Informed-decision making within the context of geohazards for immediate action, policy making, or designing adaptation and mitigation measures, requires a robust model-based risk assessment. Depending on the decision’s context and scope, a model-based assessment can either focus on short-term prediction or it is future-oriented, which impacts on the way simulation-based prediction and their inherent uncertainties need to be managed. This calls for a dynamic framework that allows to collect, project and enrich decision-relevant data by means of predictive computational models, conceptualised as a geohazard digital twin (GDT). The vision of a GDT as flexible and extensible decision-support tool has attracted great interest in recent years and is seen as an opportunity for significant improvements in disaster management [1]. A GDT relies on bidirectional data flow and facilitates dynamic geohazard risk assessment that is capable to react to changes in the physical counterpart, e.g. regarding slope or precipitation pattern. In this contribution, we focus on a specific challenge associated with GDTs, namely how to identify the most informative data given a specific decision-relevant prediction task. First, we formalize the problem setting and describe the software ecosystem that our study is based upon including computational models, a surrogate-based acceleration strategy and automated Bayesian calibration routines. This ecosystem is utilized to demonstrate how to investigate the information content of multiple sensor selection and placement strategies aiming for a decrease of uncertainty, thereby an increase of predictive reliability that yields better-informed decisions [2]. We finally demonstrate our approach by modelling runout of gravity-driven synthetic mass-wasting events in synthetic and real-world topographies. Our findings show how observation type, resolution and size control the efficiency and quality of runout model calibration. We conclude that model-based optimal design of sensor network for geohazard DTs can improve their predictive quality and efficiency. [1] Ford, D.N., Wolf, C.M., 2020. doi:10.1061/(ASCE)ME.1943-5479.0000779. [2] Yildiz, A., Zhao, H. and Kowalski, J., 2023. doi: 10.3389/feart.2022.1032438