Mathematics and Computers in Simulation 32 (1990) 113-118 North-Holland 113 STOCHASTIC WEATHER MODELLING: A PHENOMENOLOGICAL APPROACH L. GUENNI, D. CHARLES-EDWARDS, C.W. ROSE, R. BRADDOCK and W. HOGARTH Division of Australian Environmental Studies. Griffith University, Nathan, QLD, Australia 1. INTRODUCTION Daily sequences of weather data are required inputs of many agronomic and hydrologic models. Long-term sequences of daily weather records coupled with such models, are important tools in evaluating the responses of an agricultural or hydrological system to climatic variability. Daily weather records of long duration are not always available for each location and the evaluation of a particular system to climatic variability may not always be possible. Stochastic models capable of producing weather sequences which are statistically indistinguishable from the observed data, can be used to overcome this problem. The use of stochastic models acknowledges that randomness is an inherent characteristic in the weather and climate. More realism can be added if the the basic interactions of the physical system are incorporated in the model. In the present paper, a model for the stochastic generation of long-term sequences of daily climatic variables is presented. The variables of interest are daily rainfall, maximum and minimum temperature and insolation amounts. The model is designed to represent the main interactions between these weather variables, resulting from many physical processes occurring in the atmosphere and the earth/air interface. 2. MODEL DESCRIPTION For modelling purposes, the main interactions identified among the variables are: - rainfall occurrence and amount of cloudiness. - cloudiness and maximum and minimum temperatures. The model has three main components or sub-models. The year was divided into 26 periods of 14 days each, and the appropiate climate data was reduced to the same temporal basis. The model parameters were estimated for each 14-day period. A description of each component and an outline of the parameter estimation procedures follows: 2.1 Rainfall sub-model The rainfall sub-model is based on a Rectangular Pulses Poisson model [4]. The rainfall occurrence is assumed to follow a Poisson process of rate X per unit time. To each occurrence, there is associated a rectangular pulse of random duration L and random height X, which represents the rainfall intensity. It is assumed that the duration and intensity of each pulse are mutually independent, 0378-4754/90/$3.50 0 1990, IMACS/EIsevier Science Publishers B.V. (North-Holland)