Improving radar rainfall estimation for flood risk using Monte Carlo ensemble simulation

Abstract

Weather radar is a crucial tool in rainfall estimation, for flood forecasting and urban drainage design. While rain gauges give sparsely distributed ground observations over time, a weather radar provides a picture of the spatial distribution of rainfall. The Met Office have a C-band weather radar network, which is integral for the identification of high-intensity rainfall events. This covers most of the U.K., with snapshot images every 5 minutes and a pixel resolution of approximately 1km2 . Radar rainfall estimates are subject to many error sources - ground clutter, attenuation, beam blockage, the vertical profile of reflectivity and the drop-size distribution - resulting in complex correction procedures, often relying heavily on ground observations. In June 2012, the city of Newcastle suffered extensive flooding due to a high-intensity rainfall event, during which a month’s worth of rainfall fell in just two hours. This event caused traffic chaos and millions of pounds worth of damage. Radar rainfall images showed areas of unexpectedly low rainfall amounts with ground-based estimates telling a different story, caused by a dampening of the radar signals. The extent of this signal loss, known as attenuation, was due to intervening high-intensity rainfall. This occurred to such an extent that volumes of rainfall were missing, referred to as rainfall ‘shadows’, which cannot be accurately estimated from attenuated reflectivity measurements. The information is lost, with the frequency and extent of rainfall shadows unknown. To investigate the error structure (and resulting shadow effects) in radar rainfall esti mation, a flexible stochastic model for simulating spatio-temporal rainfall event fields is designed and implemented. This model can be used to generate rainfall, and in this case is used to simulate ‘true’ reference rainfall fields. Simulated fields satisfy key aspects of a spatial rainfall field, namely the spatial correlation structure, anisotropy, marginal dis tribution and advection. Standard weather-radar processing methods are inverted, with uncertainties imposed on simulated rainfall fields, through a combination of a stochastic drop-size distribution field, random errors and path-integrated attenuation effects, result ing in an ensemble of radar images. Applying a radar rainfall estimation procedure to simulated rainfall fields allows for the investigation of the resultant error structure be tween simulated and corrected rainfall rates. This provides realistic weather radar images, of which we know the true rainfall field, and the corrected ‘best guess’ rainfall field which would be obtained if they were observed in the real-world case. This flexible and efficient model performs well at recreating existing rainfall events with prescribed properties, both visually and quantitatively, for a large range of event types. A hierarchical clustering scheme is implemented on event properties, to identify poten tial problematic event types for a given region, allowing the improved parametrisation of simulations. Radar estimation errors are heavily dependent on the rainfall intensity, as expected, mostly due to the impact of attenuation effects. Variability in errors is fairly homogenous, except where the radar signal is fully attenuated. Rainfall shadows are de fined as areas of high-intensity simulated rainfall with corrected rates less than 10% of the original rate. In cases where rainfall shadows occurred in one ensemble member, they were likely to occur for all ensemble members. Half of simulated weather radar images have at least 3% of significant rainfall rates shadowed, and 25% had at least 45km2 containing rainfall shadows. This highlights the importance of gaining an improved understanding of rainfall shadows, as this gap would result in underestimation of potential impacts of flooding. This simulation tool has been demonstrated to provide a model framework for investigating the behaviour of errors relating to the radar rainfall estimation process. Applications of this framework are considered in detail, and the hydrological impacts of results, working towards the improvement of radar rainfall estimation.