Stochastic disaggregation of daily rainfall into one-hour time scale

Abstract

The Australian SILO Data Drill facility generates continuous daily rainfall data from 1889 to current date for any set of coordinates on the Australian continent. For the daily rainfall data to have any appeal to users, such as farmers and environmental modellers, a robust disaggregation model that generates sub-daily time series fully consistent with the daily totals while preserving multiple sub-daily time scale stochastic structure is required. A model, which incorporates repetition techniques and a proportional adjusting procedure [Koutsoyiannis, D., Onof, C., 2001. Rainfall disaggregation using adjusting procedures on a Poisson cluster model. J. Hydrol. 246, 109–122] into a regionalised hybrid model [Gyasi-Agyei, Y., 1999. Identification of regional parameters of a stochastic model for rainfall disaggregation. J. Hydrol. 223(3–4), 148–163], has been demonstrated to have such capability. The model is structured such that clusters of consecutive wet days can be disaggregated together during the generation of the binary wet and dry sequence and/ or intensity phases. Two ad hoc remedies to prevent overestimation of the variance and the extreme values, and underestimation of the autocorrelation that could be potentially caused by daily rainfall totals in excess of the hourly maximum have been proposed. The model was evaluated with a 5-year time series of hourly rainfall observed at an erosion control experimental site. All modes of operation of the model reproduced the dry probability very well. Clustering does not seem to affect the dry probability, variance and the intensity-duration–frequency (IFD) curves. The reproduction of the autocorrelations certainly improved with clustering. Capping the hourly rainfall depths to the observed maximum values reproduced near perfect dry probability, variance, autocorrelation and the IFD curves for all months. With this confidence, the 114-year synthetic daily rainfall data set for Rockhampton generated by SILO Data Drill facility was disaggregated to one-hour time scale. The pattern of the results from this data set was identical to that of the observed 5-year data set.