Comparing the climate experienced during the Cicerone farmlet experiment against the climatic record

Additional keywords: available soil moisture, climate variability, farming systems research, growth indices, soil moisture stress days.

When interpreting the results of a dryland farming systems experiment, the climate experienced during the experimental period needs to be considered (Walker 1988). In the Cicerone Project’s farmlet experiment, individual paddocks were allocated without bias across the experimental area (159 ha) to ensure contiguous farmlets (Scott et al. 2013d), and hence, we have assumed that each farmlet experienced the same climate over the trial period. However, the question remains whether the climate over the 6.5 years of the trial was sufficiently representative of the long-term climate to permit adequate testing of the farming system treatments.

Typically, field experiments report the climatic conditions during an experimental period but rarely consider how it relates to climatic records over the long term (Laws et al. 2000; O’Reagain et al. 2011). Climate is especially important when the treatments being evaluated, such as changes in soil fertility and botanical composition, are known to take many years to reach equilibrium (Walker 1988; Jones et al. 1995).

The effect of climate on farming systems research conducted as field experiments has received relatively little attention in the literature, with most recent farming systems studies employing modelling to investigate issues of climate (Keating and McCown 2001; Moore 2009). The cumulative effects of rainfall, evaporation, and temperature during an experimental period are difficult to express over time, and there is a need to consider variability as well as averages (Tucker 1975; Johnson 1997; Katz and Brown 1992).

Temperature and rainfall variability affect farm performance largely through their impact on soil water content, which, in Australia, is arguably the most important constraint on pasture production (Stafford Smith and McKeon 1998; Lodge and Johnson 2008a). Using a simulation model, Stafford Smith and McKeon (1998) identified total soil moisture as the single best measure of drought and grassland stress events in Australian rangeland systems. Similarly, Lodge and Johnson (2008a) used the Sustainable Grazing Systems pasture model (Johnson et al. 2003) to simulate daily soil water content in four different temperate pasture types. They defined a drought period as having ≥50 consecutive days of soil water deficit, which corresponded to a volumetric soil water content at which pasture growth would be limited in the range 9.5–30.5% across the different soil types.

McCaskill and Blair (1988) modelled soil water on the Northern Tablelands of New South Wales (NSW) and found that, when available soil moisture was <17% of water-holding capacity, the conditions were likely to severely restrict growth of the best adapted legume in the region, white clover (Trifolium repens), with consequences for nitrogen fixation and, ultimately, pasture growth. Using this threshold, McCaskill and Blair (1988) calculated the number of such days per year, which they termed ‘critical clover days’ (CCDs), and compared average CCDs per year over different decades. They reported that a period of relatively low stress conditions for white clover occurred during 1950–79, which coincided with the time when large areas of the region were sown to introduced species. McCaskill and Blair (1988) suggested that white clover may not persist under the climatic conditions more typical of long-term records, and recommended that climatic experiences over the long term need to be considered when assessing the likelihood of pasture or animal growth in response to treatments.

The Cicerone Project’s farmlet experiment was created to investigate the sustainability and profitability of three farm-management systems on the Northern Tablelands of NSW (Sutherland et al. 2013). The experiment was conducted on three farmlets, each of 53 ha, over the period July 2000–December 2006. Farmlet A represented a high-input, flexible grazing system; farmlet B represented a moderate-input system, typical of district practice, with the same flexible grazing system as farmlet A; and farmlet C represented an intensive rotational grazing system with more than four times as many paddocks as the other farmlets but with the same moderate inputs as the typical-practice farmlet (B).

The climate of the Northern Tablelands is representative of the temperate, summer-dominant, high-rainfall region of south-eastern Australia; typically, the rainfall is highly variable between years and seasons (Johnson 1997). There is evidence that the rainfall in this region is changing, both between years and between seasons (Russell 1981). Carberry et al. (2005) reported that, over the 52 years between 1952 and 2004, Armidale experienced higher winter rainfall than the long-term average (1872–2004) in only 15 years, whereas below-average winter rain was recorded in 37 years. Similarly, erratic autumn rainfall may have been unfavourable for the sowing of pastures over the period 1972–2004 compared with the preceding 20 years (Carberry et al. 2005). The timing and distribution of rainfall events and its interaction with botanical composition, seasonal evaporation, and transpiration rates can produce quite different levels of plant growth (Johnson 1997; Carberry et al. 2005).

