Welcome to the UK Reconstructed Flow Data Explorer! This app allows you to explore the reconstructed flow data produced for the Historic Droughts Project. The flow simulations are available for 303 UK catchments from 1891 to 2015, and were produced using the GR4J hydrological model. A separate Hydrological Drought Explorer app has been produced for the exploration of the drought data that has been derived from this flow data. Please see the About the Data tab for more information on how the data were derived.
Catchment ID numbers used in this app relate to those used by the National River Flow Archive.
This page lets you explore the model performance in different catchments to help guide you in selecting an appropriate reconstruction for your study. The evaluation metrics are described in more detail in the 'About the Data' tab.
The calibration period (1982-2014) is displayed on the x axis, the Evaluation Period (remaining years on observational record) is displayed on the y axis. The top left corner of the plot represents the optimum metric value.
In this radial plot, the centre represents poor model performance, with the outside of the plot representing optimum scores. The dark red, bright red and yellow lines represent the highest, medium and lowest evaluation metric thresholds respectively (see 'About the Data' ). Catchments are ordered geographically clockwise around the country, and grouped by region.
This map represents the 'Containment Ratio': The percentage of days that the observed flows fall within the upper and lower model uncertainty bounds (see 'About the Data' for information on how these bounds were quantified).
Please see the 'About the Data' tab for details on the columns in this table.
This page lets you view the timeseries data interactively. Select a catchment using the drop-down below.
Please be patient, this graph takes a while to load as it processes 125 years of daily data.
This graph displays both the single best model run (turquoise line), and the range of the 500 ensemble members (shaded turquoise). The flow observations are also shown in orange. See the 'About the Data' tab for more information on the modelling methods.
The reconstructed flow data are provided on a daily timestep from 1891-01-01 to 2015-11-30. The data are provided for 303 UK catchments, though two series are provided for the Thames at Kingston (39001) and the Lea at Feildes Weir (38001) as the model has been calibrated to both observed and naturalised flows in these catchments. For the naturalised flow calibrations, the station ID is appended with an additional '0'. The data are provided in two formats: a single best model run; and an ensemble of 500 model runs. For the top single run, the upper and lower daily limits of the 500 ensemble members are also provided. Please see the Model Calibration details below for more information on how this ensemble was created.
The data are available for download from the Environmental Information Data Centre (EIDC) here. Registration is required to access the data.
The inputs required to the GR4J model are daily precipitation and potential-evapotranspiration (PET). The daily rainfall data were provided by the UK Met Office who combined newly rescued and digisted data over the period 1860-1961, with the UK GEAR dataset post 1961. The temperature-derived daily PET data was provided by Tanguy et al (2017).
model has been calibrated over the 32 year period 1982-2014.
The model was run over this calibration period using 500,000 Latin Hypercube Sampled model parameter sets.
These model parameters were assessed against observations using a multi-objective approach comprising of six evaluation metrics:
- Nash Sutcliffe Efficiency (NSE), optimum value = 1,
- NSE on log flows (log NSE), optimum value = 1,
- Mean Absolute Percent Error (MAPE), optimum value = 0,
- Absolute Percent Bias (PBIAS), optimum value = 0,
- Absolute Percent Error in Mean Annual Minimum flows over a 30 day accumulation period (MAM30), optimum value = 0, and
- Absolute Percent Error in the flow exceeded 95% of the time (Q95), optimum value = 0.
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 trade-offs between metrics, the runs were re-ordered according to limits of acceptability.
The thresholds were set to:
1) NSE > 0.5, log NSE > 0.5, MAPE < 50, Absolute PBIAS <10 , MAM30 APE < 50, and Q95 APE < 50,
2) NSE > 0.4, log NSE > 0.4, MAPE < 75, Absolute PBIAS < 15, MAM30 APE < 75, and Q95 APE < 75,
3) NSE > 0.3, log NSE > 0.3, MAPE < 100, Absolute PBIAS < 20, MAM30 APE < 100, and Q95 APE < 100,
4) NSE < 0.3, log NSE < 0.3, MAPE > 100 , Absolute PBIAS > 20 , MAM30 APE > 100, and Q95 APE > 100.
The top single run, and an ensemble of the top 500 runs are provided in this flow dataset. For the top single run, the upper and lower daily limits of the 500 enemble members are also provided.
The six evaluation metrics were also calculated over the remaining observational period for each catchment (prior to 1982).
The results of these calculations are shown both in the scatter plot and the data table in the
'Model Performance and Parameters'
Two further calculations were performed over the entire period of record (the calibration and the validation years, up to Nov 2015) to calculate the:
- Uncertainty Width - the range of the minimum and maximum daily flow values of the 500 ensemble members, relative to the midpoint, and the
- Containment Ratio - the percentage of days for which the observed flow values sit within the range of the minimum and maximum flow values of the 500 ensemble members.
This research was funded by the RCUK Drought and Water Scarcity Programme, via the Historic Droughts (NE/L01016X/1) and IMPETUS (NE/L010267/1) projects.
The contents of this site are subject to CC-BY copyright: © NERC (CEH) 2020 Crown Copyright.