Until the end of the last century, if a biosecurity official intercepted an organism which could possibly be a quarantine risk, he or she had two options for identifying it: sending the specimen to a relevant taxonomic specialist or using published images or dichotomous keys to make an identification on their own. The explosion of digital hardware and software technology during the past 50 years or so has potentially transformed the situation. A range of digital identification tools have been developed which, when published online and as mobile apps, provides the potential for biosecurity officials and others to identify potentially invasive pests themselves. At the same time, rapid biotechnology developments have resulted in a range of laboratory and field-based molecular diagnostic techniques, providing additional tools for pest identification. The development of AI software, focussed on image recognition, presents new opportunities for on-the-spot identification.

Clearly, these developments provide new opportunities to support pest identification by quarantine staff and other plant protection decision makers. However, how accurate, appropriate, resilient, and sustainable future digital identification tools will be depends on the context in which they are developed and funded, and the way in which they are implemented. These issues are addressed in this final section of the review, under the following headings:

• Are specific DITs fit for purpose given a range of users and situations? 

• Integration of DIT’s with alternative diagnostic tools and other support systems 

• Strategies for efficient and sustainable DIT development and deployment 

• The future of digital pest identification 

4.1. Are specific DITs fit for purpose given a range of users and situations?
Digital identification tools developed for biosecurity purposes aim to provide support to a range of users – from entomologists, plant pathologists, and weed scientists, to experienced quarantine officers, testing laboratories, farmers, crop advisors, Landcare groups, and citizen scientists. 

The speed and specificity required of identification tools will depend on many factors, including the speed and precision required for operational purposes and the appropriate level of detail required. For example, identification of a Trogoderma species detected, in a warehouse containing imported and locally sourced goods, would probably need a definitive answer within hours, while the identification of the contents of a fruit fly trap from a remote, little-visited, offshore island might not require such urgency.

In both of these examples, a digital tool would need to support challenging, identification decisions, such as providing diagnostic images of dissected structures, in the case of Trogoderma, or depicting variation in colour markings in the case of fruit flies. Less precise identification may suffice in other situations. For example, if a surveillance officer detects a leaf-mining caterpillar in an Australian tomato crop, an initial identification to genus-level only (which would reveal whether the caterpillar was the endemic Phthorimaea operculella or the exotic Phthorimaea (formerlyTuta absoluta) might be sufficient for initial, decision making. 

In other cases, for instance where a digital key has been developed to identify snails, a relative novice may be able to perform an identification quickly and with confidence. For some situations, such as a digital tool developed for identifying certain insect species that requires the dissection of genitalia to make a positive identification, practical training will probably be required before the user could be regarded as proficient.

Where DITs are to be used for pest management in the field, they might need to include non-pest species, including beneficial insects (natural predators of pests and pollinators) and neutral species that pose no threat but may provide additional food for beneficial species. By contrast, tools developed for use at quarantine stations may require a very specific focus – that is, what pests may follow a particular pathway? This may include species that would not naturally be found in a particular commodity but may wind up there due to the unique circumstances that occur with worldwide shipping. Training tools may have yet another focus, as these tools may need to include more basic taxonomic or quarantine information to ensure they are usable for a less experienced audience.

Having the flexibility to deploy DITs online, as well as via mobile apps, can be critical in meeting the requirements of different users in different situations. Filterable image libraries and mobile apps, for example, may be useful for the officer undertaking field surveillance and needing to categorise a field observation, while online keys and molecular protocols and voucher sequences may be more appropriate for the laboratory technician who receives the surveillance officer’s samples. DITs that include comprehensive information about taxa, such as detailed fact sheets and images, offer opportunities for online learning, particularly in association with distance training courses for improving the identification skills of inspection officers and technical specialists. 

 

4.2. Integration of DIT’s with alternative diagnostic tools and other support systems

While a digital identification tool can provide an early and accurate result, using a different DIT device or an alternative identification technology, such as a molecular tool, may be necessary in some situations to provide confirmation of the result. Indeed, where feasible, combining different identification tools on the same platform is recommended. Earlier in this review various thrips DITs were described; in some cases, the thrips morphological identification key is supported with sections providing relevant molecular analysis techniques for the taxa involved.

A more recent example shows how a digital identification and diagnostic tool for Phytophthora (IDphy) brings together a series of different types of identification tool on the same platform. Developed by a large team of scientists over a period of 8 years, IDphy is one of the most comprehensive tools ITP has ever published, covering all the valid, described species within the genus (as of 2022).  

