Practice note: Calibration of Crop Water Models

This practice note is one of a set developed to provide consistency and transparency of river system models being used within the Murray–Darling Basin. The notes cover modelling practices, such as naming conventions for folder structures, to model methods, such as for flow routing and residual inflow estimation, and have been developed through a collaboration between the MDBA and Basin States.

Produced in collaboration with:

This practice note, 'Calibration of Crop Water Models', describes the general principles and a high-level method that should be adopted when irrigation demands are to be estimated via crop water models that maintain a soil moisture account.

Background

Irrigation demands may be a significant component of the reach water balance and generation of these demands is important for understanding the total surface water diversions. How irrigation demands are represented in the final model calibration and scenario modelling will depend on the initial model conceptualisation. In some situations, irrigation demands will not be represented using crop water models but by using metered diversions or a time series or pattern. Modelling the on-farm water balance (see Practice Note 'Estimation of Unmetered Irrigation Diversions (On-Farm Water Balance))' provides an understanding of the different sources of water available for irrigation. Depending on the location of the irrigator within the river system, the water sources available for irrigation may vary, and thus the complexity of the crop water model will also vary.

In some river reaches, water for irrigation may only be available by pumping directly from the river (metered), while in other reaches, water may be available from a variety of sources including: direct pumping from the river (metered), groundwater (metered), rainfall-runoff harvesting (not metered) or floodwaters (not metered). While the method for calibration for the on-farm water balance varies between reaches, understanding demand for irrigation water is essential in order to use the river system model to evaluate the likely impact of different policy decisions on water users.

For this practice note, the calibration of crop water models may refer to the calibration either of a single farm or multiple farms that have been lumped together. The lumping or not of irrigation demands when using crop models should be determined during reach conceptualisation (Practice note: Reach Conceptualisation - Identification of key fluxes).

General principles

The modeller should develop an understanding of the sources of water available to meet irrigation demands during model conceptualisation and should decide whether irrigation diversions should be lumped or represented as individual farms. See Practice note: Reach Conceptualisation - Identification of key fluxes.
The modeller should collate any information available required to understand the on-farm water balance (see the 'Estimation of Unmetered Irrigation Diversions (On-Farm Water Balance)' Practice Note).
When floodplain harvesting is a significant component of the on-farm water balance, the modeller should consider what information is available to inform the estimation of total water use by these farms and thus estimate the unmetered take (see the 'Estimation of Unmetered Irrigation Diversions (On-Farm Water Balance)' Practice Note).
When calibrating crop water models in Source, the modeller should use the irrigator module (Irrigator Demand Model - SRG ) in preference to other methods available for estimating crop water demands.
Calibration of the crop water model should occur systematically, with any known values set to observed during the initial calibration.
The calibration period should focus on the years closest to the target date of the model to ensure representativeness regarding on-farm efficiency, infrastructure and planting behaviour. However, the calibration period should also aim to include periods of full and limited water availability to capture the range of climatic conditions.
Crop parameters should be set prior to calibration to limit the number of parameters that are adjusted during the calibration process. Where parameters are changed from industry standard values, justification and documentation is required to explain why.
Appropriate checks should be undertaken following calibration to ensure that the diversions from metered and unmetered water sources, planted area and crop water use (ML/ha) are within the expected ranges.

Recommended High-Level Method

Conceptualisation

During reach conceptualisation (Practice note: Reach Conceptualisation - Identification of key fluxes) the following should have been decided:

How irrigation demands are represented in the model. The conceptualisation should have considered:
1. That lumping of irrigation demands best suits the conceptual premise of the crop models, so farms with similar behaviour within a reach should be lumped into a single irrigation node.
2. The current and likely future uses of the model.
Where the irrigation demands are located in the reach; this affects access to water from residual inflows and flood runners.

Data Requirements

Where unmetered diversions are a significant component of the reach water balance, available data should have been collated during the estimation of unmetered diversions (see the 'Estimation of Unmetered Irrigation Diversions (On-Farm Water Balance)' Practice Note).
Where unmetered diversions are not a significant component of the reach water balance, the amount of data available to calibrate the crop water model may vary between locations.
Before calibration, the modeller should establish what data is available from the following list.

