Skip to content

Climate only TFN model tutorial

Tim Peterson edited this page Apr 19, 2018 · 28 revisions

Overview

This tutorial presents the steps for build a time-series model of a groundwater hydrograph using a transfer function noise time-series model where the groundwater level elevation is only influenced by rainfall and evaporation. The tutorial uses data from an example model built into the HydroSight GUI, specifically Bourke Flat. Follow this tutorial to build your own model using this data, or alternatively use the HydroSight GUI example titled TFN Model - Landuse change to inspect the complete model.

The major steps are:

  1. Preparing the input data: for input the time-series modelling.
  2. Building the TFN model: by inputting the input data files, selecting the bore to model and defining the form of the model.
  3. Calibrating the model: by automatically tuning the model parameters to best first the observed hydrograph.
  4. Assessing the calibrated model: to determine if the model results, and predictions, can be deemed acceptable.
  5. Perform simulations: using the calibrated model to, say, interpolate infrequent water level observations or explore different climate scenarios.

In the following each of these steps is detailed.

The input Data

The TFN models require the following three input .csv files. See Data Requirements for general format requirements. The tasks for this step are:

  1. Download the zipped example data from here.
  2. Create a project folder on your computer.
  3. Unzip the example data into your project folder.
  4. Open each of the three .csv files within a spreadsheet and observe the following aspects of the HydroSight data requirements:
    • The head file (obsHead.csv) data column BoreID allows multiple bores to be within a single file. This file has two bores listed.
    • The head data can be of an irregular monitoring frequency.
    • The water level observations must be an elevation, not depth.
    • The water level observations start in 1983.
    • The forcing data file (forcing.csv) left three columns are Year, Month and Day. The columns are essential and must be in this order and named as such.
    • The forcing data file right three columns are Rain_mm, FAO56_PET_mm and Landuse_change_frac. Theses columns can have any name and are the forcing data available for input to the time-series model.
    • The forcing data file has the following requirements: a daily time step, no time-step gaps and the first data point many years before the first water level observation, which in this tutorial is 1983.
    • The coordinates data file (coordinates.csv) must have the following three columns: SiteID, Easting and Northing. The rows must list the location of all observation bores within the head .csv file and all columns within the forcing data .csv file and the SiteID inputs must be identical to that within the respective files.
    • The coordinates file easting and northing is, currently, only used when modelling pumping drawdown. For climate only models the coordinates can be any number.

Building the Model

To start building a time-series model of a groundwater hydrograph complete the tasks below:

  1. Open HydroSight.
  2. Click on the tab Model Construction.
  3. Set the project folder to where the input data was saved using the menu File > Set Project Folder.
  4. In the column Model Label, enter a title for the model. This can be anything but must start with a letter and be unique to the other model labels within the project. In this example let's input "controlCatchment_model1".
  5. Input the head file by placing the cursor in the next cell to the right (column Obs. Head File). This should open a file selection window showing the files within the project folder. Use it to select the file "obsHead.csv".
  6. Input the forcing file by placing the cursor in the column Forcing Data File. This should also open a file selection window showing the files within the project folder. Use it to select the file "forcing.csv".
  7. Input the coordinates file by placing the cursor in column Forcing Data File. This should also open a file selection window showing the files within the project folder. Use it to select the file "coordinates.csv".
  8. Select the bore to model by placing the cursor within the column Bore ID. This should cause a white box to be displayed on the right that lists all bore IDs within "obsHead.csv". Click on "Bore_6416" to select it for this model.
  9. Select the type of time-series model using the drop-down in the column Model Type. Notes, when you select a model type, a text box should be shown on the right giving a summary of the selected model type. For this tutorial select the option "Model_TFN".

Now that the inputs and type of model are defined, the form of the TFN model can be defined by placing the cursor within the column Model Options. This will display input box(es) on the right that are specific to the type of model (see screen shoot below). For TFN models the primary inputs are to transform the input forcing data (e.g. transform precipitation into free-drainage), weight input or transformed forcing data to fit the observed hydrograph. Below are the tasks for defining the model options:

