Waterhole persistence

The model operates on a daily time step. For each day, the volume of a waterhole is determined based on net inflows (rainfall and runoff, groundwater inflow) and outflows (evaporation, seepage and anthropogenic extraction). The waterhole volume is converted to a waterhole depth based on a look-up table derived using surveyed or estimated waterhole bathymetry data. The waterhole volume/depth determination is conducted only on days when there is no flow recorded in an associated daily flow record. When flow is recorded (i.e. a value in the daily flow time series of anything other than zero), the waterhole is reset to full on that day and remains full until flow is next zero, when the water balance model is again activated. The output is a daily time series of modelled waterhole volume, depth, area and groundwater volume.

The volume of a local groundwater store surrounding the waterhole is also calculated at a daily time step. The groundwater store is topped up by rainfall and contributes a percentage of its volume to the waterhole volume each day as groundwater inflow. Similarly, a daily volume is lost from the local groundwater store to deep drainage. These parameters can be set to zero if there is no groundwater influence on water balance in the waterhole being simulated.

Each component of the water balance model has a combination of input data and parameters by which the input is transformed. Parameters must be entered to establish the model, and these can be set to zero to exclude their influence, estimated based on field observations of the particular waterhole or approximated. The parameters may be altered to calibrate the model to a specific waterhole (See Figure 1, above)

Optionally, parameter values can be optimised against observed daily waterhole depths within the calibration module of the software (see Section ‘Optimisation’ below). Once established by setting parameters and saving the model, a waterhole model can then be run using simulated flow time series data to assess the effects of different flow management scenarios on waterhole depth and drying frequency. Note: If measured depth data for calibration are unavailable, the calibration step can be skipped and a simple, uncalibrated model built (Table 1).

Additionally, a ‘run to empty’ component exists, where you can customise relevant parameters and run the model in a no-flow state. This can be used to determine the total drying time for a calibrated or uncalibrated waterhole. See the ‘Run to empty’ section below for further details.

Table 1. Input files needed to build water loss model for calibration

Model Input	Definition	Notes
Recorded depth	Daily depth logger time series data	If depth data is unavailable then the calibration step can be skipped and uncalibrated model built
Flow	Gauged flow daily time series in ML/day	Information needed for calibrated and uncalibrated model.
Rainfall	Rainfall time series in mm/day	Information needed for calibrated and uncalibrated model.
Evaporation	Evaporation time series in mm/day	Information needed for calibrated and uncalibrated model.
Bathymetry	Digital Elevation Model – Datum, depth, volume and surface area data	Information needed for calibrated and uncalibrated model.

Model Calibration

A work area is provided in the model builder interface to conduct a parameter calibration process. Simply click the ‘Calibrate’ button within the relevant site to access this tool. Where direct field measurements of daily waterhole depth have been taken, these can be loaded and the predicted vs measured depth values can be compared. If waterhole depth data were collected at a time-step more frequent than daily (e.g. using an automatic depth logger or stream gauge), the software automatically converts this to a daily time series. The user can then interact with specific parameters to manually calibrate the model by observing the similarity between plots of observed and modelled waterhole depth on a built in graph, and by changes in the many available fit statistics. All available fit statistics (NNSE, RMSE, R^2, PBIAS and RSR) contain tooltips providing details on their use and value range. Additionally there is a genetic optimisation algorithm included in the software, to automate the calibration process. This optimisation algorithm has a handful of settings, including the fit statistic to optimise on. A definition of a genetic optimisation algorithm is provided in the Optimisation section below.

Run to empty

A work area is provided in the model builder interface to conduct a run to empty analysis. This section can be accessed via the ‘Run to empty’ button visible on a waterhole site. This tool runs a modified version of the model, which computes waterhole depth without considering re-filling measures. This allows the total drying time of a waterhole to be determined. This interface includes relevant parameters, such as the evaporation type to use (the main component in waterhole drying). Available options are:

• Data
• Mean
• Value
• Data + Value
• Data + %

Additionally, a run to empty depth (m) can be specified (default 0). Once run, the total days for the waterhole to deplete from full to the specified empty depth is noted. Additionally, a graph of the modelled depth and evaporation over time is provided. See the image below for an example of this interface.

Calibration input data

The following section outlines key points to consider for each input file needed to develop the water loss model (See Table 1, above).

