Previous PageTable Of Contents

Simulating Pasture production to better manage for climate variability in NSW

John Crichton

New South Wales Agriculture
Kite St, Orange,
NSW 2800, Australia
Tel +61 2 6391 3140, Fax +61 2 5391 3767
john.crichton@agric.nsw.gov.au

Abstract

New South Wales Agriculture is using the CSIRO decision support tool, GrassGro, to model pasture production in the New South Wales Tablelands relative to the historical record. GrassGro simulations are run monthly for some 45 locations in the study area using daily weather data from SILO, a suitable sward and soil data from the locality.

We compare the results from each locality graphically with herbage intakes from the same locality over the period from 1957 to the present, so that current seasonal pasture production at a place can be seen in its historical context. The different quantity and quality of feed that is available at different times of year is accounted for by integrating it into metabolisable energy via the grazing animal.

The method appears to be valuable in summarising current seasonal conditions over the temperate high rainfall pasture lands of New South Wales and probably Victoria. It could be used to alert farmers and others to the severity of developing seasonal conditions and possibly their duration. It has the advantage over rainfall-triggered measures of the severity of dry times that it takes the evolving condition of stock and pasture into account. This project is part of the Australian Grassland and Rangeland Assessment by Spatial Simulation (Aussie Grass) project.

Introduction

The variability of the Australian climate is legendary. A glance at the early history of settlement in South Australia or the westward spread of cropping in NSW following good seasons in the mid-50’s and subsequent events make interesting reading. Not only the events themselves, but the challenges and hardships faced by the early pioneers are remarkable. Even more remarkable if we try to identify with the people who faced the challenges in those early days.

Analogues of these events took place elsewhere, in other continents, at different times through history and pre-history, but the dryness of the Australian climate made the local experience particularly difficult.

Figure 1, (Mavi, H. pers. com.) illustrates the effect of rain that fell during the growing season on the yield of representative crops at 18 localities in the crop belt of NSW in a wet and a dry year. Over a number of years, the correlation between rainfall and yield is high, but a number of other factors, including the timing of the rainfall, are important too. Graziers have similar problems to arable farmers, but their situation is complicated by the presence of stock, their husbandry and their interaction with the growing herbage.

This paper reports the results of the use of a pasture simulation model to shed light on the possible management and effects of climatic variability on temperate grazing enterprises. It is based on work reported in the High Rainfall Zone of the Aussie GRASS project (Hall et al, 2001).

Managing with climate variability on the farm

Accurate estimates of the West Australian wheat crop are routinely made using linear models of rainfall and time (Stephens, 1996). Rainfall analysis alone, however, does not necessarily reflect the quantity and quality of pasture available on the ground. In the 1991-95 Queensland drought, for example, rainfall analysis misclassified the drought-affected south-western areas and some drought-free coastal areas of the State (Brook, 1996). The researchers found that measures of rainfall effectiveness expressed as plant biomass were required for drought definition. In some areas where rainfall alone may account for about 50% of the variance, over 70% may be explained when other factors, such as evaporation and temperature are taken into account. One way to do this for pastures is to use a growth model.

Such a model could be used on the farm to illustrate the effects of stocking rate on gross margins and to help balance the risks and benefits of increasing stock numbers. The model could also be used to simulate the effects of different management decisions under varying climatic conditions and thereby help the farmer to manage for climatic variability. Before applying the results of such an exercise it would be prudent to “hindcast” by modelling the past to test that the assumptions made and the modelled outcomes were reasonable.

Pasture simulation models

Several simulation models are available and some of their characteristics can be illustrated by comparing two examples, GRASP (Littleboy and McKeon, 1997) and GrassGro (Moore et al., 1997).

GRASP was developed to model the growth of tropical herbage in Queensland and the Australian “Top End” whereas GrassGro was developed for temperate regions. These models differ in their implementation and the assumptions used. GRASP models the growth of a simulated sward from its initial conditions in a daily step, according to daily weather information, by means of a set of parameters which specify the characteristics of the sward and the soil. It combines two successful approaches to modelling plant growth, by calculating both a soil water budget and a daily plant growth index, which are used to estimate growth. Animal intake is assumed constant. An alternate set of sward parameters is being developed to allow GRASP to be used in temperate areas, too.

