Synthetic daily rainfall data generation

Abstract

Long sequences of daily rainfall are often needed for simulation. These (when available) are cumbersome to input. Also in many cases historical data are to short to include all possible patterns. Hence the need for synthetic data. Discovering the stochastic structure underlying daily rains is the key to devising method for synthetic generation of data. Some works have been reported that treat daily rains as multi-state Markov chain. This is useful in studies where one needs for instance the distribution of only the dry and wet spells etc. However, for use in simulation of run-off from a watershed, or for moisture budgeting and crop planning, or for scheduling of irrigation etc. one needs the magnitudes of rainfall and not just an interval. For these applications it appears necessary to look at daily rains as a Markov process as was done for instance by Carey and Haan for Kentucky. In this paper we report the results of using C&H method to generate synthetic data for Panchmahals district of Gujarat, a drought-prone area.