A correlation–regression method for tracking rainstorms using rain-gauge data

Abstract

This paper focuses on the short-term tracking of rainstorms, using data from a network of 20 tipping-bucket rain-gauges located in a 600 km 2 region near Bolton in Northwest England. The tip-time data are first converted into rainrate profiles (the moving averages of hourly rainrates). The cross-correlation between pairs of profiles is examined to obtain estimated inter-gauge lags. Estimates from gauges that correlate poorly are discarded, since these gauges may be malfunctioning. The remaining estimated lags are modelled using linear dependencies on the spatial displacements of the gauge-pairs and the velocity of the rainstorm is deduced using standard formulae. Sums of squared deletion residuals give a second opportunity for identifying gauges associated with misleading estimates. Empirical confidence intervals are obtained by multiple analyses using different combinations of tip sequences. All rainstorms between February and December 2000 were analysed. Of these, 171 had a mean rainrate of at least 0.4 mm/h. Apparently accurate estimates of storm velocity were obtained for 107 storms. The mean speed was 34.5 km/h (9.6 m/s) with mean direction 80°.