Modelling the temporal and spatial variation of evapotranspiration from irrigated pastures in Canterbury

Evapotranspiration is a critical factor for local and regional planning, in terms of both water quality and quantity, to inform decisions around water catchment management, irrigation, water storage, and resource sustainability. Despite its importance, understanding of evapotranspiration from irrigated dairy pasture in Canterbury has to date been relatively limited. The focus of this research was, therefore, to improve understanding of evapotranspiration from ryegrass-based irrigated pastures under grazing. This was achieved through quantifying relationships between actual evapotranspiration and canopy development and evaluation and validation of methods commonly applied in the estimation of potential canopy evapotranspiration (PETc) for grazed perennial ryegrass (Lolium perenne L.) pasture. A network of nine lysimeters located at three sites across the mid to north Canterbury Plains was used. Pasture canopy measurements were taken throughout the study at one of the sites, and the biophysical model ‘DairyMod’ used to simulate pasture growth at all three sites. The ‘DairyMod’ and ‘HYDRUS-1D’ models were used to simulate soil water flow, and used to support lysimeter-based estimates of actual evapotranspiration. Methods examined for modelling PETc included the use of a crop coefficient time series and a number of commonly applied single-layer models including Penman-Monteith (PM), FAO-modified Penman-Monteith (PMFAO) and Priestley-Taylor (PT), and the dual-layer dual crop coefficient (DCC) and Shuttleworth-Wallace (SWW) models. ‘DairyMod’ and ‘HYDRUS’ and selected PETc models were validated with data collected under a controlled, perennial ryegrass and white clover (Trifolium repens L.) pasture experiment at Lincoln University, with two levels each of irrigation and nitrogen fertiliser. At all three lysimeter sites, the pasture production was nitrogen-limited, with herbage yields of 10.8-14.9 t DM/ha/y, below optimum yields achievable for Canterbury. The results suggested under-fertilisation of pasture to be prevalent across the region. ‘DairyMod’ was successful in accurately simulating pasture growth under a commercial dairy operation when compared with the measured lysimeter data. However, limitations within the model were identified. Specifically, the calibrated model failed to account for the mechanistic relationship between nitrogen and leaf extension at the temperature stress parameter values required to achieve a reasonable fit with the observed data. Accordingly, within the model, temperature became more limiting than nitrogen. However, this was able to be overcome through the reliance on the empirical relationship between temperature stress and photosynthesis whereby temperature stress functions in the model could be manipulated to achieve the ‘correct’ yields. ‘HYDRUS-1D’ was found to be superior to ‘DairyMod’ in the simulation of soil water flow. This was due to the closer predictions of drainage and soil moisture content with that observed compared with DairyMod. The simulated drainage highlighted issues in the lysimeter design. Where lysimeters were installed without rubber rims around the top of lysimeter casings, there was the potential for surface redistribution of water to occur. This ultimately led to discrepancies in the lysimeter data through unaccounted for water losses and therefore a reduction in drainage. A crop coefficient time series was developed from the lysimeter water use data. The time step over which water use measurements were made was the dominant contributing factor to variation in monthly crop coefficient values between lysimeter sites. When daily estimates of water use are used rather than weekly or greater, which were calculated with the SWW model, the spatial variability was largely eliminated. Temporal variations were found to be seasonally driven. When a mean crop coefficient time series from the three lysimeter sites was used to predict PETc, estimates were within 1-11% of the actual evapotranspiration (AET), determined using the observed lysimeter data. When used to predict PETc in the Lincoln experiment, estimates were within 3-13% of AET when water was not limiting. The results highlighted that, due to temporal variations, use of a single crop coefficient value could not be supported, which led to the rejection of the null hypothesis. However, the site averaged time series could be used for water allocation management purposes over the irrigation season months. The single layer PM, PMFAO and PT models over predicted water use. There was a strong systematic error to the daily PM estimates. When the canopy was small PM under-estimated AET but at a leaf area index greater than approximately 1.3, PM over-estimated AET. This led to total over-estimations of 8-29%. However, this was an improvement on the climate-based PMFAO predictions, which over-estimated AET by 31-58%, and the PT model, which over-estimated AET by 17-30%. The failure of these methods to accurately predict water use was due to their inherent assumptions of the canopy that are not representative of a typical grazed ryegrass pasture in Canterbury. The dual crop coefficient model provided estimates of water use within 1-24% of the observed AET. However, the SWW model predicted water use within 9% of AET. This was largely owing to the separation of soil evaporation from canopy transpiration, enabling the influence of the canopy on the potential soil evaporation to be adequately accounted for. These results highlighted the benefit of using the SWW model for irrigation management purposes over other PETc models and the need to actively account for the canopy, being the primary factor controlling water use. The method of estimation was also found to be significant, whereby under a grazed system the process of canopy transpiration needs to be separated from the soil evaporation. Finally, irrigation schedule scenario testing across the three lysimeter sites indicated ~35% of drainage losses could be avoided through optimising irrigation scheduling. Doing so involved applying 10-20 mm per irrigation event, with a minimum return interval of 2-4 days, and delaying irrigation by 10-14 days following grazing of the paddock. Through optimising irrigation, the total irrigation applied was reduced by up to 64% without compromising herbage accumulation. However, while less irrigation was required, more nitrogen was necessary for optimum yields to be achieved. It is likely that this finding is applicable to many commercial dairy farms region wide.