Shallow water is a complicated sound propagation medium due to multiple reflections by water surface and bottom, imprecisely measured sound speed, noisy environment, etc. Therefore, in order to localize a shallow water sound source, classical signal processing techniques must be improved by taking these complexities into account. In this work, the multiple reflections and uncertain reflectivity of water bottom are explicitly modeled. In the proposed model, a measured signal is a mixture of the direct propagation from the source and the multiple reflections. Instead of solving the Helmholtz equation with boundary conditions of reflections, each signal is interpreted as a superposition of signals emitting from the physical source and its image sources in a free space, which results in a fast computation of sound propagation. Then, the source location, along with its amplitude, reflection paths and power loss of bottom reflection, is estimated via the iterative beamforming (IB) method, which alternatively estimates the source contributions and performs beamforming on these estimates until convergence. This approach does not need to compute the sound propagation for all the possible source locations in a large space, which thus leads to a low computational cost. Finally, numerical simulations are introduced to illustrate the advantage of the proposed model and the source estimation method. The sensitivity of the proposed method with respect to model parameter uncertainties is also investigated via a full uncertainty quantification analysis. The localization error of IB is proved to be acceptable in the given error range of sound speed and water depth. Besides, the IB source estimate is more sensitive to the sound speed while the matched-field processing methods have a stronger sensitivity to the water depth: this result can guide the choice of source localization method in different cases of model parameter uncertainties.