This dataset is model output from the GR4J lumped catchment hydrology model. It provides 500 model realisations of daily river flow, in cubic metres per second (cumecs, m3/s), for 303 UK catchments for the period between 1891-2015. The modelled catchments are part of the National River Flow Archive (NRFA) (https://nrfa.ceh.ac.uk/) and provide good spatial coverage across the UK. These flow reconstructions were produced as part of the Research Councils UK (RCUK) funded Historic Droughts and IMPETUS projects, to provide consistent modelled daily flow data across the UK from 1891-2015, with estimates of uncertainty. This dataset is an outcome of the Historic Droughts Project (grant number: NE/L01016X/1).
The data are provided in two formats to help the user account for uncertainty:
(1) a 500-member ensemble of daily river flow time series for each catchment, with their corresponding model parameters and evaluation metric scores of model performance.
(2) a single river flow time series (one corresponding to the top run of the 500), with the maximum and minimum daily limits of the 500 ensemble members.
Publication date: 2018-03-12
This dataset is part of the following
The GR4J model (v 1.0.2) was run over the calibration period (1982-2014) using 500,000 Latin Hypercube Sampled model parameter sets. These model parameters were assessed against observations from the National River Flow Archive (NRFA). For two catchments (the Thames at Kingston, and the Lea at Feildes Weir) the model was also calibrated against naturalised flows. The model was calibrated using a multi-objective approach comprising of 6 evaluation metrics: Nash Sutcliffe Efficiency (NSE), NSE on log flows (log NSE), Mean Absolute Percent Error (MAPE), Absolute Percent Bias (PBIAS), Absolute Percent Error in Mean Annual Minimum flows over a 30 day accumulation period (MAM30), and Absolute Percent Error in the flow exceeded 95% of the time (Q95).
The 500,000 model runs were then ranked by each evaluation metric, the ranks were summed, and the runs were reordered according to this final rank. Finally, in order to prevent uneven trade-offs between metrics, the runs were re-ordered according to thresholds of acceptability.
Reconstructed flow timeseries were then run for the top 500 ranking model parameter sets, using PET (Potential Evapotranspiration) (Tanguy et al., 2017: doi https://doi.org/10.5285/17b9c4f7-1c30-4b6f-b2fe-f7780159939c), and reconstructed daily rainfall data, provided by the UK Met Office.
The modelled data, and the supporting metadata files, were exported from the R software programme as comma separated value files (.csv), and ingested into the EIDC in this format.