With the development of modern agriculture, large amount of fertilizer and pesticide outflow from farming land causes great waste and serious pollution to surface water and groundwater, and threatens ecological environment and even human life. In this paper, laboratory experiments are conducted to simulate adsorbed Cr(VI) transfer from soil into runoff. A two-layer in-mixing analytical model is applied to analyze laboratory experimental results. A data assimilation (DA) method via the ensemble Kalman filter (EnKF) is used to update parameters and improve modeling results. In comparison with the experimental data, DA updated modeling results are much better than those without the updating. To make predictions better, the inflation method with a constant inflation factor via DA method was used to compensate the fast decrease of ensemble spread partially related to filter inbreeding. Based on the used rainfall and relevant physical principles, the updated value of the incomplete mixing coefficient γ is about 14.0 times of the value of the incomplete mixing coefficient α in experiment 1 and about 7.4 times in experiment 2, while the difference between the flow rate of runoff and infiltration is not so large even after reaching stable infiltration condition. The results indicate the loss of Cr(VI) in soil solute is mainly due to infiltration, rather than surface runoff. With the increase of mixing layer depth, soil adsorption capacity will increase and the loss of soil solute will decrease. The study results provide information for reducing and even preventing the agricultural nonpoint source pollution.