Narrowband direction-of-arrival (DOA) estimates for each time-frequency (TF) point offer a parametric spatial modeling of the acoustic environment which is very commonly used in many applications, such as source separation, dereverberation, and spatial audio. However, irrespective of the narrowband DOA estimation method used, many TF-points suffer from erroneous estimates due to noise and reverberation. We propose a novel technique to yield more accurate DOA estimates in the TF-domain, through statistical modeling of each TF-point with a complex Watson distribution. Then, instead of using the microphone array signals at a given TF-point to estimate the DOA, the maximum likelihood estimate of the mode vector of the distribution is used as input to the DOA estimation method. This approach results in more accurate DOA estimates and thus more accurate modeling of the acoustic environment, while it can be used with any narrowband DOA estimation method and microphone array geometry.