Incoherent and coherent wideband MUSIC algorithms are used to perform high-resolution direction of arrival (DOA) estimation on ground vehicles that emit broadband acoustic signatures (e.g., turbine engine vehicles). The algorithms are analyzed with experimental data collected from a small baseline, circular array [T. Pham and B. Sadler, 8th IEEE SSAP Workshop (1996)]. GPS ground truth location data are used to evaluate and compare performance. The nonstationary nature of the detected signals, due to environmental and source variations, requires adaptive selection of a set of operating frequencies. In the incoherent method [M. Wax, T. Shan, and T. Kailath, IEEE Trans. ASSP 32 (1984)], MUSIC is processed individually at each narrow-band frequency, and results averaged to estimate the DOAs. In the coherent method [H. Wang and M. Kaveh, IEEE Trans. ASSP 33 (1985); also J. Krolik, Adv. Spect. Anal. Array Proc., Vol. II (P-H, 1991)], correlation matrices at different frequencies are combined at each look direction to form a ``focused'' correlation matrix, and MUSIC is performed on the focused correlation matrix to estimate the DOAs. Processing gain is evident for both methods compared to narrow-band MUSIC. Computational complexity is dominated by eigendecomposition; it is a function of the number of frequency components for the incoherent case, and the number of look angles for the coherent case.