The accuracy of radar‐derived rainfall measurements in hilly terrain

Abstract A weather radar has been used to measure rainfall over hilly terrain in north Wales. The radar was a standard Plessey Type 43S, which has a wavelength of 10 cm and a beam width to half power points of 2°. Measurements were compared with those based on a raingauge network consisting of 62 tipping bucket gauges distributed over 1,000 km2. This density of gauges was inadequate to define the hourly areal rainfall sufficiently accurately on some occasions and in these circumstances the radar‐derived pattern was used to interpolate between the gauges so as to obtain an ‘optimum’ estimate of the actual sub‐catchment rainfall which was then used to evaluate the accuracy of the radar measurement. Errors in the radar estimates were excessive unless the radar was calibrated hourly using raingauge measurements from one site. The accuracy was further increased when the horizontal drift of the rain in the wind between the radar beam and the calibration site was allowed for. Using these procedures, the mean percentage difference regardless of sign, Y, between the radar and optimum estimates of three‐hourly rainfall over sub‐catchments of typically 50 km2 varied from about 15% close to the calibration site to about 20% at a distance of 20 km. Differences decreased as the period of integration was increased; Y was 13% for 6 hour periods compared with 20% for 2 hour periods. Y also varied with the area of comparison. For point measurements hourly estimates differed by 37% but over an area of 500 km2 around the calibration site they differed by about 13%. The main causes of error in the radar measurements seem to be (i) variations in the drop size distribution relation, (ii) the spatial variation in the growth (or evaporation) which occurred between the beam and the ground and (iii) the horizontal drift of the rain in the wind. Probably all of these factors, but in particular (ii), are more important in the hilly terrain of this study than would be the case in flatter terrain. A further factor contributing to the differences between the radar and optimum estimates was the (unavoidable) error in each optimum estimate of the actual rainfall. The accuracy of the radar estimates decreased markedly if the beam intersected the melting layer. The increased errors were reduced by introducing an empirical correction factor which is a function of the range of the sub‐catchment from the radar and the height of the melting layer, but the spatial and temporal variations of the melting layer were such that it was not possible to obtain the same accuracy as when the beam was entirely within rain.