CRC 1404/2: Efficient DAW Execution Using Incremental Data for Forest Disturbances (SP B07)

In this subproject, we address the specific but crucial problem of detecting and evaluating forest mortality using Machine-Learning-heavy DAWs to complement and accelerate expert-based analysis of satellite images. DAWs implementing such a monitoring must analyze a steady stream of high-volume data from different remote sites, which demands frequent model updates to adapt to the new, most recent data. Stretching the update intervals as well as optimizing the training process may reduce costs and allow an energy-efficient analysis. To this end, we will research new methods a) for energy- and cost-efficient deployment of DAWs to extract indica-tors for forest mortality from diverse remotely sensed and climate streaming data and b) for effi-cient model updates using incremental data incoming in different sites at different times over multiple sites.