As a young research group at TU Ilmenau, the Data-intensive Systems and Visualization Group (dAI.SY), with currently more than 20 scientists and many student collaborators led by Prof. Patrick Mäder, combines diverse expertise in computer science, engineering and related scientific fields. In addition to machine learning and reliable software, biodiversity informatics has emerged as a focus of the department in recent years. In this way, the research team aims to contribute to the preservation of biodiversity. UNIonline spoke with Prof. Mäder about research in this area.
With your research focus on biodiversity informatics, you are working at the interface of computer science, biology and big data, thus bringing together the topics of digitization and sustainability. What motivated you to conduct research in this area and specifically on the topic of biodiversity?
In 2006 and 2012, I participated in expeditions to Siberia lasting several weeks under the leadership of Prof. Christian Wirth, the founding director of the German Center for Integrative Biodiversity Research iDiv, and Prof. Ernst-Detlef Schulze, founding director of the Max Planck Institute for Biogeochemistry. Together we conducted ecological research under adventurous conditions that was very memorable and changed my awareness of biodiversity: Climate change is a major threat to humanity, and biodiversity loss goes hand in hand with it. We need to stop thinking in silos and bring together the expertise of different scientific fields to study these complex relationships, understand them, and develop foundations for solutions. Our Flora Incognita project is doing just that.
Until recently, biologists did not have access to very large amounts of data to analyze. However, this has changed in recent decades, allowing researchers like you to use these data to investigate ecological questions. What kind of data are you working with?
We work with very different types of data, primarily photos, but also location data, multispectral image data that capture more than the three color channels perceivable by humans and thus information not visible to the human eye, and point clouds , which arecollections of very many measurement points generatedby laser scanners. Probably best known are the observation data that we have been using for years for automatic plant identification in the Flora-Incognita app: Millions of users worldwide ensure every day that we can link image evidence of plants with their locations, enabling us to analyze and predict the distribution of species.
This information is supplemented by curated observations from selected experts via the Flora Capture app. But our research group is not limited to plants. We process microscopic image data for the detection of phytoplankton and insects. In addition, we use special multispectral image data for the automatic identification of pollen. And for evaluating forest stands, we use point clouds generated with LIDAR sensors, a method related to the Radar related method for optical distance and speed measurementWith such data, we can then, for example, make statements about the water quality or help make urban areas more bee-friendly.
As part of the interdisciplinary research group KI4Biodiv - Artificial Intelligence in Biodiversity Research, you would like to work with the Max Planck Institute for Biogeochemistry to further develop and improve these AI methods and technologies in order to monitor biodiversity in different habitats and landscapes efficiently, quickly and automatically. To what extent is biodiversity monitoring a particular challenge for you as a researcher?
To understand the challenges, we first need to look at what biodiversity monitoring means in the first place: it allows us to perceive and document changes in the spatiotemporal occurrence of species. This includes, of course, the distribution of species: Where is diversity declining, where is it increasing? But biodiversity monitoring also includes other things, such as phenology: when do plants bloom, when do they bear fruit, and when do autumn leaves turn colorful? Such monitoring data indicate changes in biodiversity, they are used to investigate the causes of these changes, and they indicate whether strategies and measures to protect biodiversity are working.
Of course, such monitoring also poses major challenges, especially in three areas: it is expensive, requires a lot of time, and requires excellent taxonomic knowledge. Thus, numerous methods and concepts are needed to conduct effective biodiversity monitoring and to overcome the above-mentioned challenges - which is why automated recording and evaluation methods are the focus of research.
How important is interdisciplinary collaboration between scientists from different disciplines in this context?
It is very exciting to see what sources of monitoring data are already available and how they will be shaped in the future. To realize the full potential, an interdisciplinary research approach is of great importance here: biologists, for example, usually collect rather small-scale field data on selected species, with manageable data volumes. These in-situ recordings can, in turn, calibrate and validate the large data sets generated in remote sensing via aerial and satellite data. Computer science now holds a large repertoire of machine learning and artificial intelligence methods that can be used to ingest and analyze such datasets. Thus, the application and advancement of AI methods can and will create key momentum for further development in environmental and biodiversity research.
What benefits do you hope biodiversity monitoring will have for its conservation? Everything we research ultimately serves one major goal: to preserve global biodiversity. Monitoring comes first, because only if we know how diversity is doing - locally, regionally, in a temporal context - can we conduct causal research. Based on these results, measures can finally be developed that can prevent or at least reduce the loss of biodiversity. Biodiversity monitoring is thus a prerequisite for political decisions and ultimately also serves to evaluate the success of such measures.
Prof. Dr. Patrick Mäder
Head of Data-intensive Systems and Visualization Group