This paper investigates the problem of angle-of-arrival estimation in low-angle target tracking over the sea. For the angle of interest, the radar must contend with a reflected signal consisting of both a specular and a diffuse multipath. Conventional radar tracking methods used in sampled-aperture radar only consider the effect of specular component and ignore the diffuse component. Diffuse multipath, while arising out of the scattering of electromagnetic waves from rough sea surfaces, can affect the angle estimation performance. In order to increase the accuracy of angle estimation, the effect of the diffuse multipath must be considered. In a previous paper, we showed that diffuse multipath may not be a truly random phenomenon as is usually assumed. Instead, it may be modeled as a deterministic chaotic process. Based on this observation, a novel radar tracking technique is introduced. The new technique is based on the use of the local linear prediction method to estimate the power spectrum of the signals, and the use of the correlation dimension to estimate a suitable predictor order. We evaluate the performance of the new method using real-life radar data, and compare the results to those obtained by the conventional low-angle tracking technique.