iMOD User Manual version 4.4 (html)

10.14Data Set 14: Parameter Estimation – Main settings

Data Set 14



MXITER can have different meanings:


iMODFLOW will be run a single run and adjust all parameters accordingly and than stop.


If PE_MXITER is equal to zero, a sensitivity matrix will be computed yielding Jacobian values (finite difference between the change in head and the parameter update) for the entire zones. Those values will be written to disk in .\head\head_{date}_l{i}_sens_{param}_ils{ils}.idf. Those values can be helpful to estimate the adjustment to a parameter to yield a desired improvement of the head and/or flux (assuming the model act linearly). The process will stop whenever all parameters are perturbed.


Maximum number of iterations.


Stop criterion whenever decrease of objective function J becomes less or equal to the ratio J\({}_{i}\)/J\({}_{i}\)\({}_{-1}\). Entering a value of 0.1 means than the optimization stops whenever the objective function value J\({}_{i}\) for the current optimization step i, is reduced less than 10% of the last objective function value J\({}_{i-1}\).


Enter the acceptable sensitivity for parameters to be included in the parameter upgrade vector, e.g. PE_SENS=0.5 mean that parameters that have less than 0.5% sensitivity will be left out until they achieve a higher sensitivity.


Enter the number of periods. If PE_NPERIOD \(>\) 0, than repeat Data Set 15 for each period.


Enter the number of batch files to be included during the parameter estimation. Each batch file can have its own fraction that determines the weigh for the total objective function value.


Enter a fraction for each target:


The difference for each stress period between an available measurement and its corresponding observation


The difference between the measurement dynamics and the observational dynamics

The entered fraction should be entered relative to each other since iMODFLOW will recomputed the normalized values for the fraction. e.g. entering 1.0 and 2.0 will yield the fraction values 0.33 and 0.66, they will be summed equal to one. Whenever PE_NBATCH\(>\)0 (see Data Set 16), the entered weigh values for each batch file will be included in the final normalization of the fractions.


Enter a scaling option:


No use of scaling/Eigenvalue decomposition (SVD)


Only use of scaling


Use of scaling and Eigenvalue decomposition (SVD)


Only use of Eigenvalue decomposition (SVD)

In case a SVD decomposition is used (PE_SCALING=2 and PE_SCALING=3), eigenvalues that explain at least 99% of variance are included.


Enter the stopping criteria for Parameter ADJustment, e.g. PE_PADJ=0.05 means than whenever the parameter adjustment vector is less than 0.05, the optimization will stop. By default PE_PADJ=0.0 which means that the optimization will stop only whenever to parameters adjustment is applied.


Enter the minimal acceptable absolute residual used for the objective function. Absolute residuals smaller that PE_DRES will not be included in the objective function and therefore not influence any parameter adjustment. By default PE_DRES=0.0 which means that all residuals will be included.


Enter the type of Kriging to be used (whenever the PilotPoint concept is used). By default Simple Kriging is applied (PE_KTYPE=1), select PE_KTYPE=2 for Ordinary Kriging. The latter is used whenever a trend exists in the PilotPoints.


Enter the size of the range (meter) which is used whenever pilot points are used, e.g. PE_KRANGE=5000.0.


Enter PE_REGULARISATION=1 to apply a regularisation for the prior-estimated parameter values. Any deviation of the prior-estimated parameter value (see ..) will punish the objective function quadratic as \((p_{\rm ini}-p)^2\).


Specify the multiplication factor for the prior-estimates of the parameter values. This acts as a weighting value to scale the deviates of the parameters (\(w_p(p_{\rm ini}-p)^2\)) to the deviates of the measurements. Suppose the measurements are weigthed with a maximal value of 1000, the PE_REGFACTOR should be 1000 as well.