It is known that the least-squares support vector machine (LS-SVM) can model nonlinear temporal dynamics with high accuracy, but it is not appropriate space information. Here, a spatiotemporal LS-SVM approach is proposed to address the time-varying modeling problem and small samples caused by space and cost constraints, in which the space kernel function (SKF) is developed for the nonlinear correlation between space locations, and temporal coefficient model is designed for nonlinear temporal dynamics of the DPS. In this way, integrating the SKF and temporal model, the nonlinear spatiotemporal dynamics is reconstructed. Due to accounting for the time dynamics and the space distribution nature, the proposed method can adapt well to real-time spatiotemporal variation, and the modeling effect is presented by a practical curing thermal process.