Download the TFM demonstration software click here

TFM is a program for the analysis of rainfall-discharge catchment data based on Transfer Function Model concepts, similar to those used in the IHACRES model of Jakeman et al. (1990, 1993, Jakeman and Hornberger, 1994) and the bilinear power model of Young and Beven (1991, 1994) (See Chapter 4). The opening screen of TFM reveals three buttons: one to QUIT the package, one to go to the Load File option and one to open a Log File. The Log File is used to record the data files used, any data transformations carried out and the results of Model Identification and Parameter Estimation options.The data is saved in ASCII text format for later editing or use in reports. The basic options used in TFM is as follows:

Load File Option

The Load File option on the main TFM screen or the New Data option on the Plot Data screen allows the entry of file names for the Input and Output data sets to be used in the analysis. TFM requires input and output data are fixed time intervals to be available. Conflicts between the input and output data files are identified and flagged by warning messages. Where the two data sets have different file lengths, the length is set to the length of the input data file. It is assumed that both files start at the same time. Successfully loading the input and output data files brings up a Plot Data screen from which the other options available can be chosen. Example data files can be downloaded with the program.

Transform Option

The Transform option in the Plot Screen allows two nonlinear transformations to be made to the input data, corresponding to the bilinear power model and the Storage or soil moisture index approaches (the latter is a simplification of the IHACRES approach). These transformations are required to create an effective rainfall input that may be more linearly related to the output than the original rainfall data. Each require a single parameter value (either the power in the bilinear power model or the time constant of the storage. These are easily optimised for any data set by making repeated runs of the model. Other options allow an initial run of data to be excluded from the analysis, or the data file to be reduced by skipping every n th value. The transformed file can be saved or the original data can be restored.

The Identify option

The Identify option allows a wide range of model structures to be evaluated quickly. The models are specified in terms of a range of numbers of a parameters, a range of numbers of b parameters, and a range of time delays. The models are ranked in terms of the value of the Young Information Criterion or YIC (see Box 4.1). This should be as negative as possible, indicating both a good fit between observed and predicted outputs and well defined parameter values. A value for the coefficient of determination RT2 is also given. This is 1 for a model giving a perfect fit, 0 for a model that is no better than assuming a mean output value. Underlying the model identification step is the idea that the data should be allowed to indicate which model structure is most appropriate rather than specifying a structure beforehand. There may be many models giving an acceptable fit to the data. The simplest model giving the most negative YIC value and a physically reasonable transfer function should then be chosen. A double click of the mouse on any of the models listed after the Identification process will take control directly to the Estimate option screen and will estimate the parameters of the model.

The Estimate option

This option carries out the final estimation of the parameters of the chosen model. Results from the most recent estimation can be plotted. The most recent results are shown on the table of results shown on screen. If a Log File is open then the results are also written to the Log File. Parameter estimation is carried out using a Simplified Refined Instrumental Variable (SRIV) technique which has been shown to be very robust to data errors (see Young, 1984). The results may be evaluated using the YIC and RT2 criteria. A number of other options allow graphical and statistical examination of the results. Plot Model gives a plot of the observed and predicted outputs, together with a plot of the observed error series. The modelled time series can be saved to a file Plot TF gives a plot of the transfer function for the chosen model. This should be checked to ensure that the model transfer function is always positive and does not show any significant oscillatory behaviour or instabilities. The transfer function may be saved to a file. Composite gives a combined plot of the data, model fit and transfer function. This may be printed as a summary record of the analysis. The transfer function may be saved to a file. Validate allows the model to be tested against a new data set. The Validate option can only be chosen after the Estimate option has been performed and a model is available. It may be used in two ways depending if both input and output data are available. If both are available, the model is assessed in terms of fitting the new data for the validation period in terms of an RT2 value. If only input data are available, the model may be used to predict the missing flow data. To use the Validate option, fill in the file name fields (full path and extension required) and choose calculate. Once the new outputs have been simulated, the data may be saved to a file.