FLOW

Obtain gauged flow data (ML/day) from appropriate source (i.e. Qld water monitoring information portal, NSW real time data) and set up a table format with – date and daily flow. Before uploading input file, check for the following:

• Gaps in flow data used for calibration should be replaced with 0. Do not use -9999 for this purpose. The model identifies any non-zero days as flow days and automatically resets waterhole volume to full, which is not an appropriate response to flow data gaps during a drying phase. If gaps are long, flow events could be missing, so it may be better to exclude the dates from the calibration.
• If you think the flow data does not differentiate between zero flow and very low flow, you may wish to set a cease-to-flow discharge threshold larger than zero ML/day. In most circumstances, it is probably best to use zero for calibration. Local hydrological knowledge would be needed to set a value other than zero.

RAINFALL

Obtain rainfall time series in mm/day from:

• nearest BOM weather station,
• SILO Data Drill interpolation (http://www.longpaddock.qld.gov.au/silo/datadrill/) based on location of the waterhole; or
• local information.

SILO is generally used as it is an enhanced climate database, constructed from observational records provided by BOM. Historical records are from 1889 (current to yesterday). If the weather station is unlikely to geographically represent the locality of the waterhole, try to obtain local information. This version of the plugin has been modified to reflect any rainfall event that falls within the local catchment area in which the waterhole is located, though some rainfall events may be un-gauged.

EVAPORATION

Obtain daily evaporation time series in mm/day from:

• nearest BOM weather station, or
• SILO Data Drill interpolation (http://www.longpaddock.qld.gov.au/silo/datadrill/) based on location of the waterhole.

As with rainfall, the weather station should geographically represent the locality of the waterhole. Note: The evaporation rate (adjusted by a scaling factor, see below) is applied across the surface area of the waterhole. The groundwater store is also subject to a proportion of the surface evaporation rate when it is close to full.

BATHYMETRY

This data file includes datum, depth, volume and surface area data at water height intervals through the waterhole (Table 2). Data is based on a DEM for the specific waterhole or from an estimated waterhole shape where this data does not exist. 10cm depth intervals are often used however the resolution may vary depending on the available data and the precision of outputs required. A smaller resolution (in depth) is preferred to decrease the interpolation error when reading the Bathymetry plane table (Table 2). The bathymetry data provided can be views in the optimisation interface (‘Toggle View’ button)

• Bathymetry table format: column header names do not matter, but the order of columns does –water level (m; see point below), water depth (in m; CTF = maximum depth, empty = 0), volume (m3), area (m2; this is the 2D area from ArcMap, i.e. the waterhole surface area, not benthic area which is the 3D area in ArcMap).
• The first column contains the water level relative to a height datum. Where the bathymetry has been surveyed, this may be a standard altitude datum such as AHD. Often, as in Table 2, actual elevation is unknown in which case an abstract datum (e.g. waterhole level at CTF = 100 m) can be used.
• The bathymetry table can be viewed by clicking the ‘Bathymetry’ tab on the graphing pane.

Table 2: Example of bathymetry plane table. DEM developed with 1cm depth intervals.

Waterhole level (RL(m))	Depth (m)	Volume (m3)	Area (m2)
100	2.69	445.22	869.3
99.99	2.68	377.62	635.79
99.98	2.67	315.4	609.98
99.97	2.66	255.56	586.84
99.96	2.65	198.17	560.67
99.95	2.64	143.82	524.24
99.94	2.63	94.84	445.82
99.93	2.62	56.4	320.11
99.92	2.61	30.06	213.49
99.91	2.6	12.72	135.29
99.90	2.59	3.1	57.37
99.89	2.58	0.61	8.92
99.00	2.57	0.07	2.64
98.99	2.56	0	0

DEPTH LOGGER DATA

Consider the following measures when preparing the depth logger data as an input file:

• Data file format: column header names do not matter, but the order of columns does – date (dd/mm/yyyy) then depth (in metres) (Table 3).
• The frequency of measurements must be at least daily. Where the time series contains measurements at more frequent intervals, pre-processing is not required; Eco Modeller will aggregate measurements as necessary into daily averages for modelling.
• Depth logger data must be pre-processed to relate it to the bathymetry datum. Because the depth logger is unlikely to be positioned right at the bottom of the waterhole or at exactly the deepest point, depth logger data should be adjusted so that the depth logger cease-to-flow (CTF) level matches the depth of the deepest point on the digital elevation model (DEM) at CTF level. e.g. If DEM maximum depth at CTF is 3.5 m, and depth logger level is 2.5 m, you will need to add 1m to all depth logger data points to relate the two sets of data.
• Waterhole empty = 0m, waterhole at CTF = maximum depth (e.g. 3.5m). (It is wrong to adjust depths for CTF = 0 and depths below this, negative values).
• The calibration start and end date fields should be used to select a suitable period of data without gaps when performing the calibration.