Climate data (Rainfall and FAO56 Evapotranspiration)
Soil Data
Crop information
- Planted areas
- Crop types
- Planting dates
- Typical rates of water application (ML/day)
- Maximum crop area
- Minimum crop area
Observed regulated water availability
- Trade information
- Account balances
- Effective allocations
- Allocation announcements
- Observed diversion data
Information about on-farm practices
- Understanding the on-farm infrastructure
- Pump capacities
- Irrigation efficiency
- On-farm storage information (including observed storage volumes/heights)
- Ability to provide temporary storage on-farm during flood events
- Use of groundwater (recorded use by groundwater licences).

Selecting a method for calibration of the crop water model

The method of calibration depends on the data available and the ability of the water users to access different water sources. Depending on the type of farm and the available data, calibration aims to match the observed diversions and the area planted. Typically, for reaches where there is no on-farm storage or ability to capture floodwater, the on-farm water balance is calibrated to the observed diversions. When farms are known to capture floodwater for irrigation purposes, calibration is more difficult due to the limited data available on floodplain harvesting. This section provides a high-level method for three situations:

Calibration when all available water is regulated (and metered) and there is no on-farm storage
Calibration of the crop water model in river reaches with access to regulated water only with on-farm storage
Calibration of the crop water model for river reaches with rainfall and floodplain harvesting. This calibration requires an estimate of the on-farm water balance (see the 'Estimation of Unmetered Irrigation Diversions (On-Farm Water Balance)' Practice Note).

Calibration when all available water is regulated (and metered) and there is no on-farm storage

Figure 1 shows the components of the on-farm water balance for a regulated reach with no access to floodwaters or ability to store water on-farm. Where all irrigation diversions in the river reach are metered, the method of calibration of the crop water model will depend on the data available and the period between metered diversion readings.
The main steps in calibration for these crop water models are:

Data acquisition: Collate all data relevant for calibration of crop model
Data manipulation: Data needs to be input to Source at the appropriate timescale. Data manipulation may include:
1. Observed available water – required as daily time series.
2. Establishing a time series of crop area for the season. Where area data are missing, the estimation of crop areas may be required. Methods for infilling data include: area based on observed take during the irrigation season (observed take divided by typical application rate) or interpolation of areas with between observed data points, or estimation from remote sensing.
3. Selection of planting date – should be based on textbook values or local expert knowledge.
Develop Source Model:
1. The initial model run should fix as many parameters as possible to observed or textbook values.
2. Where crop types and areas are known for the calibration period, an initial calibration is undertaken with observed crop areas. This model run involves
  1. Running the model and assessing the diversion results at a mean annual timestep to see if the bias matches. If the bias does not match, then the efficiency is adjusted. Efficiency should be maintained within an expected range based on your understanding of factors such as soils and infrastructure.
  2. The seasonal pattern should be assessed, and if this does not match the observed, then the target depletion should be adjusted.
  3. An assessment should be made of the ML/ha usage against published figures and values in adjacent reaches for this initial model.
3. Once efficiency and target depletion have been set, a planting decision is incorporated to determine the area planted based on available water. This planting decision should consider:
  1. Available water on the planting decision date.
  2. Permanent and temporary trade information.
4. Following the incorporation of the planting decision, the following checks should be undertaken and reported in the calibration report:
  1. An assessment of the crop water use in ML/ha. Does this look reasonable against published figures for that climatic region and when compared to other crop water use models in the area?
  2. An assessment of the annual, monthly and seasonal patterns.
  3. An assessment of how the crop water model responds to climate variation.
  4. An assessment of how well the model represents the mean diversion.

Figure 1: On-farm water balance for water users on a regulated reach with no on-farm storage. It should be noted that in some cases irrigation returns are not explicitly modelled in the calibration but are accounted for in the efficiency and application rates.