Figure 1. Seasonal rainfall and crop production in 1993-94 and 1994-95 seasons

GrassGro is also driven by daily weather information. The growth of the several components of the pasture sward is modelled taking into account competition among the species present and the effects of grazing animals. The performance of the sward is the sum of the performances of its component species. GrassGro currently has parameter sets for about 22 grasses, legumes, fodder crops and fodder weeds eg. capeweed.

Setting the initial conditions specifies the composition of the sward, the stage of growth of its components and soil conditions at the start of the simulation. Subsequently, the daily growth of each component and the competition between them is modelled by the program. Consequently, the composition of the simulated sward changes year by year in response to the seasonal climate, as one component or another is favoured by conditions. In this study, to maintain an adequate pasture at each locality over the 43 year span for which weather records were available, the sward was simulated by a single representative species, commonly Phalaris, an improved perennial.

As part of CSIRO’s Grazplan package, GrassGro models the management, growth and production of the grazing animals in detail. In poor years, the simulated animals are fed maintenance supplement as required to maintain a preset minimum condition score. To help evaluate possible alternative management strategies, GrassGro can be run tactically as well as historically. Gross margins can be calculated to facilitate comparisons between the outcomes of different strategies.

Daily weather information was obtained from the patch point database in SILO (Mullen et al, 1998) on a monthly basis. Any missing values in the incoming data are replaced by estimates in the SILO database. Soil information was as far as possible real data from the locality being modelled.

Modelled Regions

Initially, modelling was concentrated in three broad areas of eastern NSW, where the bulk of pasture production is concentrated, the Southern Tablelands, the South-West Slopes, and the Northern Tablelands. Two other areas, the Southern Riverina and the South Coast were added later. Outside the temperate high rainfall areas, modelling was precluded by the lack of suitable parameterised pasture species for GrassGro. The extent of the modelled area is shown in figure 2.

The Southern Tablelands, westward from the Blue Mountains, is typical of the temperate grazing country on which much of the State’s wool and meat are grown. The elevated areas are largely long-term pastures characterised by autumn, winter and spring growth and rainfall in all seasons. Modelling was based on Phalaris, which is a common component of the pasture. Phenologically, it is similar to the other major contributors to pasture biomass in this area. In summer, when phalaris is dormant, annual grasses can make a significant contribution to biomass in some areas but the bulk of growth takes place in the cooler months.

In the Northern Tablelands, rainfall becomes more seasonal. North of the Bathurst and Orange districts, summer rain and higher temperatures combine to make Tall Fescue a suitable species for modelling. Unlike phalaris, the ecotype of fescue used grows all year round given suitable conditions. Fescue was used in all the northern areas in the early runs. Following feedback from district officers, however, phalaris was substituted in elevated areas, notably the Tamworth and Armidale districts, where most growth occurs in the non-summer period.

The South-West Slopes stretch west of the Southern Tablelands as far as the 500-mm isohyet and are devoted largely to cropping. In southern and eastern parts, extending north of Wagga, phalaris remains a suitable indicator species for pasture modelling. In the pasture leys of the west and north, annuals are more representative than phalaris, and Annual Ryegrass was used as the indicator species.

The Southern Riverina was modelled using Phalaris. Rainfall is on the low side at 450mm but it is evenly distributed throughout the year.

The North-Western Slopes and Plains have a subtropical climate and were not modelled in this study. GrassGro is an inappropriate model to use as no suitable species have been parameterised. For similar reasons, it was not feasible to model production in the North and Central Coasts due to the lack of validated parameter sets for C4 grasses, typical of these areas. GRASP may provide better pasture simulations of the hot, dry and coastal areas mentioned.

It may be possible to better model the South Coast with GrassGro when Kikuyu parameters are more fully tested. For this project, phalaris was used.

Pasture Performance Index

The GrassGro model is capable of calculating a range of soil, pasture and animal outputs. A measure of pasture growth and availability was needed to compare pasture status in different areas and at different times of year. In spring, the available mass of green herbage could be used and in summer, available dead herbage and litter would be satisfactory. There is no measure of herbage that could be used as an indicator all year round, however, as the quality and quantity of available feed vary throughout the year in response to growing conditions, grazing and management.

Rather than using any of the direct measures of pasture, we used livestock performance to integrate the relative quantity and quality of available pasture over time. The measure used was the metabolisable energy intake (MEI) of the simulated livestock, which is proportional to herbage mass and digestibility up to certain limits.

