A computationally efficient framework for stochastic prediction of flood propagation

This paper presents a computationally efficient method to forecast floods stochastically. The main purpose of the method is to encapsulate prior knowledge as off-line calculations leading to earliest possible warnings. The computational efficiency is improved through exploiting the stereotypical features of hydrology and its dependence on topography by combining the parallel water-flow processes into parallel calculation through a probability transition matrix. Efficiency is improved further with the use of properties of Markov matrices in the general equation of the model. Extensive simulations on real rainfall data over parts of Queensland, Australia, during January 2012, revealed that this method was capable of improving the calculation efficiency by over 18 times with respect to gradient based calculations.