The spatial and temporal variation of rainfall over small scales substantially affects the estimation of runoff particularly in urban catchments, such as those in Hong Kong, where the time of concentration is typically short. A rainfall model that can adequately capture and describe the spatio-temporal variation of rain-field characteristics is highly desirable. A three-part framework for analyzing the dynamic time-space evolution of the rainfall consistent with 6-min, 1-km² pixel radar images is proposed and examined in this thesis.

The first part is to extract the characteristics of spatio-temporal variation of rain-field from the radar data. Geostatistical approach is employed to analyze the spatial structure of the rain-field. Indicator variograms of rain-fields are used...[ Read more ]

The spatial and temporal variation of rainfall over small scales substantially affects the estimation of runoff particularly in urban catchments, such as those in Hong Kong, where the time of concentration is typically short. A rainfall model that can adequately capture and describe the spatio-temporal variation of rain-field characteristics is highly desirable. A three-part framework for analyzing the dynamic time-space evolution of the rainfall consistent with 6-min, 1-km² pixel radar images is proposed and examined in this thesis.

The first part is to extract the characteristics of spatio-temporal variation of rain-field from the radar data. Geostatistical approach is employed to analyze the spatial structure of the rain-field. Indicator variograms of rain-fields are used to deal with the non-Gaussian distribution issue and are found to be adequately modeled by an exponential variogram model. The spatial structure of rain-field is found to be highly anisotropic and should be adequately considered in the model. The second part is to estimate the rainfall intensity over the selected study domain with the sample values and the information of the spatial structure (e.g., anisotropic variogram) derived from the radar rainfall data. The cumulative distribution functions at all unsampled locations are obtained by Indicator Kriging which provides estimations of exceedance probabilities for different rainfall intensity thresholds. A procedure for interpolation of cumulative distribution function values between discrete rainfall thresholds is implemented by fitting them to theoretical probability distributions. The performance measures and visual comparison between the observed rain-field and Indicator Kriging-based estimation suggest that the proposed method can provide satisfactory estimation of indicator variables and is capable of producing realistic time-space evolution of rain-field during rainstorm events. In the third part, autoregressive integrated moving average models are used to describe the within-storm temporal variations of the rainstorm characteristics describing its spatial structure. This framework could potentially provide useful basis for critical operation and management decisions in agriculture, hydrology, hydraulic structural design and other areas which depend on the reliable estimation and predictions of precipitation.