Empirical models and process-based theoretical or numerical models are valuable tools for analyzing sediment dynamics by reproducing the suspended sediment concentration (SSC) observations, but they are either limited by simplified assumptions or computationally complex. Some data-driven models have been developed to predict SSC but are still weak in explaining the causes of its changes. A field observation of silty sediment dynamics was performed in the Yellow River Delta, China. Then, a vector autoregressive (VAR) model is employed to analyze and predict the SSC variations. Using the impulse response function, the VAR model quantifies the transient and cumulative responses of the SSC to storm and current shocks (i.e., the impulse input), which reveals the key mechanisms of SSC changes from a new perspective. Results show that the storm effect on SSC in the study area lasts for ca. 85 h, and the SSC peaks at around the 36th hour with a cumulative increase of about 1.3 g/L due to the storm’s cumulative effect. Quantitative analysis reveals that waves contribute 73.78% to the SSC, while tidal currents contribute 26.22%. Fine SSC is significantly impacted by horizontal advection. A single reciprocal motion of the tidal current affects the SSC for approximately 6.25 h, while the overall duration exceeds 100 h. The proposed model physically reveals that the storm-induced high concentration gradient has been pushed back and forth by the tidal currents, affecting the SSC at the observation site, and the horizontal advection effect lasts for more than 100 h. Moreover, the VAR model achieves high-accuracy interval predictions of SSC for the next 5 h using only historical hydrodynamic data, with a 100% coverage probability of the 95% prediction interval. This paper provides a simple and efficient method for sediment dynamics analysis, which is of great significance for marine environment protection and disaster warning.