Usage of the package

The SVP can be used as a stand-alone program or from within a C++ or TCL program as a submodule. The operation in both cases depends on a definition file which needs to be written by the operator. The file is used to specify symbolic names and attributes for each one of the variables that the operator intends to test. Attributes for each variable are:

• reference (observed) data, either directly or read from files, or other criteria to be met according to the kind of tests one wants to perform;
• actual (model output) data, either directly or read from files, to compare with the reference data using one or more tests. In the stand-alone program these data must be available before the invocation of the program; in the embedded (C++ or TCL) version of the SVP, they can be specified for each variable when they become available;
• a weight expressing the relative importance of each variable in the computation of the global score;
• one or more tests to perform, each one with optional parameters (when applicable) and a weight expressing the relative importance of the test in the computation of the variable's score. All tests return values in the interval $[0,1]$, with 1 expressing perfect agreement between reference and actual data, or full meeting of the test's conditions;
• one or more output format definitions to enable the user to write scores, and statistics pertaining to the variable to the console and/or to files.

Test weights and output specifications can be also defined outside of a variable definition, to be used as defaults values or to report global informations (such as the global score) to files.

For each variable $v$ specified, a variable score $V_v$ is computed as a weighted average of the scores $T_t$ of each test $t$ performed:

$$V_v = \frac{\sum_t W_t T_t} {\sum_t W_t}$$ (1)

where $W_t$ represents the user-defined weight (relative importance) of test $t$. The global score is computed similarly as

$$S = \frac{\sum_v W_v V_v}{\sum_t W_v}$$ (2)

with $W_v$ being the user-defined weight of variable $v$. Weights allow the experimenter's priorities to be accounted for, and are all 1 unless specified. The normal use of the SVP is to carry out a number of simulations with different values of the parameters and to sort the parameter lists according to the global score. Of course, it can be used just to obtain informations about the data. It is designed for flexibility and ease of use.