The aim of this paper is to examine the climate experienced during the experimental period July 2000–December 2006, using a range of climate measurements and indices, to compare it with long-term climate records and consider some of the implications for interpretation of the results of the farmlet experiment. In doing so, it is hypothesised that the farmlet experiment experienced a climate typical of that expected over the long term.

The Cicerone Project farmlet experiment was located 17 km south of Armidale, NSW (30°37ʹS, 151°33ʹE), at an elevation of ~1050 m above sea level, with terrain described as flat to undulating (Scott et al. 2013d).

Due to resource constraints, no meteorological records were collected at the experimental site over the period 2000–06; therefore, the long-term climatic data have been sourced from the combined Armidale data recorded by the Bureau of Meteorology (BOM). Over the time that records have been kept for Armidale, four collection sites have been used for BOM data, and this has contributed slight variation to data over time. The current Armidale BOM station (Tree Group Nursery; station 056037) has complete daily temperature records to World Meteorological Organisation (WMO) standards from 1 January 1981 onwards, and hence, data from this site were used for calculation of the number of frost days per month. Severe frost days were defined as those with a screen temperature ≤−5°C. A comparison has shown that this site has a climate similar to the site of the Cicerone farmlets (George et al. 1977), although there may be slight differences on a daily basis due to site-specific rainfall events and cloud cover. Detailed historic analysis of the climate of the CSIRO McMaster Research Station from 1946 to 1975, where the Cicerone farmlet trial was located, has been provided by George et al. (1977).

During the Cicerone farmlet experiment, decisions in response to climatic events were made by a producer-led Board (Sutherland et al. 2013), which, in conjunction with the farm manager, managed the running of the experiment according to the guidelines detailed in Scott et al. (2013c). Supplementary feeding strategies were implemented based on data regularly collected on both pastures and animal liveweights and fat scores (Scott et al. 2013c). It is acknowledged that, as on real farms, compromises sometimes had to be made in managing stock and pastures, as various factors required adjustment including stocking rate, supplementary feeding, pasture renovation, resting of pastures, and maintaining critical livestock fat scores according to the treatment protocols (Scott et al. 2013c). However, every effort was made not to compromise the integrity of the farmlet treatments (Scott et al. 2013c), which were essential to the conduct of the experiment. The winter stocking-rate decisions were made each autumn (April) and were set by the number of ewes to be joined and the number of wethers to cull. Trading cattle were brought onto the farmlets as opportunities arose to capture surplus growth over the spring–autumn period. Further details of the pasture renovation carried out, the levels of fertiliser applied, and the stocking rates and grazing regime have been described by Scott et al. (2013c), whereas soil fertility changes over time on each farmlet have been described by Guppy et al. (2013).

Funding constraints limited the range of data that could be collected, particularly with respect to soil moisture and water-balance measurements such as runoff and deep drainage. To overcome this constraint and provide insight into the effect of climate on the results of the Cicerone farmlet experiment, daily available soil water was estimated using a modelling approach as part of investigations into optimising grazing management and pasture renovation (Behrendt et al. 2013a), as well as soil fertility (Behrendt et al. 2013b).

Modelling of soil moisture was carried out using the model AusFarm (CSIRO 2007), previously known as Farmwi$e (Moore 2001), which is a component of the GrazPlan suite of decision support tools (Donnelly et al. 1997, 2002). AusFarm is a complex, multi-component, biophysical model that utilises the climate, soil moisture, pasture, and livestock sub-models from other GrazPlan decision support tools (Moore 2001). The AusFarm simulations utilised Armidale’s long-term daily climate data to estimate available soil water as well as pasture and livestock production during both the experimental period and over the long term. The modelled soil water data during the experimental period were compared with both calculated percentiles and cumulative soil water values.

In the AusFarm simulations, a default duplex soil profile with a depth of 700 mm with five layers was used to define the soil type for the Cicerone Project site (A horizon: 0–15, 16–150, 151–300 mm; B horizon: 301–500, 501–700 mm). The choice of this soil profile is supported by earlier research that found duplex soils to be dominant in the experimental area (Schafer 1980; McLeod et al. 2006). Through comparisons of modelled pasture herbage mass with measured data from the farmlet experiment (Shakhane et al. 2013a), primary pasture variables were manipulated in AusFarm iteratively to calibrate the model. This resulted in rooting depth being restricted to a total depth of 500 mm and the fertility scalar being set at 0.85, in order to limit modelled herbage mass to levels similar to measured values. The calibration of the AusFarm model against Cicerone farmlet data also required additional management scripts to suppress the winter growth of phalaris (Phalaris aquatica), the dominant species within the desirable sward.