We calculated a relative livestock performance as the metabolisable energy intake (MEI) of the stock present compared to the MEI percentiles over a 43-year period. MEI was plotted in the trend graphs as a 14-day running mean, for smoothing purposes. The seasonal conditions map, (eg. Figure 2) produced monthly as a summary of pasture condition in the study areas, was based on the current state at the end of the month. In the figure, the colours represent current seasonal pasture conditions relative to historical records as measured by the MEI percentiles; areas of light green for example, indicate that current pasture condition lies between the 15th percentile and the median. Use of a relative index rather than an absolute measure allowed a more meaningful extrapolation across the landscape since the sward was represented by a single species. The relative index also compensated for the different pasture qualities and management strategies that occurred within a district.

As GrassGro is a point model, the mapping units used were Rural Land Protection Districts and their divisions, modelled using climate data from 45 long-term weather stations.

Figure 2, Seasonal Pasture Condition

Initial conditions

Rainfall measurements have been recorded widely in Australia since the 1840’s, but computerised records of maximum and minimum temperatures, evaporation and radiation have only become available recently. In the project, long modelling runs on which the percentiles are based began in January 1957 and ended in December 2000. This is the longest period for which complete weather data were available from climate surfaces, and representative of late trends of increasing rainfall.

Model runs were started on 1/1/1957. Initial values for standing dry matter, litter and underground biomass were set at median values for the locality, based on trial runs. Soil moisture was set half way between wilting point and field capacity. Dry feed was distributed equally among the available digestibility pools. In areas where the sward was annual grass, an adequate seed pool was established.

Weather updates

The simulations were run in batches. When weather updates were received, early each month, records for the 65 weather stations were updated to the first day of the new month. Updates were rolling two-year revisions that included the current month’s data and adjustments to earlier records, the result of quality control by the Bureau of Meteorology (BoM) and Silo. Previous records held within the update period were overwritten by updated data and lost.

Each month, when the weather records were updated, all the GrassGro simulations for the entire 43-year period were re-run. Two tables were extracted from each simulation; (i) daily herbage MEI percentiles from 1957 to 2000 and (ii) daily MEI intakes from 1997 to the present. The two tables were imported into a spreadsheet, which was used to smooth the values and plot a relative livestock performance graph for each locality. The spreadsheet was also used to calculate the pasture classes needed for the Seasonal Condition map (figure 2).

The graphs of relative livestock performance effectively put current herbage intakes in their historical context. They have proved to be a useful summary of seasonal conditions. The graphs were circulated to experienced officers in the districts both to obtain local feedback on their accuracy and to improve the presentation.

The relative livestock performance graphs (eg. figure 3) consisted of curves of daily intakes from the previous three years superimposed on the five fixed percentiles of MEI (5, 15, 50, 85 and 95). The 5th percentile corresponds to a one in 20 year event. In 19 years out of 20, conditions are better. The 1998 curve was included as early 1998 was part of the most recent El Niņo event, when conditions in several districts were poorer than the 5th percentile. This was important initially to enable our collaborators to interpret the graph in dry conditions. Graphs presented here consist of a single year for clarity.

Livestock and stocking rate

The indicator livestock used at a locality, wethers or steers, were intended to have been characteristic of the grazing enterprise of the district. Attempts to model a replacement strategy and age structure of steers to impose a relatively constant stocking rate on a pasture throughout the year were unsuccessful. Consequently, wethers were modelled in all areas. This substitution is of no consequence as the animals are merely used to integrate the quantity and quality of available pasture to a common index. To reduce all areas to a common level and eliminate the effects of under or over stocking, the stocking rate for each locality was adjusted to achieve approximately 35% long-term pasture utilisation.

A 35% utilisation rate is representative of the rates achieved by conservative practice in many areas. There was no way of implementing a flexible stocking rate and indeed this was not desirable since the purpose of the project was to monitor seasonal conditions. Changes in stocking rates in response to seasonal conditions would only serve to mask the results.

In good years, the chosen stocking rate resulted in fat animals and potential intakes not being reached. This has the effect of narrowing the distribution above the 80th percentile in the best seasons. In poor years, with sub-optimal pasture availability, the MEI curves closely mirrored pasture condition. For the purposes of Aussie Grass, it was desirable to model poor conditions and shortages of feed rather than times of plenty because of the need to minimise losses and overgrazing of pastures.

