Articles | Volume 21, issue 12
https://doi.org/10.5194/hess-21-6541-2017
https://doi.org/10.5194/hess-21-6541-2017
Research article
 | 
22 Dec 2017
Research article |  | 22 Dec 2017

Development and evaluation of a stochastic daily rainfall model with long-term variability

A. F. M. Kamal Chowdhury, Natalie Lockart, Garry Willgoose, George Kuczera, Anthony S. Kiem, and Nadeeka Parana Manage

Abstract. The primary objective of this study is to develop a stochastic rainfall generation model that can match not only the short resolution (daily) variability but also the longer resolution (monthly to multiyear) variability of observed rainfall. This study has developed a Markov chain (MC) model, which uses a two-state MC process with two parameters (wet-to-wet and dry-to-dry transition probabilities) to simulate rainfall occurrence and a gamma distribution with two parameters (mean and standard deviation of wet day rainfall) to simulate wet day rainfall depths. Starting with the traditional MC-gamma model with deterministic parameters, this study has developed and assessed four other variants of the MC-gamma model with different parameterisations. The key finding is that if the parameters of the gamma distribution are randomly sampled each year from fitted distributions rather than fixed parameters with time, the variability of rainfall depths at both short and longer temporal resolutions can be preserved, while the variability of wet periods (i.e. number of wet days and mean length of wet spell) can be preserved by decadally varied MC parameters. This is a straightforward enhancement to the traditional simplest MC model and is both objective and parsimonious.

Download
Short summary
Stochastic rainfall models are required to be be able to assess the reliability of dams used for urban water supply. Traditional Markov chain stochastic models do well at reproducing the mean and variance of rainfall at daily to weekly resolution but fail to simultaneously reproduce the variability of monthly to decadal rainfall. This paper presents four new extensions to Markov chain models that address this decadal deficiency and compares their performance for two field sites.