First lag autocorrelation Number of sign changes
Minimize w.r.t 9 Minimize w.r.t 9
(Count the number of times the sequence of residuals changes sign)
nonhomogenous variance. The advantage of the ML approach is that the validity of the underlying assumptions can be verified by a postcalibration residual analysis. It is important to note that each of these scalar measures defines a different way to gauge the "size" of the error vector, and the minimum value for each will define a simulated streamflow sequence that is "close" to the observed data in a different way. If a certain scalar measure is selected, then it is possible (in principle) to find a single parameter set (or a small region of the feasible space) that minimizes that measure. This makes the model calibration procedure much easier to automate so that the accuracy, speed, and efficiency of a computer can be exploited. Nonetheless, while more efficient than visual comparison, the use of scalar measures is perhaps no less subjective.
Was this article helpful?