MEI, an indicator of Seasonal Variability

For consistency, the performance of small merino wethers was modelled in all areas. The graphs show MEI intake against time. The five lines that do not cross are the (fixed) MEI percentiles (from the bottom, 5%, 15%, 50%, 85% and 95%). The current year line is bold. Figure 3 shows the situation at Bombala in March 1998. On the graphs, the vertical axis increases from zero to 20MJ/head/day. Below about eight MJ/d wethers lose weight. As conditions worsen, at some point depending on their initial condition score and available pasture, the model will feed supplement to maintain a (low) condition score. Feeding supplement had little effect on the graph as only nutrition from herbage intake is taken into account. At higher MEI values, the lines are flatter and the percentiles closer together as the stock, with a surfeit of feed available, cannot eat beyond their capacity.

All the areas modelled show a spring or early-summer maximum in availability and reliability of feed. Feed is reliable when the MEI percentile lines on the graph are close together, ie when there is not a large difference in MEI between a good and a poor year. Data for Bombala (Figure 3), for example, showed that on average, adequate, reliable feed is available year-round (>10MJ/head/day) in six years out of seven (the 15th percentile). The graph also shows that in one year in 20, at this stocking rate, a cold, dry winter can lead to severe feed shortage until mid-September.

A comparison between the colours and line thicknesses used in figure 3 and subsequent graphs (eg. Figure 4) shows some of the improvements in presentation that evolved as a result of feedback from district officers.

Figure 3. Relative seasonal conditions for Bombala, in March 1998.

In the Northern Tablelands, Glen Innes (Figure 4) and Armidale (Figure 5) have good feed availability for nine months of the year in all but the worst years. The reason that the least reliable time of year was autumn at Armidale and spring at Glen Innes was that Phalaris was used as the indicator species at Armidale and Fescue at Glen Innes. Either of these species could have been used for simulation at Glenn Innes. Farmers would use the mixture of grasses and legumes that characterise their own pastures for runs of a few years. A limitation of the model at present is that mixed swards may not coexist well over long runs.

Figure 4. Relative seasonal conditions for Glen Innes, 2000.

Figure 5. Relative seasonal conditions for Armidale, 2000.

Tamworth (Figure 6) and Wagga (Figure 7) have MEI graphs of similar shape, which show a feed shortage beginning in January, which in the worst years, continues until August. Late May is the least reliable time of year. Relatively low intakes early in the year at Tamworth, illustrate the need for a summer-growing grass to grow in concert with Phalaris, as summer growth contributes proportionally more biomass in the non-seasonal to summer-dominant rainfall environments than it does in cooler areas.

Figure 6. Relative seasonal conditions for Tamworth, 2000.

Figure 7. Relative seasonal conditions for Wagga, 2000.

The Relative Livestock Performance graphs have proved useful as records of week by week seasonal conditions. Since 1998, global Southern Oscillation Index (SOI) conditions have been generally neutral and this has translated into above average pasture growth in most areas and in most seasons (eg. figure 2).

In addition to their role as seasonal indicators, however, pasture models can also be used to help manage seasonal variability.

Seasonal Forecasting

Seasonal variability in agriculture is a fact of life. Farmers, graziers and pastoralists are familiar with the losses that can accompany a late break or a dry spring. MEI graphs show seasonal variability well. Serious situations take time to develop, and the graphs illustrate their progress. When the year curve at a district drops below the 15th percentile, equivalent to 1 year in 7, it serves as a warning to focus interest on the district and add it to a short list for possible action if the line continues a downward trend. For farmers it would be a time to investigate feed and stock prices and costs of transport, to review options, make plans and decide on an action date. When the line drops below the 5th percentile, equivalent to a 1 in 20-year event, conditions are critical and urgent action is required.

While such dry conditions prevail, the natural response is to wonder how long they will last and how long before pastures recover and green feed is again available. Forecasting is notoriously unreliable. One way to make a forecast was to use weather records from an analogue year to extend the current year line into the future for a period of time.

We found that 3-months was an appropriate time. When a longer period was used, the possible outcomes multiplied and a poor forecast was the result. We concluded that the analogue procedure was subjective, both in the choice of the forecasting system used to select the analogue year, whether Southern Oscillation Index, Sea Surface Temperatures (SST) or a combination of the two, and in the identification of the specific analogue year from a number of possible analogues. Had the SOI phase been used, for example, a different analogue year would have been suggested as nearly every month went by.

