Adaptive management and optimisation of stream monitoring network design through an integrated monitoring and modelling approach

• Monitoring network spatial design greatly influences model predictive performance. • Greedy Exchange finds optimal network designs to minimise model prediction error. • 25 % of the existing network sites may be removed without increasing model error. • Strategically relocating redundant sites reduces model prediction error by 67 %. • Prioritisation based on model error rapidly reduces RMSPE following a disruption. Stream monitoring networks provide data essential to understand waterway condition and direct management actions. However, methodologies tailored to support adaptive management and optimisation of stream monitoring networks post-establishment are limited. This study presents an approach that integrates monitoring data with a river-based geostatistical model to assess how the inclusion or removal of observation locations (sites) in an existing monitoring network affects model predictive performance and to identify optimal network designs. The approach is tailored to address key adaptive management challenges faced by waterway managers: 1. Eliminate redundancy and reduce the number of sites in a network, 2. Efficiently acquire additional sites to improve model prediction, and 3. Effectively prioritise responses to sampling disruptions. We apply the approach to stream temperature data collected from a large river water quality monitoring network in South-East Australia. An optimal subset of 66 sites from the existing network provided equivalent prediction performance to the existing 89 site network. Strategically replacing redundant sites with new sites acquired using model uncertainty (Kriging variance), was shown to reduce prediction error to less than 40% of that of the original network. Applying the approach to a simulated disruption to sampling demonstrated that preferentially restoring sampling for sites identified as high priority, improved prediction error substantially more than restoring low priority sites. The approach may be adapted for different optimality criteria and applied to commonly measured parameters including water quality, macroinvertebrates or streamflow, providing a practical and pragmatic method to support adaptive management of existing stream monitoring networks.