Regresar

Improving extreme value behaviour of fine-scale stochastic point process models

Abstract:

Urbanization and climate change encourage water managers to improve their ability to pbkp_redict possible future rainfall events. To study impacts on urban drainage and river systems and assess their vulnerability, long term simulations at fine time scales, including extreme rain storms of high return periods, are of critical importance. Bartlett-Lewis rectangular pulses models are considered to provide such long term simulations. These models have proven to be capable of repro-ducing general historical rainfall characteristics but tend to overestimate extremes at higher levels of aggregation, and underestimate them at lower levels of aggregation. Furthermore, unrealistically large rainfall events are occasionally generated during simulation. This might lead to serious implications when the simulated rainfall series are used for im-pact analysis in urban hydrology. The presented research focuses on ways to improve extreme value behaviour of the Bartlett-Lewis models by introducing the third order moment of rainfall intensity in the objective function. By doing so, the tail of the rainfall distribution is represented better during calibration. The extreme values generated by a standard Bartlett-Lewis model is analysed using the Peak-Over-Threshold method. Secondly, the occasional simulation of unreal-istically large rainfall events is addressed by an adjustment to the model structure. By truncating the gamma distribution responsible for the simulation of cell durations, the probability of sampling extremely long rainfall events is drastically reduced.