Predicting Daily Runoff Using Stochastic Rainfall Data Generation And Rainfall-Runoff Models

Runoff data generally has short record lengths; however rainfall data with long records is available. Therefore, rainfall data will include wider range of conditions compared to runoff data. This paper will use stochastic rainfall data with conceptual rainfall-runoff model to generate synthetic runoff data for long term water resources design. In this paper, the ability of generating runoff data in the future using statistics of historical rainfall data will be analysed. For this purpose, rainfall data were divided into two samples: referred to as the Earlier Period and the Later Period. The rainfall data will be split according to the length of runoff data. The Later Period will be of the same length with the runoff data, whilst the Earlier Period will be all data before the runoff data. For the stochastic rainfall data generation, Transition Probability Matrices model will be used. For the rainfall-runoff modelling, Soil Dryness Index model will be used. Daily data will be used as the period of available data is long. Catchment used in this study was Kangaroo Valley, Australia. Comparison will be made between synthetic and recorded runoff data (daily, monthly and annual statistics; daily, monthly and annual maxima; monthly and annual minima; lag-one serial correlation coefficients). Daily, monthly and annual means and standard deviations were reproduced satisfactorily. However, lower values of skews tended to be generated indicating more normally distributed data were generated. Daily, monthly and annual maxima generated tend to be higher than the recorded values. Monthly and annual minima generated tend to be lower than the recorded values. This behaviour was desirable since it gave values of runoff outside the range of recorded runoff, which may not cover a complete range of possible values. The lag-one serial correlation coefficient was satisfactory as it is of the same range (slightly lower) as the recorded values. Generally, the use of historical rainfall data (Earlier Period) statistical properties to generate data for the future (Later Period) and then applying rainfall-runoff model to generate runoff data was found to be satisfactory.