Point process models for weather radar images

A framework for analysing weather radar (DBz) images as spatial point processes is presented. Weather radar images are modelled for the purpose of predicting their evolution in time and thereby providing a basis for short-period precipitation forecasts. An observed image sequence is modelled as a set of individual rain cells that are the outcome of a marked 2+1D spatial point process. To each point giving the place and time of maturation of a rain cell is assigned a vector of possibly time-varying features such as intensity, duration, extent, shape and velocity. The point process is a doubly stochastic spatial point process with a clustering mechanism determined by the mesoscale situation. Also determined by the mesoscale situation are prior distributions for the elements of the feature vector. A scheme for fitting this type of model to an observed sequence of weather radar images is presented