Botanical composition data for individual species measured in the Cicerone farmlet experiment were categorised into functional groups to help identify representative species of the most important groups for modelling in AusFarm (Behrendt 2008). This approach is supported by Kelly and Basford (2000), who described the process of grouping species on a functional basis so that the difference between species within groups is less than between groups. In the Cicerone farmlet experiment, pasture composition was assessed using the BOTANAL dry weight technique (Tothill et al. 1978; Shakhane et al. 2013b). The allocation of species to functional groups was based on species descriptions documented by Whalley et al. (1976), Kahn et al. (2003), Wheeler et al. (1982), and Auld and Medd (1987).

Wavelet transformation of rainfall and modelled available soil water data was used as a statistical technique to explore the statistical significance of differences between the experimental period and the long-term climatic data. Estimates of available soil water and records of rainfall over the years in which the Cicerone farmlet experiment was conducted (2000–06) were compared with the data calculated for the years from 1929 to 2007, excluding the 2000–06 data. The profiles of these measurements over time were derived by aggregating daily data into fortnightly averages (available soil water) and totals (rainfall) and smoothed using the wavelet transform procedure of Nason (2008). The wavelet transform requires that the time series be of a length that is a power of 2, and this was obtained by the subset of the long-term climate data (1890–2007), from 1929 onwards. The coefficients produced by this transform provide a measure of the rainfall or available soil water at fortnightly frequencies. Minor coefficients, corresponding to noise, are eliminated by the process and the smoothed time series is reconstructed from the detail (or signal) coefficients. The random component of the time series (residuals) was calculated from the difference between the actual and smoothed time series, and this was used to calculate the average standard error and hence the confidence regions for the mean profile from the years 1929–2007 (excluding 2000–06). The areas between the profiles and the 95% confidence region for both rainfall and modelled available soil water were also calculated in order to compare the extent and duration over which profiles were either below or above the confidence region.

The CCD concept developed by McCaskill and Blair (1988) was modified to allow the calculation of what we have termed soil moisture stress days (SMSDs). These represent days when the overall pasture, consisting of both perennial grasses and clover, would most likely have been subjected to moisture stress. SMSDs are those days when all soil layers down to 300 mm of soil depth were found to have a modelled soil water content of <17% of the plant-available water-holding capacity, which is similar to the threshold used by McCaskill and Blair (1988). This summary variable in AusFarm ranged between 0, which represents the lower limit of plant extraction of soil moisture (i.e. wilting point), and 1, which is the drained upper limit (i.e. field capacity) (Anon. 2006). Thus, the AusFarm simulation data from the long-term climate data sequence, which included the experimental period, were used to calculate the number of SMSDs during the experimental period against those calculated from the long-term climatic data (1890–2007).

The monthly SOI values for both the experimental period and over the long term were obtained from the BOM (Bureau of Meteorology 2011).

The methods developed by Fitzpatrick and Nix (1975) allow the calculation of growth indices without the need to employ the full AusFarm model; these indices, which describe the climatic limitations to pasture growth across Australia, were used to indicate the extent to which the climate experienced during the farmlet experiment might have restricted pasture growth compared with the long-term climatic record. In this way, we were able to readily calculate growth indices which were not only of use in placing the climate experienced in context but were also used as climatic covariates in other related papers in this Special Issue of Animal Production Science concerning the balance between pasture supply and animal demand (Shakhane et al. 2013c) and pasture herbage mass and growth (Shakhane et al. 2013a).

As long-term daily solar radiation data were not available for Armidale, daily solar radiation data from 1971 to 1983 (George et al. 1977; with supplementary CSIRO research station data supplied by CSIRO staff) were used to construct an average daily light index using the equation for a C₃ metabolic pathway provided by Medd and Smith (1978). The temperature index was calculated from the same daily maximum and minimum temperature data for Armidale from 1890 to 2007 that were used as inputs to the AusFarm modelling. The soil moisture index data were derived from a weighted average of the available soil water in the three model layers comprising the top 300 mm of soil for ‘desirable’ pasture species modelled using AusFarm. The daily growth indices were calculated by multiplying the average daily light index values by the long-term daily temperature index and soil moisture index values.

Figure 1 compares the monthly mean climatic data for the experimental period against the long-term climatic record. The mean annual rainfall over the years 1890–2007 was 745 mm, with ~66% of this rainfall falling between October and March. In addition, Fig. 1 shows that rainfall received during winter was generally lower than the long-term median for all years except 2006.

