In order to establish quantitative timbre representation of underwater noise, this paper tries to build a relationship between essential timbre scores and central auditory responses to stimulus based on partial least squares regression, and use regression coefficients to interpret the physical meaning of each dimension. In order to verify the utility of this method, this paper extracts a large amount of timbre descriptors as independent variables for comparisons, and it is shown that the predictive ability of the central auditory model is better. Finally it is found from two types of timbre representations that the first three dimensions in the timbre space can be respectively interpreted as high-frequency energy ratio, spectral flatness and temporal continuity; however, dimension 4 cannot be related to any acoustical features.