Calibration when all available water is regulated (and metered) and there is on-farm storage with rainfall harvesting (Figure 2)

The main steps in calibration for crop water models where there are on-farm storages but no unmetered diversions are:

Data acquisition: Collate all data relevant for calibration of crop model (see the 'Practice note: Estimation of Unmetered Irrigation Diversions (On- Farm Water Balance)).
Data manipulation: Data needs to be input to Source at the appropriate timescale. Data manipulation may include:
1. Observed available water – required as daily time series.
2. Crop area planted on the planting decision date. Where area data are missing, the estimation of crop areas may be required. Methods for infilling data include: area based on observed take during the irrigation season (observed take divided by typical application rate) or interpolation of areas with between observed data points.
3. Selection of planting date – should be based on textbook values or local expert knowledge.
4. Sizing of on-farm storage, based on available information and gaining an understanding of the filling and emptying behaviour for the on-farm storage based on any observed data.
Develop Source Model: Fixing as many parameters as possible to observed or textbook values. The following steps should be followed to calibrate the crop water model:
1. A run of the model should be undertaken forcing the following inputs:
  1. Planted areas
  2. Setting rainfall harvesting parameters to the best available information on runoff rates under fallow and irrigated conditions.
2. Following this run, an assessment of the on-farm storage behaviour should be undertaken to see if the storage is behaving as expected and the regulated diversions match the metered diversions. If not, then the rainfall harvesting parameters, efficiency and target reserve should be reviewed and revised.
3. Once the behaviour of the on-farm storage and rainfall harvesting seems reasonable, the forced planted areas are replaced with a planting decision function that determines the planted area based on the available water (both in accounts and OFS).
4. Following the incorporation of the planting decision, the following checks should be undertaken and reported in the calibration report:
  1. An assessment of the area planted against any observed information.
  2. An assessment of the crop water use in ML/ha. Does this look reasonable against published figures for that climatic region and when compared to other crop water use models in the area?
  3. An assessment whether the annual pattern makes sense (monthly behaviour and seasonal behaviour in particular).
  4. An assessment of how the crop water model responds to climate variation.
  5. An assessment of how well the model represents the mean annual diversion.

Figure 2: On-farm water balance for water users on a regulated reach with on-farm storage but no access to floodwaters for irrigation. It should be noted that in some cases irrigation returns are not explicitly modelled in the calibration but are accounted for in the efficiency and application rates.

Calibration of the crop water model for river reaches with rainfall and floodplain harvesting (Figure 3)

The main steps in calibration for these crop water models are:

Data Acquisition, Review and Manipulation: Collate all data relevant for calibration of crop model (see Practice note: Estimation of Unmetered Irrigation Diversions (On- Farm Water Balance)).
Set up a model with as many parameters as possible set to predefined values based on literature and expert advice. Set rainfall harvesting parameters to the best available information on runoff rates under fallow and irrigated conditions. The suggested method is to undertake an initial model run with the following inputs forced to observed/estimated:
1. Metered diversions
2. Planted areas (forced to observed or estimated).
Assess the on-farm water balance with the initial model configuration. The requirements for this assessment are detailed in the 'Estimation of Unmetered Irrigation Diversions (On-Farm Water Balance)' Practice Note; this should include an assessment of the water use in ML/ha, an assessment of the contribution from metered and unmetered water sources, and the timing of water diverted from unmetered water sources.
The model has been calibrated to reflect the water supplied from unmetered water sources with forced areas and observed metered diversions. If the results for unmetered diversions and total water use by the crops are considered acceptable, the forced areas should be replaced with an appropriate planting decision function. It is suggested that the planting decision function should be based on total available water (i.e. account balance + OFS volume). The model with the planting decision functions should be assessed to ensure that there are no unexpected changes in outputs for the unmetered diversion.
The final step in the model calibration is to replace metered diversions data with simulated diversions for metered diversions.
Assessment of the final model should include undertaking appropriate checks for reporting in the calibration report. These would include:
1. An assessment of the area planted against any observed information.
2. An assessment of the crop water use in ML/ha. Does this look reasonable against published figures for that climatic region and when compared to other crop water use models in the area?
3. An assessment whether the annual pattern makes sense (monthly behaviour and seasonal behaviour in particular).
4. An assessment of how the crop water model responds to climate variation.
5. An assessment of how well the model represents the metered diversion.

Figure 3: On-farm water balance for water users on a regulated reach with on-farm storage and access to floodwaters for irrigation. It should be noted that in some cases irrigation returns are not explicitly modelled in the calibration but are accounted for in the efficiency and application rates.

Practice note: Calibration of Crop Water Models