had experience in native forests while others had experience primarily in managed lands, it is possible to compare the WRA results with survey results from people working in native forests (Table 2) versus people more familiar with managed lands.   In general, the degree of agreement between the WRA decision and the expert surveys is similar for both groups of surveyors.


WRA numerical score in relation to pest seriousness

            An answer for each question in the WRA can potentially add or takeaway one or more points from the total WRA score. Answers that add to the WRA score reflect information about a species that increases the risk that it will be a pest. It would be natural to infer that a species with a higher WRA score is a bigger threat than one with a lower score, even if both scores fall within the same decision region. For example, a species scoring 15 theoretically has more potential to do greater harm than a species with a WRA score of 8, although both would be “rejected” because their scores are > 6.  Unfortunately, we cannot make a powerful test of this idea using our data because we did not ask our surveyists to make a quantitative assessment of the pest status of the survey species. Instead, the survey responses were categorical (major pest, minor pest, or not a pest).  We can convert these categories into numbers (2 = major, 1 = minor, 0 = not a pest) and average them across surveys to obtain a pseudo-quantitative measure of pest status for each species. However, we must use caution in interpreting this measure. For example, if one surveyior reported that a species is a “major pest” (score =2) and another says “not a pest” (score =0), the average is 1 (making it “minor pest”), which might or might not be a reasonable conclusion.  In general, converting a categorical variable into a quantitative one will also create a lot of scatter around any predicted relationship between the two quantitative variables. Nevertheless, when mean survey score versus WRA score was plotted, we found significantly, positive linear regression (Figure 3).  There is a fair amount of scatter around the regression line, but on average, a higher WRA score is predicted for species with higher expert survey scores. This remains true when the regression is limited to smaller WRA score intervals (e.g. 0-8 or 8-15), indicating that even within WRA classification categories like “Reject”, higher WRA are statistically predictive of increased pest status. However, we also need to keep in mind the fairly high degree of scatter around the regression line. There is a statistical relationship between increasing survey score and increasing WRA score, but we can't expect the quantitative WRA score for any one species  to precisely predict the total harm that might be caused by a species.



Pheloung, P. C., P. A. Williams, and S. R. Halloy. 1999. A weed risk assessment model for use as a biosecurity tool evaluating plant introductions. Journal of Environmental Management 57:239-251.