Table 3. Example of recorded depth data

Date	Logger Level (m)
4/02/2006	2.676512821
5/02/2006	2.661444444
6/02/2006	2.644722222
7/02/2006	2.632236111
8/02/2006	2.624666667
9/02/2006	2.613736111
10/02/2006	2.619611111
11/02/2006	2.610791667
12/02/2006	2.601680556
13/02/2006	2.591013889
14/02/2006	2.580138889
15/02/2006	2.572319444
16/02/2006	2.649430556

Model Parameters

By entering values characterising the water input and loss rates at the waterhole of interest, the basic waterhole model can be calibrated to accurately predict waterhole depth/volume over time, based on real or modelled flow, rainfall and evaporation time series. A list of all calibration parameters is provided in Table 4. When a new waterhole model is created, it contains default parameter values, which are broadly based on calibrated test waterhole models from the Moonie and Border Rivers in the Queensland Murray Darling Basin. These default values provide a rough guide (at the order of magnitude scale), but measures or estimates of the parameters specific to the waterhole of interest are likely to be required in order to develop a reasonable model. However, because a single figure is used for parameters that are likely to be highly variable and influenced by a range of factors and interactions, there is unlikely to be a precise ‘correct’ value for some (e.g. infiltration). By estimating a value and then using the optimisation, a reasonable figure can be reached for the purposes of building the model.

As the calibration of parameters is limited by the amount and quality of data available, it is advised to start with a parsimonious calibration (i.e. all calibration parameters are set to zero) and add complexity if needed. If there is little or no usable data to calibrate a parameter, then start the calibration process by assuming that the evaporation multiplier is one and all other parameters are zero.

If there is data, then start the calibration process by first modifying the evaporation multiplier and then modifying another parameter, one at a time. Which additional parameters to modify will be dependent on what type of recession events are observed during calibration. Points to consider for calibrating a parameter are provided below.

Table 4. Summary of calibration parameters

Function	Calibration Parameters	Definition
Time series	Calibration start/end date	Selects which parts of time series to use for calibration of model parameters.
Gauges	Flow Lag	Shifts the calibration period when flow will impact the waterhole.
Evaporation	Evaporation Scaling	An adjustment factor for values in the evaporation time series, which accounts for additional sources of water input or loss, and differences in actual evaporation rate due to estimation/interpolation error and local fine-scale factors (e.g. due to waterhole micro-climate, orientation, shading etc.).
Seepage	Seepage Rate (mm/d)	The amount of water lost from the base of the waterhole into deep drainage. A daily seepage rate (mm/d) is specified for the waterhole and is applied across the water surface area.
Direct Rainfall	Local catchment area (m2)	Considers the surface area of the waterhole’s local catchment that might contribute to inflow to the waterhole.
	Maximum daily infiltration (mm/d)	Sets the threshold of a rainfall event beyond which all rain enters the waterhole.
Groundwater	Local GW catchment area (m2)	Sets the size of the larger area around the waterhole in which water can be temporarily stored following a rain event, gradually contributing to the waterhole over a period following that event.
	Inflow to waterhole (%)	Percentage of the groundwater that reaches the waterhole.
	Lost to deep drainage (%)	Percentage of the groundwater that is lost to deep drainage rather than reaching the waterhole.
Extraction	Commence pumping depth (m)	Waterhole depth determining start of pumping period.
	Cease pumping depth (m)	Waterhole depth determining end of pumping period.
	Extraction rate (ML/d)	The volume of water that may be extracted in ML/day.
	Maximum annual take (ML)	The total volume of water that may be taken in a water year based on pumping licence conditions, in ML.
	Water year start month	Sets start month of the water year, within which the annual take limit applies.
	Cease-to-flow threshold (ML/d)	Waterhole volume at cease-to-flow.

