Traditionally, biologists access the identity of an organism through optical cues and identification rules. While arguably the traditional identification method is working well and forms the basis of almost all current biodiversity and biomonitoring campaigns, we are reaching the end of the road for a number of reasons.
Traditional taxonomy is expert work and comes at a relatively high price. There are fewer expert taxonomists and training of future experts is in decline. Expert identification of microscopic organisms is slow and the time lag between sampling and obtaining results is long. And in most countries funding for biomonitoring campaigns is in decline.
Computers have already won the first two rounds against humans
Fortunately, we are witnessing breakthroughs in two technical domains that can solve the aforementioned problems - genetic methods and computer vision/machine learning. Take the often used example depicting the progress of machine learning. Computers outperformed humans for decades in chess but now also do so in the Japanese strategy game Go. Only a decade ago it was still thought impossible for computers to ever match humans in Go but the advent of deep learning in 2012 and its subsequent developments changed this paradigm. In 2016, computers gained the first ever unaided win over a human Go grandmaster.
At first glance one might fail to see how Go ties to biological taxa identification, but there is a parallel. While there have been calls for the use of computer vision in taxa identification (MacLeod et al. 2010) most traditional biologists viewed this task too complex for computers. Thus its pursuit was thought to lead to unreliable results and was deemed downright dangerous for bioassessment.
However, times have changed. In the project DETECT we show in a comparative study that for e.g. freshwater macroinvertebrate taxa, deep learning based identification already equals the average accuracy of human experts (Ärje et al. manuscript). Developments in DETECT and a recently funded project AID (Danish Villum foundation, starting 2018), will extend these techniques to both the marine as well as terrestrial environments.
Machine vision will soon provide identification matching that of human experts at a negligible cost and in almost real-time. Numerous other possibilities for the use of computer vision in taxa identification exist, such as the use of satellite and drone images for wildlife species monitoring, plant or even algae species identification.
Will genetic methods deal the final blow?
The good news for taxa identification does not stop there. While genetic approaches to taxa identification have often been seen as more reliable, they have mostly been dismissed for being too costly in the past. In reality, genetic methods have now become more and more viable for routine use. Much of this development is due to the immense decline in sequencing costs. Considering e.g. aquatic monitoring, particularly new developments in the field of metabarcoding show promise.
SYKE has recently taken part in the first ever comparative test of traditional vs. genetic identification for field samples gathered for routine aquatic biomonitoring (press release). A main finding was that the DNA-based assessment protocol detected twice the amount of species at equal costs when compared to the current routine biomonitoring method. A joint, in-depth, Scandinavian follow-up effort is planned for 2018.
Another important field for genetic techniques is environmental or eDNA. Again, while this approach is not yet mature, it is already successfully being used to identify whether either an endangered or invasive species is present within an area. More generally, genetic methods using both DNA and RNA will provide more information on biodiversity and hidden biodiversity. The Academy of Finland recently awarded funding to SYKE’s researchers for two distinct projects which will help to explore the lesser known genetic aspects of aquatic biodiversity of freshwaters and the Baltic.
To integrate the additional wealth of taxa information that is unleashed through the use of genetic methods, a lot of standardization and basic research is needed. Standardization is not only needed with respect to the methodology used to collect, process and analyze genetic data but also on how to assure accuracy and to improve current reference databases. To this end, SYKE is actively involved in the large European COST action DNAqua-Net which has already attracted significant global attention from scientists and stream managers of 43 countries. DNAqua-Net gathers researchers across disciplines, to identify gold-standard genomic tools and to develop novel eco-genomic indices and metrics for routine application for biodiversity assessments and biomonitoring of European water bodies. Furthermore, DNAqua-Net will provide a platform for training of the next generation of European researchers, preparing them for the new genetic technologies.
Research is still needed before novel identification methods become routine tools of environmental data acquisition. But it is certain that in the near future human experts will become almost obsolete in taxa identification. However, the same experts can look forward to shifting their focus from mere data production to the ecological interpretation of ever more reliable and extensive data.
Senior researcher Kristian Meissner is head of Jyväskylä office and the manager of the environmental data acquisition and usage research program at SYKE. He is a member of the core group of MONITOR2020, the EurAqua network, a research coordinator of the PEER network and management committee member of the DNAqua-Net. Meissner currently serves as an external member on the faculty council of the Faculty of Mathematics and Science of the University of Jyväskylä.
Tel. +358 295 251 422
firstname.lastname@ymparisto.fi
Opinions of blog contributors do not necessarily reflect the official views and opinions of the Finnish Environment Institute.