TFN modle options

  1. Within the top-left box (see red "1" above), use the drop-down to select the type of transform to use. In this tutorial use "climateTransform_soilMoistureModels" (see here for model details). This will add a single storage soil moisture model to the time-series model.
  2. Place the cursor within the column "Input Data". This will display a grid below it. The left column of the grid lists the possible inputs for the model, which are precipitation, potential evapotranspiration and the fraction of the landscape influenced by the land cover being investigated. In this tutorial only precipitation, potential evapotranspiration are required. Hence, use the drop-down option in the right column to input Rain_mm to the first row and FAO56_PET_mm into the second row. For the input TreeFraction input "(none)" to turn this feature off within the model.
  3. Place the cursor within the column Options to display a table of options for setting the soil model parameters (see red "2" above). The column Parameter lists the parameters, of which there are upto eleven and the column Fixed or Calibrated? is used to specify which of the eleven are to be calibrated. In this example, set only the following parameters to "Calib": SMSC, k_sat and beta. The other parameters will be fixed at the Initial Value.
  4. Within the top-right box (see red "3" above) column Component Name, input a label for driver of heads to be simulated within the use the drop-down to select the type of transform to use (e.g. "Recharge").
  5. Use the drop-down within the column Weighting Function to select "responseFunction_Pearsons". When yo do this, a text bow should be shown below giving a summary of the function.
  6. Place the cursor within the column Input Data to select which forcing time-series to weight using the Pearsons response function. When you do this, a white box should appear at the bottom that lists all input data from the forcing file plus all of the possible fluxes and state variables from the "climateTransform_soilMoistureModels" function. Select the item "climateTransform_soilMoistureModels : drainage normalised" to use the normalised free drainage daily time series.
  7. Click on the button "<" (see red "4" above) to push the set of model option back to the left hand table.
  8. Within the left hand table of models, use the checkbox to select the model you've been defining.
  9. Click on the button Build Selected Models to build the model. If successful, then the column Built Status will show "Model built."

Calibrating the Model

To attempt to fit a model to the observed hydrograph, complete the calibration tasks below:

  1. If you're unfamiliar with numerical calibration, please read What is calibration?.
  2. Click on the tab Model Calibration. A model labeled "Bore_6416", which was built above, should be listed within the calibration table and model bore ID and the start and end date of the observed hydrograph should be shown within columns 3 to 5.
  3. Set the start and end date of water level observations that are to be used in the calibration using columns Calib. Start Date and Calib. End date. Selecting, say, a calibration end date prior to the end of the observed record omits the observations beyond this end date from the calibration. Importantly, once the calibration is complete HydroSight predicts the omitted water level observations. This provides a valuable assessment of the model predictive reliability. In this step, change the calibration end date by placing the cursor in the cell. This should open a calender window. Use it to set the end date to 1st January 2009. This date was selected to assess if the calibrated model can predict the recharge event in 2010.
  4. HydroSight has multiple calibration schemes built in (see HydroSight Calibration Schemes). Use the column Calib. Method to set the method to _"SP-UCI".
  5. Select the model using the left checkbox and then click the button Calibrate Selected Models to open the following calibration window. The above HydroSight Model Calibration window allows the user to input settings for each calibration method and, once the calibration is running, to observe the iterations. Below is a summary the red labels items:
    1. The settings for each calibration method is shown as a separate tab. Only those calibration methods that are used in at least one of the selected models are active.
    2. For each calibration method, the settings are shown on the right. The default methods should produce a reasonably robust calibration.
    3. The buttons start the sequental calibration of all selected models. Note, the button _HPC Offload _ and HPC Retrieval are only shown when HydroSight is ran from within MatLab.
    4. Once the calibration is started, the settings and parameters to be calibrated are shown followed by results from each iteration.

Calibration GUI

  1. To start the calibration, click on Start calibration. Importantly, global calibration requires 100,000s of model runs. Within HydroSight this is undertaken using all of the CPUs on your PC. Despite this the calibration can take between multiple minutes and hours depending upon the PC and the model form.

Assessing the Model Calibration

There is no receipt book for assessing the calibration of any environmental model. However, it is good practice to assess if the model can make reliably predict observations omitted from the calibration and if the internal mechanisms of the model are sensible. HydroSight provides the tools for such analysis (and much more). Below, tools for the two aforementioned tasks are outlined. Used them to assess if the predictions are bias and if the internal model fluxes and stores are sensible.

Assess the Model Predictions

The screenshot below shows the Calib. Results tab. The red numbers denote the following three important aspects:

  1. There are five tabs for exploring different aspects of the calibrated model.
  2. All of the observation data, calibration and prediction results are shown in the table. To save the data, right click inside the table and use the drop-down menu.
  3. The model fit to the observed hydrograph can be explored in many ways. The right drop-down allows for plotting of the results in many ways.

Calibration results tab

Assess the Model Fluxes

The screenshot below shows the Forcing Data tab. This tab allows examination of all model forcing data and, more importantly, all model fluxes such as those from a soil moisture model. The red numbers denote the following six important aspects:

  1. The forcing data shown in the table can be presented at various time-steps such as annual, monthly or daily, using the drop-down.
  2. When the forcing data is up-scaled from daily the up-scaling can be undertaken in many ways, for example the mean, miniumum or maximum value within the time-step can be calculated. Use this drop-down to explore a range of time-steps.
  3. The time span for the data shown in the table, and plotted, can be set using the text boxes. The format for the dates is DD/MM/YYYY.
  4. When the format for the forcing data table is decided upon, multiple types of plots can be produced. Use this drop-down to select a plot type.
  5. Each column within the forcing data table can be plotted. Use this drop-down to specify the data to be on the x-axis.
  6. Similar to item 5 above, the y-axis data can be specified using this drop-down. Once specified, click on Build model to plot the data.

Calibration forcing data analysis window

Model Simulations

Clone this wiki locally