Generally there is a change in slope of the recession when conditions go from “flowing” to “evaporating”. The recession rate for an evaporating waterhole is less and should approximately match the “theoretical” evaporation rate. If the rate of modelled water loss appears higher than the evaporation loss rate, check whether the cease-to-flow level used is too high. If necessary adjust it, to ensure that water loss during the falling limb of the flow event hasn’t been included in the calibration data.

1.CALIBRATION START AND END

Points to consider when defining the calibration period:

• The default date range is the period of the depth logger data time series, however use of a specific subset may be desirable.
• Setting these to correspond to the start and end of a long drying sequence in the depth data may be helpful to calibration.
• If long gaps in flow or depth data had to be interpolated, it may be best to select a period that excludes these.

2. FLOW LAG

Points to consider if flow lag (days) is used as a calibration parameter:

• If the waterhole of interest is distant from the nearest flow gauge, such that flow events are likely to occur at the waterhole several days before or after they are gauged, including an estimate of the flow lag (in whole days only) may improve model calibration. Set the lag to zero if flow events at the gauge occur at the waterhole on the same day.
• A positive value is used when the waterhole is downstream of the gauge, to push forward the date of flow events, and a negative value is used when the waterhole is upstream of the gauge.
• To identify if flow events should be omitted from the time series, look for unexpected “rises” in the recession. Note that these ‘rises’ could also be releases from an upstream dam, localised rainfall event causing some small streamflow or could be a sensor problem.
• Also look for unexpected “falls” from the time series that could be either a sensor error or someone pumping from the waterhole.

3.EVAPORATION SCALING

Points to consider when adjusting evaporation scaling:

• Values in the evaporation time series are multiplied by the ‘evaporation scaling’ factor to increase or decrease the influence of evaporation. A factor of 1 leaves values in the time series unchanged.
• Adjust the evaporation multiplier, and associated parameters, if unrealistic, i.e., maybe if multiplier is not in the range 0.5 – 2.0 (approximately). For a very large multiplier, set to a more reasonable value and consider modifying seepage. For a very low multiplier set a reasonable value and modify infiltration. If there are still differences, especially faster recession at high waterhole levels and slower recession at lower levels then modify groundwater connection parameters.
• Because the scaling factor acts like a ‘black box’ for balancing a range of influences, it is unlikely to be known exactly. Optimisation should be used to select the most appropriate value (see Optimisation, below).

4.SEEPAGE RATE

If evidence suggests that seepage is unlikely, this parameter can be set to zero. Otherwise, consider the following points when adjusting seepage rate:

• If water loss rates are notably higher than expected by evaporation, or if they vary significantly between nearby waterholes, seepage may be responsible.
• The rate of seepage can vary spatially and temporally and is influenced by a range of factors such as the levels of surface and ground water, the shape of the river channel, geological features and the nature and thickness of river bed deposits (Jackson 2005).
• There are a range of methods for field estimation of seepage that could be used to inform the parameter value. These include measures of water quality (e.g. using isotopes; observing changes in conservative ion concentrations with changing water volume during the drying phase), use of piezometers, monitoring of temperature fluctuations in the sediments underlying waterholes.

LOCAL CATCHMENT AREA

The waterhole surface itself, along with the surrounding banks and slopes that contribute direct run-off during a rain event (Figure 2).

• 100% of rain that falls on the waterhole surface is added to the waterhole volume. The waterhole surface area on each day is identified from the bathymetry table.
• If a number smaller than the waterhole surface area is entered (e.g. zero), then there is no contribution of runoff from the surrounding banks, and only rainfall on the waterhole surface is included in the model.
• If a local catchment area larger than the waterhole surface area is specified, rain that falls on the difference in area between the two either infiltrates into the soil (see Maximum daily infiltration, below), or runs off into the waterhole (Figure 2).
• The area of surrounding banks and slopes may be estimated, surveyed in the field or identified from satellite images/aerial photos (such as using Google Earth). Alternatively, if it is considered unimportant to the water balance, it can be left out of the model (set to zero).
• If there are “spikes” in the recession of the time series that coincide with rainfall events then you might need to increase the local catchment area.

