Search for a command to run...
Abstract Environmental Decision Support systems use models to make tailored predictions about local circumstances based on user input. Citizen Science systems accept contributions about local circumstances to improve models. Each system type emphasises data flow in one direction: system to user or user to system, respectively. We introduce a new system type that synergises data flows in both directions. Reciprocal Environmental Decision Support (REDS) systems process user-contributed information to (1) provide tailored predictions and (2) improve the model that generates the advice. We assessed existing decision support systems and found no REDS system for ecosystem management. For our proof-of-concept REDS system, Garden Advice, we began with an initial Bayesian species-habitat association model (for the House Sparrow Passer domesticus ) and solicited data from UK residents about habitat structure and sparrow observations in domestic gardens. The model made predictions about sparrow presence, and effects of planned garden changes, while updating parameter estimates based on contributed data, generally becoming more confident. One salient parameter update likely reflects reality (a positive association with grass), while another (a reduced association with roof proximity) likely reflects observation bias. Cross-validated comparison indicated that the final model predicted the solicited data better than the original model. Thus, untrained observers can provide data of sufficient quality to refine a model already parameterised with trained observer data. Notwithstanding important questions about distinguishing observation bias from ecologically meaningful information, our system demonstrates that important synergies can be obtained from the REDS approach. Later users of the system obtained better advice thanks to contributions from earlier users and automated Bayesian updating. Lay Summary A summary for non-specialists is available at https://bit.ly/reds_brochure .