iForest

Runtime: 11/2023 - 10/2024
Team: Dr. Daniel Doktor (PI), Dr. Maximilian Lange

Despite the increasing threats to essential forest ecosystem services posed by the biodiversity and climate crises, robust monitoring of biodiversity has been lacking. This is due to limited accessibility, heterogeneous formats of valuable data and observations, and the absence of suitable analysis methods (along with the necessary hardware). The goal of this project was therefore to integrate the available repertoire of geo- and citizen science data (e.g., drone data, Flora Incognita, iNaturalist) to utilise these previously untapped biodiversity information sources. By combining them with remote sensing products and applying artificial intelligence (AI) methods, the project aimed to derive forest communities and, ultimately, biodiversity at a landscape scale. This approach leveraged standard development environments for AI and data science, enabling flexible integration of extremely large datasets with predictors of different data structures while handling incomplete data with minimal software development effort.

The project results provided, for the first time, an assessment of the added value of citizen science data for forest biodiversity research. Including these data improved the classification accuracy of tree species - and, consequently, forest communities - by 10%. However, due to the project's short duration, the functional relationships underlying the increased classification accuracy could not be conclusively determined. Additionally, the overall classification process (independent of citizen science data) was improved in terms of accuracy (+20%) and processing time.

A new AI workflow was developed to integrate diverse observational components from citizen science, environmental, and remote sensing data. This workflow can serve as a blueprint for future studies with different thematic focuses, as it emphasizes the integration of heterogeneous data sources.