Multivariate spatiotemporal data, where multiple variables are measured over time and across space, present modeling challenges due to complex dependencies. Traditional approaches often rely on assumptions like stationarity and separability, which may be too restrictive. This work introduces a flexible alternative by decomposing the random field into two independent subspaces: one stationary and one nonstationary. The proposed method, spatiotemporal stationary subspace analysis (stSSA), identifies the subspaces by segmenting the space-time domain and comparing local and global means. This decomposition simplifies modeling and enhances interpretability. A simulation study assesses the method’s sensitivity to segment construction, number of latent components and the choice of the covariance function.