The invention relates to a big data-based marine environmental disaster early warning method. The early warning method includes the following steps that: (1) the distributed web crawler method of a consistent Hash algorithm is used to obtain a marine environment historical original index spatial-temporal database DRO and a marine environment historical evaluation index spatial-temporal database DEO, and a data set Dro (Ri (xi1, xi2, ..., xij), ti) and a data set Deo (Ei (yi1, yi2, ..., yim) and ti) are obtained; (2) the data sets are standardized; and (3) a marine environment evaluation and disaster early warning model is established. The step (3) of the method of the invention further includes the following steps that: (3.1), on the basis of the Dro (Ri (xi1, xi2, ..., xij), ti) and the Deo (Ei (yi1, yi2, ..., yij) and ti), a Kringing method is used to establish a marine environment gradual change simulation model MS (Dro (Ri, ti), Deo (Ei, ti)), and then the correlation analysis of each marine environment index and the evaluation of environmental health levels are carried out, and a health evaluation model is established; and (3.2), on the basis of the Dro (Ri (xi1, xi2, ..., xij), ti) and Deo (Ei (yi1, yi2, ..., yij), ti) big data, an SVM (support vector machine) method is adopted to perform early warning for classification indexes and comprehensive indexes.