Transferability of a Bayesian Belief Network Across Diverse Agricultural Catchments Using High-Frequency Hydrochemistry and Land Management Data

Abstract

Biogeochemical catchment models are typically developed for single catchments and, as a result, often generalize poorly beyond this specific context. Therefore, evaluating their transferability is an important step in improving their predictive power and application range. We assess the transferability of a recently developed Bayesian Belief Network (BBN) that simulated monthly stream phosphorus (P) in a poorly drained grassland catchment through application to three further catchments with different hydrological regimes and agricultural land uses. In all catchments, flow and stream water P concentrations were measured sub-hourly from 2009 to present day and supplemented with 400 – 500 soil P test measurements. In addition to the original BBN, five further model structures were implemented to incorporate in a stepwise way: in-stream P removal using expert elicitation, additional groundwater P stores and delivery, and the presence or absence of septic tank treatment, and, in one case, sewage treatment works. Model performance was tested through direct comparison of predicted and observed total reactive P (TRP) concentrations and using percentage bias. The original BBN simulated the observed flow and TRP concentrations well in the poorly and moderately drained catchments, irrespective of the dominant land use (74%≤PBIAS≤81%) but performed less well in the groundwater-dominated catchments. The inclusion of groundwater total dissolved P (TDP), Sewage Treatment Works (STWs) inputs, and in-stream P uptake improved model performance (-4%≤PBIAS≤16%) in the groundwater-dominated catchments. A sensitivity analysis identified redundant parameters in some of the non- catchment specific data used. The original BBN structure could be transferred effectively only between catchments with similar hydrology. For application elsewhere, the BBN structure required modification to include representation of in-stream P removal, groundwater P concentrations, and sewage treatment works inputs. Thus, inclusion of these processes is recommended to accurately determine monthly P concentrations and aid widespread application of the proposed BBN model.

Keywords: hybrid network, expert elicitation, model universality, sensitivity analysis