Taking Chinese eastern plains as an examples, the authors construct a set of predication models for trace metal partitioning with multiple regression functions.