Released in 2019/20 as an online resource and subsequently as a mobile app, IDphy includes the following range of identification and support tools:  

  • Detailed protocols for molecular identification, including: sequence vouchers for ITS rDNA and COI, and protocols for SOPs for DNA extraction, PCR, electrophoresis gels, and sequence-based identification
  • Voucher sequences from the types, providing the most reliable source of reference material for the tool
  • Filterable image gallery
  • Morphological interactive Lucid key
  • Searchable fact sheets
  • Tabular key for quick reference
  • Background and life cycle/biology information

 

As well as combining DITs with other identification techniques, there are other opportunities to combine DITs with reporting, surveillance, and forecasting platforms. For instance, diagnostic wheat apps, both Android and Apple, developed by the Western Australia Department of Agriculture, includes 86 insect pests, diseases, and other disorders. In 2016, when a new invasive pest, the Russian wheat aphid, was first detected in Western Australia, the diagnostic apps were rapidly updated to include details and images of the Russian wheat aphid in the key to help distinguish it from other aphid pests. If a farmer or consultant thinks they have detected and identified this introduced pest, they can report the details to authorities by clicking on the “Pestfax” icon at the top-right corner of the relevant fact sheet – as shown below.

 

A further example of how extra functions can be combined with ID tools is provided by the development of international “harmonised databases”, such as that for cassava, a tool being developed by the International Centre for Tropical Agriculture CIAT. The aim is to create a coalition of platforms to make sure that wherever new incursions are detected, they are reported, and the data gets to the right people to provide a better overall picture for decision making. Therefore, the line between technical assistance with pest identification, diagnostics, remote diagnosis, and active monitoring has become blurred. The barrier between assisting with diagnostics and leveraging that information into a surveillance and monitoring platform has become more a matter of policy and politics than technology. 

4.3. Strategies for efficient and sustainable DIT development and use

Several examples of DIT development and deployment, described earlier in this publication, indicate there are various ways in which efficient and sustainable development and use of DITs can be achieved. Some of the strategies for achieving this outcome are summarised below.

Strategic development processes. In section 3 of this review, we discussed the processes developed by two government organisations for developing and implementing DIT tools: USDA/ITP and EPPO/Q-Bank. The structured and comprehensive approach adopted by both organisations provides important lessons on how to design strategies that deliver efficient, sustainable, and effective digital identification support programs.

Collaboration and sharing data and DITs. Two other case studies, described in section 3, have shown how international collaboration and the sharing of information, data, and DITs can also achieve more efficient and efficient DIT development. The collaborative projects that thrips taxonomists have engaged in for over 20 years has resulted in the development and sharing of a database of thrips identification data. This has enabled new thrips keys to be rapidly developed for use in specific regions, by combining thrips key data contained in the shared database with new information resulting from collaboration with local entomologists.

Developing platforms for both sharing and delivering DITs
. The ISMA platform, described in section 3, provides a good example of the range of support that can be provided on various aspects associated with developing and using DITs for weed seed identification. Similarly, the invasive species image database project Bugwood, mentioned in section 2.5, involves another example of collaborative data sharing, across many states in the USA, as well as internationally, providing a valuable international source of images relevant for pest identification and crop diagnostics. Uuse of web services increases the ability to share data and allow third parties to bring together and use data in new and innovative ways.

Utilising generic platforms and standards. Apart from collaborating on DIT development and the sharing of data, a more efficient development and deployment strategy for DITs can be achieved by utilising generic software platforms, rather than developing one-off, bespoke software. The development of three, generic, matrix key software platforms, described in section 2.4, has enabled many hundreds of taxonomists and other specialists to develop and deploy matrix keys on the internet and as mobile apps. Since constant upgrades and updates are required for developing and deploying DITs, especially when deployed as mobile apps, generic platforms provide a much more cost-effective way of achieving this.

