A New Method for Radar Rainfall Estimation Using Merged Radar and Gauge Derived Fields

Thursday, 18 December 2014
Mohammad Mahadi Hasan1, Ashish Sharma2, Fiona Johnson3, Gregoire Mariethoz1 and Alan Seed4, (1)University of New South Wales, Sydney, NSW, Australia, (2)University of New South Wales, School of Civil and Environmental Engineering, Sydney, NSW, Australia, (3)University of New South Wales, Sydney, Australia, (4)Bureau of Meteorology, Melbourne, VIC, Australia
Accurate estimation of rainfall is critical for any hydrological analysis. The advantage of radar rainfall measurements is their ability to cover large areas. However, the uncertainties in the parameters of the power law, that links reflectivity to rainfall intensity, have to date precluded the widespread use of radars for quantitative rainfall estimates for hydrological studies. There is therefore considerable interest in methods that can combine the strengths of radar and gauge measurements by merging the two data sources. In this work, we propose two new developments to advance this area of research. The first contribution is a non-parametric radar rainfall estimation method (NPZR) which is based on kernel density estimation. Instead of using a traditional Z-R relationship, the NPZR accounts for the uncertainty in the relationship between reflectivity and rainfall intensity. More importantly, this uncertainty can vary for different values of reflectivity. The NPZR method reduces the Mean Square Error (MSE) of the estimated rainfall by 16 % compared to a traditionally fitted Z-R relation. Rainfall estimates are improved at 90% of the gauge locations when the method is applied to the densely gauged Sydney Terrey Hills radar region. A copula based spatial interpolation method (SIR) is used to estimate rainfall from gauge observations at the radar pixel locations. The gauge-based SIR estimates have low uncertainty in areas with good gauge density, whilst the NPZR method provides more reliable rainfall estimates than the SIR method, particularly in the areas of low gauge density. The second contribution of the work is to merge the radar rainfall field with spatially interpolated gauge rainfall estimates. The two rainfall fields are combined using a temporally and spatially varying weighting scheme that can account for the strengths of each method. The weight for each time period at each location is calculated based on the expected estimation error of each method, leading to a dynamic weighting scheme. The weighted combination method reduces the MSE by close to 50% compared to a traditional power law method for using the radar reflectivity. The combined method has the lowest overall error of the three methods over the densely gauged Sydney region.