Radar rainfall forecasting for sewer flood modelling to support decision-making in sewer network operations
Date: 17 November 2017
University of Exeter
EngD Water Engineering
Radar quantitative precipitation estimates (QPEs) and forecasts (QPFs) are useful in urban hydrology because they can provide real time or forecasted rainfall information for flood forecasting/warning systems. Sewer flooding is a disruptive problem in England and Wales. Wastewater companies have reported that more than 4,700 customers ...
Radar quantitative precipitation estimates (QPEs) and forecasts (QPFs) are useful in urban hydrology because they can provide real time or forecasted rainfall information for flood forecasting/warning systems. Sewer flooding is a disruptive problem in England and Wales. Wastewater companies have reported that more than 4,700 customers are at risk of internal sewer flooding. Currently in the UK, mitigating sewer flooding before it occurs is difficult to achieve operationally because of the lack of accurate and specific data. As radar rainfall data is available from the UK Met Office, particularly radar QPFs with a maximum lead time of 6 hours, these datasets could be used to predict sewer flooding up to this maximum lead time. This research investigates the uses of radar Quantitative Precipitation Forecasts and Quantitative Precipitation Estimates to support short term decisions of sewer network operation in reducing the risk of sewer flooding. It is achieved by increasing the accuracy of deterministic radar quantitative precipitation forecasts, developing on probabilistic radar quantitative precipitation forecasts, and using spatial variability of radar quantitative precipitation estimates to estimate flood extents in sewer catchments from the North East of England. Radar rainfall data used in the case study is also sourced from this region of size 184 km x 140 km. The temporal and spatial resolutions of rainfall forecasts are important to producing accurate hydrological output. Hence, increasing these resolutions is identified to improving deterministic radar quantitative precipitation forecasts for hydrological applications. An interpolation method involving temporal interpolation by optical flow and spatial interpolation by Universal Kriging is proposed to increase the resolution of radar QPF from a native resolution of 15 mins and 2-km to 5 mins and 1-km. Key results are that the interpolation method proposed outperforms traditional interpolation approaches including simple linear temporal interpolation and spatial interpolation by inverse distance weighting. Probabilistic radar quantitative precipitation forecasts provide information of the uncertainty of the radar deterministic forecasts. However, probabilistic approaches have limitations in that they may not accurately depict the uncertainty range for different rainfall types. Hence, postprocessing probabilistic quantitative precipitation forecasts are required. A Bayesian postprocessing approach is introduced to postprocess probability distributions produced from an existing stochastic method using the latest radar QPE. Furthermore, non-normal distributions in the stochastic model are developed using gamma based generalised linear models. Key successes of this approach are that the postprocessed probabilistic QPFs are more accurate than the pre-processed QPFs in both cool and warm seasons of a year. Furthermore, the postprocessed QPFs of all the verification events better correlate with their QPE, thus improving the temporal structure. Spatial variability of radar QPE/QPF data influences flood dynamics in a sewer catchment. Moreover, combination of different percentiles of probabilistic QPFs, per radar grid, over a sewer catchment would produce different spatial distributions of rainfall over the area. Furthermore, simulating many probabilistic QPFs concurrently is computationally demanding. Therefore, generalised linear models have been used to estimate model flood variables using a spatial analysis of radar QPE. Spatial analysis involves using indexes representing specific information of the spatial distribution of rainfall. The novelty of this estimation method includes faster estimations of flood extents. The main points of success of this approach are that more detailed spatial analysis of large sewer catchments produce more accurate flood estimations that could be used without running hydraulic simulations. This makes the approach suitable for probabilistic sewer flood forecasting in real-time applications. A business case is proposed to use the outputs of this research for commercial applications. Probabilistic sewer flood forecasting is evaluated and recommended for industry application using a financial appraisal approach for Northumbrian Water Limited. The business case shows that the methods could be adopted by the wastewater company to mitigate sewer flooding before it occurs. This would support decision making and save costs with better intervention management.
Item views 0
Full item downloads 0