Note: that sometimes the local rainfall station might not capture all rainfall events (especially if the only rainfall station is some distance away). In this case it might be difficult to calibrate to all rainfall events.

6.LOCAL GROUNDWATER (GW) CATCHMENT AREA

This parameter represents the broader area around a waterhole that may catch and temporarily store rainwater as soil moisture or shallow aquifers, a portion of which can then be laterally transmitted into the waterhole following the rain event (Figure 2).When the groundwater store is particularly full (>80%), it is also subject to evaporation from the soil surface at 50% of the surface rate.

• The volume of the local groundwater store is defined as the ‘Local groundwater catchment area’ times the maximum waterhole depth times a void ratio (the relationship between the volumes of empty space and solid components of the soil) of 0.4.
• The ‘Local GW catchment area’ value entered should include the ‘Local catchment area’ and waterhole surface area within it; Eco Modeller will subtract these areas out of the model when calculating the water balance.
• If the GW catchment area entered is less than the waterhole surface area (e.g. zero), GW inflow will not be included in the model. GW can be excluded in this way for waterholes known or suspected not to have GW inputs.

7.MAXIMUM DAILY INFILTRATION

Determines the quantity of rainfall that enters the local groundwater store in a 24 hour period.

• Rain falling on the ‘Local catchment area’ first infiltrates into the soil. Once the specified maximum daily infiltration threshold has been reached, any additional rain on this area enters the waterhole as runoff.
• Rain in excess of the maximum daily infiltration rate that falls on the ‘Local GW catchment area’ does not enter the waterhole.
• A range of factors are likely to influence the infiltration rate, including the intensity of rainfall, the physical properties of local soils, the antecedent soil-moisture level, the slope, density and type of vegetation and land use (USDI & USGS 2015).

8.INFLOW TO WATERHOLE

The proportion of the volume of the temporary groundwater store that flows laterally into the waterhole each day. This creates a lagged influence of rainfall, where water inputs continue in the days following an event.

Figure 2: The relationship between the ‘Local catchment area’, ‘Local GW catchment area’ and the waterhole water balance.

9.LOSS TO DEEP DRAINAGE

The proportion of the local groundwater store that seeps down into underlying deep groundwater aquifers each day, and is no longer relevant to the waterhole water balance.

10.EXTRACTION

It is unlikely that actual water extraction will be regular and well-known enough for inclusion in the model calibration. It is recommended that these parameters, namely pumping, should only be adjusted if aspects are known and appear to be influencing measured water loss. A characteristic of pumping can be unexpected “drops” in the recession of the time series depth data. This unexpected recession behaviour can also be a sensor problem.

Where usage is clearly understood and likely to influence the calibration, pumping characteristics may be included. The inclusion of extraction behaviour within the model should be considered for scenario modelling.

Commence/cease pumping depth: Waterhole depth determining start and end of pumping period.

• CTF = maximum waterhole depth; empty = 0; 0.5m below CTF = max depth – 0.5.

Extraction rate: The volume of water that may be extracted in ML/day.

• If commence/cease pumping thresholds are to be applied, a pumping rate must be specified. For example, in Queensland Water Resource Planning, daily extraction rates are often indicated in the Resource Operations Plan (ROP) for a catchment (Table 5), otherwise a pump size may be specified (Table 6). The pump size table (Table 6) may be used as a guide to estimate an extraction rate from waterholes when specific limits are unknown.

Maximum annual take: The total volume of water that may be taken in a water year (see below) based on pumping licence conditions, in ML.

• A value for maximum annual take must be specified in order to apply extraction parameters – if left as zero then no extraction will occur, even if depth thresholds and pump rates are set. If annual usage is unlimited, enter a very large number (i.e. several orders of magnitude larger than the maximum waterhole volume from the bathymetry table) to ensure the limit does not kick in and affect modelled extraction.

Water year: Allows the user to set start month of the water year, within which the annual take limit applies.

• In Queensland, the water year is often defined as July to June because the southern hemisphere summer is the ‘wet season’ for many regions, but this may differ depending on the seasonality of climate or infrastructure operation in a region. The intent of setting a water year is to avoid splitting the wettest season of the year between consecutive years. So in a southern hemisphere river with winter rain dominated hydrology, it may be January to December. Information about the water year for a particular Queensland plan area can be found in the ROP or by consulting a local hydrologist.

