Linking monitoring and modelling: can long-term datasets be used more effectively as a basis for large-scale prediction?

Abstract

Data from long-term monitoring sites are vital for biogeochemical process understanding, and for model development. Implicitly or explicitly, information provided by both monitoring and modelling must be extrapolated in order to have wider scientific and policy utility. In many cases, large-scale modelling utilises little of the data available from long-term monitoring, instead relying on simplified models and limited, often highly uncertain, data for parameterisation. Here, we propose a new approach whereby outputs from model applications to long-term monitoring sites are upscaled to the wider landscape using a simple statistical method. For the 22 lakes and streams of the UK Acid Waters Monitoring Network (AWMN), standardised concentrations (Z scores) for Acid Neutralising Capacity (ANC), dissolved organic carbon, nitrate and sulphate show high temporal coherence among sites. This coherence permits annual mean solute concentrations at a new site to be predicted by back-transforming Z scores derived from observations or model applications at other sites. The approach requires limited observational data for the new site, such as annual mean estimates from two synoptic surveys. Several illustrative applications of the method suggest that it is effective at predicting long-term ANC change in upland surface waters, and may have wider application. Because it is possible to parameterise and constrain more sophisticated models with data from intensively monitored sites, the extrapolation of model outputs to policy relevant scales using this approach could provide a more robust, and less computationally demanding, alternative to the application of simple generalised models using extrapolated input data.