Irrigation

WRSM-Pitman (Sami groundwater)

• Irrigation modules in WRSM cannot contribute return flows upstream of the source they are irrigated from. This means that areas irrigated by a farm dam are necessarily removed from its contributing catchment area even if they are actually located upstream of the dam in reality. This restriction is not present in other tools. If the relevant irrigated area is large in relation to the dam’s contributing catchment, this could lead to inaccurate dam inflow, particularly during large rain events producing surface flow.  With many smaller dams and hence associated smaller irrigation areas this becomes less of a problem than it would be if all small dams were lumped and had a large associated area not considered part of the contributing catchment.  
• Each irrigation area can only receive water from one source in WRSM, a single channel or reservoir module. A comparison of irrigation demand with potential surface water supplies over time was done to estimate the amount of irrigation that would have had to come from groundwater to maintain the observed agricultural production. In the model, groundwater withdrawal demand rates were input for the main runoff module of each quaternary/subcatchment to match the expected groundwater irrigation demand. The actual amount that would be withdrawn would be limited by the aquifer storage in the timestep. Groundwater withdrawals are removed from the model in WRSM, assumed to be used outside the catchment. To work around this, water was added to a ‘dummy channel’ module, as would be done for a flow transfer from outside the catchment. This amount was limited by the expected harvest potential of the aquifer. Specific groundwater irrigation areas were added to the model which were supplied by this channel module. The size of the groundwater irrigation areas was selected to match the estimated groundwater irrigation demand.
• A limitation of this approach is that the amount actually pumped from groundwater in the model and the amount of water made available to the groundwater irrigation areas are separate inputs. If the model groundwater storage is too low to allow groundwater pumping at a particular time, there should be no withdrawal, but this would not necessarily be included in the externally derived inflow time series. To prevent this an iterative approach is needed to check the withdrawals achieved compared to the inflows input. This process would also need to be redone if any relevant parameters or inputs are then changed (i.e. the irrigated area, crop type, rainfall, etc).  

SPATSIM-Pitman (Hughes groundwater)

• SPATSIM, as with WRSM, only allows irrigation from a single source for a given irrigation area in a subcatchment; includes irrigation from rivers, dams, and external input sources; and considers groundwater abstraction water to be removed from the model. As such a similar approach to WRSM would be needed for irrigation from groundwater, in which an irrigation area is specified to be irrigated from a water source external to the modelled catchment. This input is created to match the separately input groundwater abstraction. Small dams internal to a subcatchment are necessarily lumped into one unit per subcatchment in SPATSIM. The tool also includes a module for a reservoir at the downstream outlet of the subcatchment. This meant that two storage reservoirs can be included per subcatchment, as opposed to the many included in the WRSM set-up. This may result in surface water supply shortages being modelled at different times than were predicted in WRSM, in which specific areas would be limited by the amount available from a smaller local storage.    

ACRU4

• ACRU4 does not include groundwater withdrawals at all and each irrigated area can only receive water from one source, channel or reservoir.  Many farm dams in a catchment are permitted, but unlike WRSM and SPATSIM, in which a user-input proportion of the sub catchment runoff can directed to a dam, dams included in ACRU4 would require specific HRUs upstream to represent the dam’s separate, or incremental, contributing catchment area. Because of this, it was decided to only include two lumped farm dams in each subcatchment, rather than the many in WRSM, to represent those fed by highland areas separately from those along lowland main channels.  This is similar to the structure in SPATSIM.  
• An approach to attempt to represent irrigation from groundwater in ACRU4 was discussed for this case study. An initial set-up for a single sub catchment was done; however, it was determined that the method would need significant testing and adjustment to perform as desired and was outside of what could be considered ‘typical use’ of ACRU4. As such it was not used in a full model of the Middle Letaba catchment.  In this approach, an additional ‘dummy’ dam is added to the subcatchment to represent aquifer water storage. A ‘dummy’ riparian zone HRU is added to effectively route the baseflow outputs from all the HRUs in the subcatchment and can be routed into the dam: flow is routed via the dummy riparian HRU, which is assigned parameters to promote drainage rather than ET. It could be seen as the vadose zone below the soil profile. The ‘aquifer dam’ is given a maximum storage volume matching the maximum aquifer storage volume, and a maximum seepage rate (occurs when the dam is at its maximum storage) equivalent to the maximum aquifer outflow rate. Shape parameters can be selected to minimise evaporative surface area and the lowest allowed evaporation coefficients are assigned, however evaporative losses can’t be completely eliminated. This is a short-coming of the approach.  In this case study, outflows from the ‘aquifer dam’ would be routed to the channel upstream of the real surface water dams such that these would not be starved of baseflow.
• The ‘aquifer dam’ will only receive inflow from the contributing HRUs at the rate that they are predicted to produce baseflow in the model, controlled by a baseflow lagging parameter.  Because storage may accrue over time in the aquifer dam, depending on the inflow versus outflow rates, there may be enough water stored to allow irrigation from groundwater in excess of the amount of baseflow that would otherwise be modelled in the river, which is the desired conceptual outcome. However, it would be conceptually valid to decrease the baseflow lag from contributing HRUs compared to normal values applied because a lag is already being imposed through storage and seepage outflow in the aquifer dam. Finding appropriate values for HRU baseflow lag and ‘aquifer dam’ seepage parameters would likely need to be a calibration exercise.    
• As with the Pitman tools, separate irrigated HRUs would be needed to represent irrigation from groundwater (from the ‘aquifer dam’) and irrigation from surface water sources. This ‘aquifer dam’ approach has a representation advantage over WRSM and SPATSIM approaches in that irrigation would be curtailed internally by the model when the modelled groundwater availability was low.  

