Use of Nimrod in flood forecasting. Contract report to the Environment Agency and the Met Office

It is common practice within the Environment Agency to calibrate rainfall-runoff models using raingauge data. Operationally these models are used in flood forecasting systems with raingauge data as input up to the time that the forecast is made. Radar rainfall forecasts are used beyond this for lead-times up to 6 hours to provide the potential for earlier flood warnings. There is an acknowledged reluctance to employ weather radar data for rainfall-runoff model calibration at the present time. This stems from concern regarding the consistency of the radar product in time and space, offset by the knowledge that raingauge networks of low density may provide poor spatial estimates of rain especially for spatially-varying convective storms. What is needed is some sound evidence that points to the value of using weather radar data in rainfall-runoff models for both model calibration and forecasting. The “Use of Nimrod in Flood Forecasting” project aimed to establish a body of evidence that will serve to clarify the relative merits of using radar or raingauge data for rainfall-runoff model calibration and forecasting. The study is timely since there have been substantial efforts in recent years by the Met Office to improve the quality of the weather radar product under the Nimrod programme. For the first time, these improved Nimrod products are evaluated systematically in this study for use in flood model calibration and forecasting. A set of four catchments in the Upper Calder part of the Yorkshire Ouse drainage basin were used as the focus of the study. The gauging locations for these catchments feature as operational flood forecast points within the River Flow Forecasting System used for flood forecast construction. The catchments are well served by the weather radar at Hameldon Hill, being at a distance of between 6 and 21 km. Three of the catchments are nested within the fourth, the Calder at Mytholmroyd (near Caldene Bridge) with a natural drainage area of 121 km2, and the smallest being 10 km2. Only one tipping-bucket raingauge is located within the larger catchment although three are located on its outer periphery. The prospect for radar being useful is therefore good. The main Nimrod radar product used was the Nimrod 5 km analysis, available as a rainrate every 15 minutes over the period of the study. This was chosen as being available over the period of the selected events from September 1998 to November 2002. Detailed quality control of the study dataset led to a set of 11 events being identified, of which 6 were used for model calibration and 5 for independent assessment. Three of the assessment events were chosen specifically to encompass spatially-varying convective rainfall. Radar might be expected to outperform raingauges in such situations, when used in rainfall-runoff models to estimate catchment runoff. The form of rainfall-runoff model used was the PDM (Probability Distributed Model), selected as representative of models used in practice and employed operationally for flood forecasting in the Upper Calder by the Environment Agency. It was found that the raingauge-calibrated models provided a basis to set the parameters of the radar-based models, with revisions only needing to be considered for the rainfall factor and time delay parameters affecting flow volumes and timing respectively. This has beneficial implications to operational models that use raingauge data up to the time of the forecast and Nimrod rainfall forecasts beyond this. A major finding of the study was that raingauge-based models provided more consistent flow simulations than those employing radar. This stemmed from the lack of consistency in the Nimrod radar product, with a tendency on occasions to over- and under-estimate rainfall. This was aggravated at times by changes affecting the product, notably the Cyclops radar processing upgrade on 1 September 2000. In contrast, the raingauge network provided a consistent rainfall measurement product, apart from the addition of one gauge over the study period. The radar-based model for the smallest 10 km2 catchment did occasionally perform best, with the nearest raingauge being 2 km away. This leads to the conjecture that for a given gauge density (or proximity) there is some threshold catchment size below which radar may, on occasions, provide superior estimates of catchment average rainfall for input to a rainfall-runoff model. There was also some evidence for flood peak timings to be better predicted, on occasions, using radar-based models although peak magnitudes were usually inferior. Temporal synchronisation of hydrometric networks and possibly quantisation of raingauge estimates are issues to bear in mind here. A sensitivity analysis was performed using the higher resolution Nimrod 2 km analysis available at a 5 minute frequency. Although this provided a minor improvement over the use of the Nimrod 5 km analysis, the overall finding relating to the superiority of the raingauge-based models was unaffected. Model sensitivity to using a Nimrod 5 km rainfall forecast, for a lead-time of one hour, in place of the Nimrod 5 km analysis was also investigated. This demonstrated the consistency of the radar-based models when used with Nimrod analysis or forecast products, and an expected overall fall in performance using the forecast product. The beneficial implications referred to above are reinforced by this finding. A third sensitivity test focussed on the impact of including an additional raingauge on raingauge-based model performance. The successful simulations obtained, without changing the model parameters, served to support the raingauge weighting strategy for catchment average rainfall estimation developed in the study. The report’s summary, conclusions and recommendations provide further details of the knowledge acquired by the study and the implications for Environment Agency and Met Office operational practices.