Statistical modelling of intermittence metrics in temporary rivers of the UK

Lack of monitoring of the IRES network limits the potential to develop current understanding of these critical landscape features. Simulation offers potential to improve the spatiotemporal extent and resolution of the available evidence, enabling further research to be performed and understanding developed. Previous research has demonstrated the potential for statistical models to accurately reconstruct the spatiotemporal dynamics of intermittent rivers.

In this study ordinal regression models were trained to simulate the hydrological regime for the first time, enabling the simulation of flowing, ponded and dry dynamics using localised environmental variables, site characteristics and seasonality. This was enabled by a dataset covering the full range of hydrological conditions of ten chalk streams in the Chiltern Hills, UK over the last 23 years.

The hydrological regime was accurately simulated using ordinal regression models, with weighted f-scores ranging from 0.759-0.955, and scores exceeding 0.935 in the six most westerly streams. This apparent west-east pattern was also present in other evaluation metrics, with probability of detection scores ranging from 0.954 to 0.973 in the westerly streams, and 0.775 to 0.908 in the east, and Correct Classification Rate ranging from 0.935 to 0.955 in the west, and 0.849 to 0.909 in the east. The apparent relationship between model performance and BFI suggests groundwater influence on the hydrological regime    

Further research provided further insights into controls on model performance, including groundwater influence, prominence of ponding and the relatively rapid response of the easterly streams. In addition to improved understanding of controls on model performance, the accurate reconstruction of hydrological regime in these rivers facilitates research from linking associated data to previously unavailable hydrological state data, to investigating climate change impacts and influence of abstraction pressures on these invaluable ecosystems.