ArcSWAT2012

• ArcSWAT2012 allows irrigation from reservoirs, river withdrawals, and groundwater. However, the software is designed for each irrigation area and can only be supplied by a single source. If this approach is used, the same total area of groundwater irrigation could be used as was assumed in the WRSM model. Using ArcSWAT this requires spatial selection of mapped irrigated areas to assign to a groundwater irrigation land cover class in the model. There may be a ‘work-around’ to allow irrigation of the same area from two sources that is now being tested.  Irrigation can be set-up with either an input application amount schedule (manual) or with irrigation amounts calculated by the model to maintain soil moisture above a certain threshold (automatic). It appears an HRU can be assigned both a manual irrigation set-up and an automatic irrigation set-up and that these can come from different water sources. A manual irrigation from surface water sources coupled with an automatic irrigation from groundwater could hopefully allow the groundwater irrigation to be automatically modelled when surface water sources are depleted and soil moisture drops.  What remains to be confirmed is that the manual irrigation is calculated first when both are included in the set-up. Appropriate irrigation rates to specify for manual irrigation then need to be estimated.
• An additional logistical challenge to setting up the Middle Letaba irrigation system in ArcSWAT is that the addition of small farm dams is more complex than for Pitman tools and ACRU. SWAT does include a ‘pond’ water storage feature which can be fed by flows by a specified proportion of a larger subcatchment, as in the Pitman tools, but this was intended for flood control structures and cannot feed irrigation in the model. Reservoirs that can feed irrigation require specific delineation of their contributing catchments using the input DEM and delineated drainage lines. This is an automated process in the model interface in which the user specifies the outlet points of the dams to be included. As such the larger dams located along the main rivers that were explicitly represented in WRSM were also explicitly represented in SWAT for this case study. However, for numerous very small dams, it becomes impractical to specify an outlet point, delineate a subcatchment, and input a storage-area relationship for each one individually.
• The work-around to represent many small dams for this case-study was to find a suitable point on a tributary stream with a contributing catchment area equivalent to the contributing catchment area assumed for the lumped smaller dams in each quaternary in WRSM and create a dummy dam at this location in the SWAT model. The selection of this ‘dummy contributing subcatchment’ should ideally take into account the land cover and topography distribution of the areas that actually contribute to the small farm dams if possible, however this may not always be an option. In this case study, cover in the uplands was assumed to be fairly uniform so this did not pose a problem.

MIKE-SHE (using simpler options and using more complex, distributed options)

• MIKE-SHE, using any complexity level, allows irrigated areas to be supplied by multiple different sources in an ordered sequence, such that when one source does not have sufficient supply available to meet the irrigation demand, the model will attempt to draw water from the next listed source if possible.
• However, the potential irrigation water sources are allowed are river withdrawals and groundwater withdrawals. Reservoirs can be represented on the river network by adding their bathymetry as a set of cross sections and the dimensions of a dam wall as a structure into the river hydraulic model. In this way a ‘river’ withdrawal would be a reservoir withdrawal. Setting up dams in this way requires detailed data for each one, so is not well suited to the case of having many small dams. A lumped ‘dummy’ reservoir could be added to the channel as the other tools do, but the input data preparation is still more intensive. The user has to work out the bathymetry cross sections, wall dimensions, and position in the channel network would have the desired impact and be stable in the hydraulics simulation. There is a simpler option to add a reservoir as a storage unit with specified volume onto a channel network in MIKE-SHE, however this does not allow irrigation withdrawal directly from the unit and it is not clear that the unit would result in enough water ‘backing-up’ in the hydraulically modelled channel upstream of it to mimic the water availability of the reservoir.