Machine-supported monitoring of plant communities and habitats

In Switzerland, 48% of habitats are threatened, but they are often poorly monitored because confident identifications of habitat types and related species are time-consuming and require expert knowledge. 

We aim to foster habitat monitoring in Switzerland by developing algorithms that can semi-automatically identify plant communities from videos and link them to habitat types. We recently demonstrated the impact of machine-learning support in biodiversity monitoring by integrating image-based species identification (FlorID) into FlorApp, a leading app to monitor Swiss biodiversity. Here, we propose to make the next step and build the foundation for a service to rapidly monitor plant communities and habitats. 

We envision 

1. vision models trained with the FlorID image database, which can handle numerous frames depicting multiple plant species simultaneously, 
2. a recommender system trained on the best vegetation survey data available to suggest expected but not yet observed species from partial species lists, and 
3. two habitat classifiers: a multimodal one, considering point observations of environmental variables and high-resolution remote-sensing images, and a vision model considering in situ photos. Given the importance of monitoring habitats, the popularity of FlorApp, and the timely nature of this innovative step, we expect the proposed tools to be of great impact.