Table 5: Example of the extraction rates and volumetric limits specified for nil flow entitlements (i.e. for extraction from waterholes during periods of no flow) in the Queensland Warrego, Paroo, Bulloo and Nebine Resource Operations Plan.

Table 6. Pump size and maximum extraction rates; example from The Queensland Warrego, Paroo, Bulloo and Nebine Resource Operations Plan.

REFINING THE CALIBRATION

Once the starting model parameter values/estimates have been entered, they can be adjusted to improve the accuracy with which the model represents the measured depth time series. The graphing pane displays the depth time series from both measured field data and predicted model data to give a visual guide to the performance of the model. The in-built optimisation function can be used to calibrate the model, however best results will be achieved when parameter values are first manually adjusted. Expert opinion should inform which parameters are tweaked and by approximately how much, so that the model is meaningful. Trial and error is then used, making a series of small changes and observing the improvements in model fit each time (see Assessing the quality of the model, below). Once satisfied that a good fit has been achieved manually, the optimisation function should be applied to some or all of the model parameters. Note: water loss models should be re-calibrated if more appropriate data is obtained i.e. depth logger data in which the depth of the waterhole has dropped to a lower level

OPTIMISATION

The Eco Risk Projector waterhole model includes a genetic algorithm function that can be used to optimise the model. The check boxes next to each parameter entry field are used to select which parameters to include in the optimisation. Any values that are unknown or estimated should be included in the optimisation. Where confidence in an entered value is high (e.g. where it is based on a direct measurement) it may be preferable to exclude it from the optimisation, so that it doesn’t get modified.

Genetic algorithm optimisation is a machine-learning technique based on the principles of biological evolution, where possible combinations of solutions, like genomes, evolve over a number of generations. At each generation, solutions can ‘reproduce’ (i.e. parts of solutions are exchanged between two individuals) or mutate, and the ‘fittest’ resulting solutions are selected to contribute to the subsequent generation. The parameter values from the best solution in the final generation are selected for the model. After the optimisation process is completed Eco Risk Projector will report the fit statistics for the final parameters.

OPTIMISATION SETTINGS

Population: The number of possible solutions generated. The default setting is 10, meaning that each generation contains 10 individual sets of parameter values. The values in each individual solution are based on, but modified from, the entered calibration starting values.

Generations: The number of times that selected individual solutions are recombined and modified (via crossover or mutation) to develop a new population of possible solutions.

• Larger numbers of generations provide more opportunities to select an optimal solution, however computation time can be considerable and returns diminish. Trials using waterhole test models suggested that very little improvement was gained beyond 15 generations.

Scaling: The scaling factor sets the acceptable bounds of change in parameters for all populations, based on the starting value entered by the user. e.g. a scaling factor of 50 means that the parameter values used in the optimisation solutions can be up to 50% larger or smaller than the calibration starting values. The more confident you are that the starting values are approximately correct, the smaller the scaling you apply.

Fit statistic: The fit statistic to use for assessing the accuracy of a parameter set. If in doubt NNSE is a good option for this, however other options (such as PBIAS) might be worth considering depending on the context of your optimisation.

Assessing the quality of the calibrated/optimised model

A range of fit statistics are reported each time a calibration or optimisation is run as an indication of the performance of the calibrated model. The available options are:

• NNSE – Normalised Nash Sutcliffe Efficiency Coefficient. This is effectively an R^2 value bounded between 0 and 1, where 1 is a perfect fit and 0 is a fit worse than linear. Any value above 0.75 is considered very good.
• RMSE – Root Mean Squared Error. The RMSE is a measure of the differences between values in the measured and predicted depth datasets. The RMSE is in the same units and scale as the value on the y-axis (depth), with a RMSE of zero indicating no difference between observed and predicted values. The smaller a RMSE, the better, however because the units are model-specific, there is no universally acceptable level.
• R^2 – Coefficient of Determination. This is a value bounded between -inf and 1, where 1 is a perfect fit and -inf is worse then linear. Any value above 0.7 is considered very good.
• PBIAS – Percentage Bias. This is a highly informative measure, which ranges from -inf to inf. A value of 0 indicates a perfect fit, whilst a negative value indicates overestimation (modelled depth is larger than observed) and a positive value indicates underestimation (modelled depth is less than observed). Any value less than |10| is considered very good.
• RSR – RMSE Standard Deviation Ratio. This is effectively a normalised RMSE value, which ranged from 0 to +inf. A value of 0 indicated a perfect fit.