Nevertheless, there are still important issues that must be addressed to sustain the viability of DIT software platforms. In the 1980s and 1990s, most land grant universities in the USA received funding to develop digital “expert systems” that provided decision support to a range of plant protection activities. In many cases, once the funding grants came to an end, and details of the expert system were published, there was no appropriate funding mechanism to support continued development, support, and practical application of the decision support tool. This has also been an issue for some of the DITs mentioned in this review. Some have survived through long-term funding from biosecurity agencies. The group responsible for the Lucid software products, who have been developing and supporting DITs for over 25 years, was initially supported as part of a Federal government Cooperative Research Centre, but subsequently had to develop a strategy involving:

  • A range of funding sources – grants, contracts, software sales, non-Lucid contracts.
  • Flexible R&D activity – initially university based  and subsequently involving a spin-off company.
  • Focus on generic software products – Lucid Builder; Lucid Mobile platform (over 100 apps published).
  • Close collaboration with authors – website, forum, help desk, workshops, ideas for new software features.

4.4. The future of digital pest identification
Before addressing specific opportunities and constraints regarding future development and use of DITs, it would be remiss to ignore the fact that recent development and use of DITs and other ID technologies has occurred at a time when there has been a dramatic reduction in the number of specialist taxonomists worldwide. Few universities now teach taxonomy, and it is not unusual for many developed and developing countries to lack taxonomists, let alone those specialising on major pest taxa. Since the precision of DITs in distinguishing specific pest species can make a critical difference to quarantine risks and the action taken, the importance of up-to-date taxonomic information about major pest species can be critical. Suffice to say that unless this taxonomic crisis is addressed, it could constitute an important constraint to the future development and effectiveness of authentic DITs.

As already discussed in this final section, there are several issues that require more attention to ensure future DIT developments to support plant biosecurity decision making are efficient and effective, including:

– Improving feedback on the use and value of DITs to practitioners. 

– Integrating different DITs with other identification technologies, such as molecular analysis, to provide a range of tools for different identification needs, enabling identification checks using different methods, and linking these identification tools to other surveillance and reporting platforms to provide agencies with comprehensive information systems.
Valkenburg et al 2023 describe a situation where both morphological and molecular techniques are required to distinguish certain Salvinia aquatic weed species.

– Adopting strategic approaches rather than ad hoc developments, encouraging increased collaboration and data sharing in the development and use of DITs, and using generic software platforms where appropriate. 


The continuing development of computer and mobile hardware and software will undoubtedly enable the range of software tools described in this review to be further enhanced as new features become available. The recent success of AI software applications in other disciplines, notably for facial recognition and the diagnosis of skin diseases, such as melanoma, has stimulated interest in the use of AI for pest identification: some examples have already been described in Section 2.6. However, three important factors have been responsible for the success of AI applications to facial recognition and skin diseases: a relatively limited range of features that need to be analysed in a controlled environment, the availability of a large quantity of images to analyse, and data to confirm the respective identifications and diagnoses. 

In the case of plant pest recognition, the equivalent of such “low hanging fruit” is likely to include weed seed recognition and the use of AI to recognise such feature/states as wing venation of certain winged insect pests.  Since the AI application currently being developed in New Zealand for the identification of the brown marmorated stink bug, mentioned earlier in section 2.6, concerns a limited number of New Zealand native species and established relatives, it is also likely to be successful. 

More ambitious applications of AI to the recognition or diagnosis of quarantine pests are not likely to be constrained by the AI technology itself but with accessing the data required for training the technology. Many projects that have embarked on an AI approach have found that acquiring enough images to give anything close to a reasonably high level of recognition or diagnostic success requires a much larger investment than initially envisaged.

Not only does imaging quality need to be improved, as well as the consistency, coverage, and availability of images, but currently, there is no efficient mechanism to authenticate and facilitate the sharing of the large number of images accumulating to train AI-assisted image recognition tools. Images must also include metadata reflecting validated identifications and other associated data to ensure AI training is based on the correct taxa and not on misidentified images. Once again, local expert taxonomic or diagnostic information associated with these images needs to be available during the AI training process.

As stated several times in this review, the future role of DITs is likely to involve a mix of various online and app keys, images, and other digital tools, together with complementary molecular and other lab-based techniques, as well as field identification and diagnostic procedures. The extent to which AI will play an important future role in this context is unclear, including the latest AI development – ChatGPT. 

Judging by the experience gained in the use of AI for medical diagnostics, chess, and other AI applications, the most effective AI results are likely to occur when AI is used in combination with human expertise, which again raises the important issue regarding the future role of taxonomy. Many current practicing taxonomists, particularly those experienced in plant and animal species of biosecurity concern, are either close to or already in semi-retirement. We believe it is critical to arrest this taxonomic decline, train para-taxonomists, or find other ways of maintaining the basic taxonomic skills necessary to authenticate future developments and applications of DITs for plant biosecurity.