A NONISOTROPIC GAUGE INTERPOLATION SCHEME APPLIED TO THE MONTREAL RAINSTORM OF 14 JULY 1987

Raingauge data from the 14 July 1987 rainstorm over Montreal are used to underline the reduced errors obtained with a non-isotropic interpol ation scheme. This scheme modifies the geometrical distance between a point and a set of gauges on the basis of the orientation and elongati on of the surface rainfall pattern. Initial guesses of these two param eters can be obtained from the two-dimensional autocorrelation pattern of radar-derived maps of surface rainfall. The set of optimum paramet ers can also be derived from raingauge data only by computing error st atistics obtained from the successive omission of one gauge from the a vailable network. The RMS error of 18% is significantly better than th at of 26% obtained with isotropic interpolation. Results from data on this day also reveal that RMS errors are relatively insensitive to the functional relationship between the weights of the optimum number of gauges and the distance to a given point. The orientation and elongati on of a rainfall pattern is of greater importance in any interpolation technique. This statement confirmed that an ''oriented'' Thiessen tec hnique that incorporates the orientation and elongation reduces RMS er rors from 34 to <21%. The latter is comparable with the magnitude of 1 8% obtained with the optimum number of gauges, (3), and with a weighti ng function that is exponentially decreasing with distance. Interpolat ed rainfall fields derived with the optimized set of parameters have r evealed that large areas of the city of Montreal received rainfall amo unts at a rate corresponding to more than twice the 100-year return pe riod's.