In the process of model verification, we found that the day to day flows are reproducible only in the stochastic sense, which can be approximated by a normal distribution function specific to each time point since.
the mean and variance of flow change significantly over time. This approximation significantly reduces the complexity in model formulation.
The strong dependence of flows on time indicates that flow forecasting at a given time point must incorporate the historical information about flows at that time point.
It was also shown in the paper that the flow increment is dependent on time as well, though the level of dependency, is lower compared with the flow.
Including the time dependent historical information of flow increments in a forecasting model would improve the performance of the model.