All options should be considered to get a complete picture of the model accuracy.

Optimisation settings

Model parameters: once saved, the calibrated model parameters are duplicated in the fields on the waterhole site.

Model input and output time series: By clicking on the save icon download all results (both time series csv and fit statistics json). By right clicking on the graph you can save an image of the current view.

Uncalibrated models

If a measured depth time series for calibration is unavailable, the calibration step can be skipped and a simple, uncalibrated model built by linking the required input data time series and water extraction parameters and estimating the water balance parameters. See the ‘Calibration parameters’ section above for information about setting each of the model parameters.

Note: there is no way to assess the quality of an uncalibrated model.

Building a calibrated model

Once the model has been calibrated, the resulting parameters are duplicated in the site, ready for application to flow scenarios. The input data and the water extraction parameters may differ between the model calibration and the model build.

Input data

Input time series will be duplicated on the calibration and site pages. The rainfall, and evaporation time series and bathymetry table are likely to be the same in both instances, however, depending on the desired application, the flow time series may be different. The flow used for calibrating the model will usually be gauged flow data concurrent with the depth logger data. The flow time series used for building the calibrated model is more likely to be modelled daily flows (such as an IQQM management scenario model).

You will need to ensure the date range of all input time series needed for running the model (flow, evaporation and rainfall) are concurrent.

It is also important to be aware that some hydrological models are poor at differentiating zero flow from very low flow and they may not give stable predictions at very low flow levels. This means that when running daily flow sequences from such models a cease-to-flow threshold greater than zero may need to be applied to represent no-flow. For example, Queensland IQQM models are not accurate at very low flow levels, so a threshold of 2 ML/day is applied to represent cease-to-flow, meaning any daily flow less than 2 ML/day is assumed to be zero flow. If a value other than zero needs to be used to represent cease-to-flow, enter this value in the ‘Cease-to-flow threshold’ parameter field, otherwise, leave it as the default value of zero. Advice from local hydrologists should inform this parameter. Important: check linked flow time series before running the model.

In some situations there may be an interest in modifying the bathymetry table before running the model. An example may be if you wish to simulate the influence of fine sediment accumulation on waterhole persistence. If field measurements indicate that fine sediment has accumulated in the waterhole and you want to simulate water depth over time both with and without this sediment, a new bathymetry table representing the ‘natural’ size and shape of the waterhole in the absence of fine sediment could be derived by manipulation of the waterhole DEM. This may then be linked to the waterhole model in place of the ‘current state’ bathymetry used for model calibration. Run outputs from the ‘current state’ and ‘natural’ bathymetry models could then be compared to assess the interaction between waterhole condition and flow management scenarios on the drying regime of the waterhole.

Water extraction parameters

Extraction from waterholes is more likely to be applied in the model builder, than during calibration. See notes in the Model calibration section (above) for further information about setting water extraction parameters. When extraction is likely, but rate is unknown, the model could be run several times using different extraction rates to examine the effect. This approach can also be used to simulate the effects different pumping rates and rules have on waterhole depth and persistence.

Running the model

Once the calibrated or uncalibrated waterhole model has been built, the run button can be used to toggle on/off scenarios, then run the model. The date range of the model run will be determined by the period of concurrent dates in all relevant time series, or limited by the input in the run period box.

Assessment method

Assessment for waterholes is failure based. Daily failure based on the daily modelled depth (I.E. depth <= 3 meters is failure). Annual failure based on a number of daily failures (I.E. >= 10 daily failures in a year is a failure). Temporal failure based on sum of annual failures (I.E. 2 failures in 3 years is a failure).

Spatial aggregation is then applied to determine the % of simultaneous failures and determine a risk level (I.E. high risk if >= 50% of sites fail simultaneously)

Inputs

• Daily flow data
• Daily evapotranspiration data
• Daily rainfall data
• Bathymetry data
• Daily depth data (for calibration)

Outputs

• Daily waterhole modelled depth, volume, area and ground water volume
• Daily time series of assessment results
• Yearly time series of assessment results
• Temporal time series of assessment results
• Spatial time series of assessment results

User interface