Prediction validation and iterative model optimization: a tool for improving long-term landscape management.

• Applicant: Dr. Tatiane Micheletti Ribeiro Silva
• Subject Area: Ecology of Land Use ; Ecology and Biodiversity of Animals and Ecosystems, Organismic Interactions
• Term: since 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 529743012

Long-term predictions of ecological systems in complex and changing landscapes can be improved by a better selection of computer-based models for a given context and purpose. Iterative, flexible, and cross-disciplinary workflows could facilitate model selection, validation, and optimization to support sustainable, long-term landscape management, including effective species conservation. Using the endangered woodland caribou (Rangifer tarandus caribou) as an example, the method of "predictive validation" will be further developed and systematically tested for this purpose. This is an "out-of-sample validation" in which the validation data are from a later calendar date than the data set used to fit the models. This method allows a robust minimization of model overfitting, which can improve their predictive ability. The proposed project aims to develop a generic and model-agonistic predictive validation tool for ecological forecasting that is both stand-alone or be iteratively integrated into a forecasting workflow. Implementing the method as an open-source tool will bring an important scientific contribution to forecasting the dynamics of exploited ecological systems in variable environments, thereby improving landscape management in the long term. Furthermore, it facilitates the highly topical establishment of the novel open-source "digital twin" approach to ecology.

• DFG Programme: Research Grants
• International Connection: Canada
• Cooperation Partner: Eliot McIntire, Ph.D.