Multivariate analysis of radar images for environmental monitoring

The identification of the relationship between radar reflectivity factor Z, expressed in mm6 m-3, and rainfall intensity R, expressed in mm h-1, is crucial for both the calibration and operational phases. The Marshall and Palmer relationship, which links together Z at the lowest constant altitude plan position indicator (CAPPI) level and R, is commonly used in operational hydrology. This relation is of the form Z = aR^(b). Coefficients a and b reflect the dependence of Z from the number and size distribution of meteors present in the volumes scanned by radar beam. However, as a and b (which depend on the type of precipitation) are affected by great variability and both Z and R are affected by errors, detailed statistical analyses of the Z-R relationship may help improving the operational capabilities of weather radar. In this framework, the aim of the paper is twofold: (1) to develop a non-parametric approach which is more flexible and offers more generalisation capabilities than the MP relationship;(2) to use a vector of all 11 reflectivity factors at different CAPPI levels. For this purposes, three kinds of neural networks are developed: the multi-layer perceptron, radial basis function networks and Bayesian networks. Models are trained and tested using a real data set of reflectivity observed by the Monte Grande weather radar (Teolo, Italy) and rainfall intensity measured at five rain gauges in the Cortina d’Ampezzo area (Northern Italian Dolomites), during the June 12, 1997 storm event (from 11.15am to 12.00pm). Reflectivity data are given at 11 CAPPI levels with 15-minute time resolution. Rainfall intensity data are measured at 5-minute time resolution and are averaged over the 15-minute time intervals of radar data to constitute integrated measurements.