It is shown that short-period seismic noise at the NORSAR seismic array can be described by a parametric model having the form
where X(t) is used to denote the noise process. This is a nonstationary generalization of the standard autoregressive models allowing time-varying autoregressive coefficients and a nonstationary white-noise residual process Z(t). The coefficients a1(t), a2(t), …ap(t) and the residual variance σ2z(t) are estimated using data from the NORSAR seismic array. It is found that the short-period noise at NORSAR is quite satisfactorily described by a third-order (p = 3) model. Also, it is found that the nonstationary character of the noise is due primarily to time variations of the residual variance and the lower order autoregressive coefficients.