UKWIR – RTM – Case Studies

Three case studies for RAPIDS1 have so far been conducted under the UKWIR-funded joint-industry RTM project, in two-stages to evaluate effectiveness in different sized catchments under different conditions; stage 1 used design rainfall and stage 2 used real rainfall. Overall RTM project coordination was performed by HR Wallingford.

Case Study Cities

Dorchester: small urban catchment (Partner: RAA)

-          Evaluation of the significance of use of soil moisture as ANN input

Portsmouth: medium urban catchment (Partner: Mouchel)

-          Island location; tidal effects; need for pumping; evaluation of effectiveness of ANN models to provide early starting of pumps – as a flood-mitigation strategy

Crossness (South London): large urban catchment (Partner: Halcrow)

-          Evaluation of model effectiveness using spatially varying rainfall as ANN inputs.

HydroMAT – Hydrographic Model Analysis Tool

In order to allow all partners to present results consistently, the HydroMAT model analysis tool was developed to provide automated assessment of ANN output using a number of standard metrics:

-          Nash-Sutcliffe Efficiency Coefficient (NSEC)

-          RMSE-Observations Standard Deviation Ratio (RSR) 

-          Percentage Bias (PBIAS)

-          Total Volume Error (TVE)

-          ANN Normalised Root Mean Square Deviation (NRMSD)

-          % Samples in Limits - All Nodes

-          Amplitude Error of Hydrograph Peak

-          Timing Error of Hydrograph Peak

-          R-Squared - All Nodes

-          Pearson Correlation Coefficient - All Nodes

-          ANN Output vs Target X-Y Plot (ATXY) - Single Node

-          ANN Output & Target Hydrographs - Single Node

-          Confusion Matrix for Peak Flood Depth Categories

-          Confusion Matrix for Flood Positives & Negatives

-          Confusion Matrix Accuracy Band summary analysis

UKWIR – RTM – Results and Discussion

In summary, results for UKWIR case studies demonstrated that:

  • (Dorchester): Use of soil moisture levels (NAPI) as ANN input demonstrated a small improvement in model performance, but this was probably not sufficient to offset additional costs of data gathering, preparation and application to ANN model.
  • (Portsmouth): Use of ANN models were demonstrated successfully to prevent flooding in the 'Morass' area of Portsmouth, when used as a trigger for early initiation of pumping at the Eastney pumping station.
  • (Crossness): Although modelling the entire 220km2 catchment using 23 raingauges was not possible, spatial rainfall as ANN-input worked will when applied to smaller areas (4-5 raingauges subcatchments). Further work is needed with spatially varying rainfall and large catchments.

Work is ongoing to develop RAPIDS1 to optimise using MOEA methods and allow a universal set of inputs to be definable at run-time, via input data files (MS Excel workbooks).


Work is at too early a stage to present results; the methodology is still under development.