Pumping rate estimation
The forcing transformation function pumpingRate_SAestimation allows the downscaling of infrequent (>daily) groundwater pumping data to daily, weekly or monthly rates. This extension to HydroSight was funded by the Department of Environment, Land, Water and Planning. The function is to be used with the [https://github.com/peterson-tim-j/HydroSight/wiki/Transfer-function-noise-(TFN)-model](Transfer function noise (TFN) model).
The downscaling is required when pumping is metered, say, only annually and when there are unmetered periods. The downscaling is undertaken by simulating the pump as being either on or off for each user-set downscaling time step. When the period is metered then the number of days pumping is used to estimate the mean daily pumping rate. The median of all metered periods is then applied to non-metered periods. This approach requires no assumption of the behaviour of groundwater users or their response to climatic conditions.
Importantly, in assessing if a time-series of pump state produces a TFN model in agreement with the observed head, the number of combinations of pump state over, say, a decade far exceeds the largest number possible on a 64 bit machine. The implications are that the calibration cannot assess all combinations and that if an identical model is re-calibrated an alternate optima may be identified, which may have a slightly different time series of pump state and model parameters.
To increase the reproducibility of the calibration, detailed trials were undertaken to quantify the reproducibility of the solution for a range of SP-UCI calibration settings. The trials suggest that the following SP-UCI calibration settings should produce a reasonably reproducible solution without requiring exceedingly long calibration time:
- Minimum number of total complexes (NGS) and the Maximum number of total complexes (NGS) set to 16. This will result in a total of 16 complexes always being used indifferent of the numer of model parameters or number of pumping wells.
- Percentage change in the objective function allows in kstop loops before convergency (pcento) set to 1e-4.
Additionally, the following modifications can be made to the model structure to reduce calibration time:
- Reduce number of years within the forcing data file. That is, reduce the number of years before the first water level observation and substantial groundwater extraction.
- Reduce the number of pumping wells used in the model. Trials for the Warrion Groundwater Management Area suggest that a radius of influence of 2km is sufficient; but obviously this is site dependent and should be assessed for each site.
The different metering scenarios are defined from the TFN input forcing data file. Any pumping period that was infrequently metered is defined by -999 values in the forcing data file followed by the total extraction volume over the preceding period of -999s. A period with no metering is defined by a period of -999 values ending with a zero pumping rate.
To illustrate the input forcing data requirements, the table below shows that Pump_1 has two periods of -999. The first period (1/3/1990 - 1/6/1990) ends with a zero value. This is interpreted as the preceding 4 days as being unmetered. Over this period the calibration will examine if the pump should be on or off and the median daily metered usage (from the metered periods) will be assigned to each day that the pump is deemed to be on. Over the second period of -999s, it end with a metered volume (always in units of m3). This is interpreted as the total pumping volume over the preceding 5 days without metered usage. Again the calibration will assess if the pump is on or off. If the pump is deemed as being on for all of the six days (includes 1/14/1990) then the daily usage will be 20 m3/day. If however the calibration deems that the pump is only on for 3 of the 6 days then the usage will be 40 m3/day for each day that the pump is on.
| Year | Month | Day | Precip | PET | Pump_1 |
|---|---|---|---|---|---|
| 1990 | 1 | 1 | 0 | 3.4 | 0 |
| 1990 | 1 | 2 | 1 | 4.2 | 0 |
| 1990 | 1 | 3 | 12 | 4.3 | -999 |
| 1990 | 1 | 4 | 0 | 5.2 | -999 |
| 1990 | 1 | 5 | 0 | 5.2 | -999 |
| 1990 | 1 | 6 | 0 | 5.2 | -999 |
| 1990 | 1 | 7 | 0 | 5.2 | 0 |
| 1990 | 1 | 8 | 0 | 5.2 | 0 |
| 1990 | 1 | 9 | 0 | 5.2 | -999 |
| 1990 | 1 | 10 | 0 | 5.2 | -999 |
| 1990 | 1 | 11 | 0 | 5.2 | -999 |
| 1990 | 1 | 12 | 0 | 5.2 | -999 |
| 1990 | 1 | 13 | 0 | 5.2 | -999 |
| 1990 | 1 | 14 | 0 | 5.2 | 120 |
| 1990 | 1 | 15 | 0 | 5.2 | 0 |
When building this model component, there are also the following options:
-
Downscaling time-step: Defines the highest frequency time step over which infrequent metering will be downscaled. The options are:
- 0: no downscaling. Over each period of -999s the pump will be modelled as either on or off over the whole period.
- 1: Monthly downscaling will be undertaken for each period of -999s. If the period is shorter than a month then the pump will be modelled as either on or off over the entire period. If a period of -999s is longer than a month then the modelling will examine turning the pump on or off for each of the months.
- 2: Weekly downscaling will be undertaken for each period of -999s.
- 3: Daily downscaling will be undertaken for each period of -999s. Peterson and Western (2019) shows that, for annual metering, downscaling to daily is unreliable. This option should only be used when the usage metering is more frequent than annual, e.g. monthly.
- Minimum number of bore meter readings: Defines the minimum number of meter readings over the entire record of pumping for the -999 periods to be downscaled. If a pump does not meet this criteria, then it is omitted from the modelling. Default is 2.
- Minimum days to downscale: Defines the minimum duration of any unmetered -999 period for it to be downscaled. That is, if a period of -999 ends without a meter reading for the preceding period and the duration is less than this criteria, then the pumping over the period is set to zero and the period is not downscaled. If the period of -999s ends with a metered volume, the it is downscaled. Default is 7 days.
The GUI includes tools to assist with the efficient selection of production bores for downscaling. To access the tools, place the cursor within the Input data cell for the pumpingRate_SAestimation function row within the model options in the Model Construction tab. Doing so should show a list of all non-date columns within the input forcing file. Right-clicking within the bottom table and selecting Wizard... to open the window shown below. It can be used to select a maximum number of pumping bores within a user defined search radius. Once you are happy with the selected bores, right click again and select Push "Required data" to component to assign the bores to a prior defined weighting function component.
The diagram below summarises the calibration algorithm for the downscaling and estimation of infrequently metered pumping. Importantly, the approach is only to be used with the SP-UCI calibration scheme. The algorithm has not been incorporated into the other calibration schemes.