An alternative approach, which we finally adopted, was to use the variance of the local weather record to obtain a graph of the temporal probability of growth using percentiles. We took the official 3-month forecast, with the greatest skill in the districts involved, and used it in conjunction with the long-term climate record for the stations concerned. GrassGro was used to analyse the corresponding three-month segment of every year since the beginning of the climate record and the resulting percentiles were appended to the year line. As a precondition, before such a forecast was undertaken, it was specified that the current MEI needed to lie between the fifth and fifteenth percentile and maintain a downward trend.

Figure 3 illustrates the procedure adopted at Bombala. By mid-March 1998, conditions were close to a one in twenty-year dry event and still in decline. How long before median pasture levels might again be achieved? Fig 8 shows the 1998 year line at Bombala, in March 1998, extended by 3 months by appending percentiles from the 5th to the 95th. Looking at the 95th percentile shows that recovery to median levels of pasture availability could not be expected before mid-May (60 days) by even the most optimistic grazier. Supplementary feed would be required for much of that time. Taking a pessimistic view, as illustrated by the fifteenth percentile, suggested a much longer period of shortage.

Figure 8. Relative Seasonal Conditions for Bombala as at March, 1998 with a three month forecast using the 5th, 15th, 50th, 85th, and 95th percentiles.

The Bureau of Meteorology’s 3-month forecast for the period March to May in 1998 stated, “the autumn rainfall over most parts of NSW is likely to exceed the median amounts”. On that basis, by following the 50% line, it would take less than about three months (90 days) for feed to recover. This is valuable information for planning purposes. Estimates could be made, feed sources investigated and feed ordered before the severity of the dry period became apparent to most people. Figure 9 shows that median pasture availability was finally achieved in early-July (112 days), nearly four months after the forecast was made.

Figure 9. Relative Seasonal Conditions for Bombala in 1998.

Information on the expected duration of a feed shortage and current livestock and feed prices could empower the farmer to assess the risks and act accordingly. Is the best strategy to destock, to spell selected paddocks and feed breeding stock or to attempt to feed existing stock for the duration of the shortage? GrassGro could be used to calculate outcomes to several possible scenarios and help the farmer to choose a number of preferred options for further analysis. Tactical runs can be run forward in time from the current status of stock and pasture condition to assess financial and practical outcomes including the cost of feed and the likely pasture condition in the end.

On the other hand, if circumstances suggest that an above average year is expected, analysis could show to what extent stock numbers might be increased to take advantage of potential high production without pasture degradation. A model like GrassGro is, however, only a decision support system. Akin to the mystical response of an Oracle, the results cannot simply be taken at their face value. Is the question you have asked likely to give you the answer you want? Did you pose the right question to result in the favourable answer you may have? The GrassGro model is a complex one. Scenarios need to be defined with care and the results need intelligent interpretation.

References

Brook, K.D. (1996). Research Summary, Final report on QPI 20 to Land and Water Resources Research and Development Corporation

Hall, W., Bruget, D., Carter, J., McKeon, G., Peacock, A. and Brook, K. (2001). Australian Grassland and Rangeland Assessment by Spatial Simulation (Aussie GRASS). Final Report for the Climate Variability in Agriculture Program, April 2001.

Littleboy, M and McKeon, G.M. (1997) Subroutine GRASP: Grass Production Model. Appendix 2 of “Evaluating the risks of pasture and land degradation in native pasture in Queensland.” Final Project Report for LWRRDC project DAQ124A

Moore, A.D., Donnelly, J.R., & Freer, M. (1997). GRAZPLAN: Decision support systems for Australian grazing enterprises. III. Pasture growth and soil moisture submodels and the GrassGro DSS. Agricultural Systems 55, 535-82.

Mullen, C. and Beswick, A, (1998). Use of SILO data for analysing and monitoring regional climate impacts. In 12th ANZ Climate Forum, Perth, Western Australia, 40.

Stephens, D.J. (1996) Using seasonal forecasts for national crop outlook; “Of Droughts and Flooding Rains.” LWRRDC Occasional paper CV03/95, 111-115.

Previous PageTop Of Page