This chapter describes how observed streamflow can be used to optimize model performance by automatically estimating parameters. HEC-HMS has two different approaches to model optimization: deterministic, and stochastic. Deterministic optimization begins with initial parameter estimates and adjusts them so that the simulated results match the observed streamflow as closely as possible. Stochastic optimization produces a collection of equally-probable parameter sets which represent a sample from the joint distribution of the parameter population.


HEC-HMS optimization tools offer two deterministic search algorithms that move from the initial parameter estimates to the final best parameter estimates. A variety of objective functions are provided to measure the goodness of fit between the simulated and observed streamflow in different ways. One stochastic procedure for generating samples from the joint distribution of the parameters is included. While parameter estimation using optimization does not produce perfect results, it can be a valuable aid when calibrating models.


Deterministic and stochastic optimization are philosophically different approaches to the optimization problem. Deterministic optimization seeks to minimize the difference between the model outputs and observed data by changing model parameters to find a single, optimum set. Parameters optimized in this way may be used as parameter values in an ordinary Simulation Run. With the same parameters, a deterministic optimization will arrive at the same optimum parameters with each trial. Stochastic optimization infers what likely model parameter values are in light of the observed data, and can only do so by creating a number of parameter sets. This approach treats the parameters with uncertainty and does not return a single set of optimized parameters. In order to use parameter sets generated by a stochastic optimization, the user must use an Uncertainty Analysis and populate tables of sampled parameters. Stochastic optimizations will result in different parameter sets with each trial.