Fig. 1.  Actual mean monthly maximum and minimum temperatures (thick lines) compared with the long-term average temperature data (thin lines), and the monthly rainfall (bars) compared with long-term average rainfall (thin line) for 1890–2007. (Source: AusFarm based on Bureau of Meteorology daily climate records for Armidale, NSW.)

The variation in daily minimum and maximum temperatures during the experimental period compared with the long-term climate record (Fig. 1) indicates that both the average daily maximum and minimum temperatures were higher during the summers of 2003, 2004, and 2006 than the long-term average maximum and minimum daily temperatures. It also indicates that, during the winters of 2000, 2002, 2004, and 2006, the average daily minimum temperatures were somewhat lower than those over the long-term climate record. Over the experimental period, there were substantial variations from long-term average temperatures, which are subsequently presented in relation to the frequency of severe frosts.

Figure 2a shows the profiles of rainfall over each year of the Cicerone experiment in conjunction with the mean profile and its 95% confidence region from the other years. Similarly, Fig. 2b shows profiles for the modelled available soil water data. To give a measure of the durations and extent to which rainfall and modelled available soil water over the Cicerone years (2000–06) were outside the 95% confidence region of the long-term averages, the areas between the confidence region and the profiles below or above them were calculated (Table 1). Over the experimental period, less rainfall was received than would have been expected based on the long-term climatic data. In addition, lower levels of available soil water were commonly experienced over the duration of the experiment. The area below the confidence region for rainfall was lower in 4 of the 7 years, while for available soil water, the area below the confidence region was lower in 6 of the 7 years.

Fig. 2.  Profiles of (a) total rainfall and (b) average available soil water index over each fortnight from January 2000 to December 2006 (thin lines) compared with the mean profiles of both parameters (thick black lines) and their 95% confidence regions (grey shaded areas) derived from long-term data (1929–2007, excluding 2000–06) following use of wavelet transformations (see text).

Table 1.  Calculated areas between the profiles of rainfall and modelled available soil water over the years of the Cicerone experiment (2000–06) when they were lower or higher than the 95% confidence region based on long-term climatic data (1929–2007, excluding 2000–06)
Areas between profiles and confidence regions shown in Fig. 2. Data in bold indicate the areas with the greater absolute value (the area between each profile and confidence region) within the rainfall and available soil water columns

The potential for pasture growth during the winter periods was severely limited by cold temperatures. Figure 3 shows the number of frost days recorded at the Armidale weather station during the experimental period compared with the medium-term climate record (1981–2011). The number of frost days experienced during the autumn and winter of 2001, 2002, 2004, 2005, and 2006 exceeded the monthly mean number of frost days over the medium-term record. Note that these frosts generally occurred during dry winter periods, which commonly experience little cloud cover, resulting in severe frosts as shown by the higher frequency of screen temperatures below −5°C during most of the winters throughout the experimental period (Fig. 3).

Fig. 3.  Observed number of frosts per month during the Cicerone farmlet experimental period compared with the monthly mean number of frosts over the period January 1981–October 2011. Upper panels, all data; lower panels, for those days when the minimum screen temperature was ≤−5°C.

The modelled available soil moisture index over the experimental period 2000–06 commonly showed months of below-median soil moisture conditions, with relatively fewer months of above-median conditions. When these data are presented as cumulative soil moisture index values (Fig. 4), it is obvious that, especially from autumn 2002 onwards, the cumulative effect of the imbalance between below- and above-median soil moisture conditions resulted in the experiment not reaching median or above-median cumulative soil moisture conditions. A line representing the cumulative 40-percentile soil moisture conditions allows the observation that the soil moisture values were approximately mid-way between the cumulative 40-percentile, which is the cumulative sum of the 40-percentile values for each month, and cumulative median soil moisture values, which are the cumulative sum of the median values for each month. The parts of the cumulative curves with the lowest slope correspond to the periods of lowest soil moisture, whereas the steepest slopes occurred during the relatively few months with above-median available soil moisture conditions. In fact, the Cicerone farmlets experienced below-median soil moisture conditions for 3, 4, or 5 consecutive months during the autumn–winter of 2000, 2002, 2004, 2005, and 2006. Thus, low soil moisture conditions had a large cumulative effect on winter feed supply in 5 of the 7 years of the experiment (Shakhane et al. 2013a) as pasture growth was constrained by both drier and colder than average conditions in winter.

Fig. 4.  Cumulative available soil water index for the Cicerone experimental period compared with the long-term median data and associated long-term 40-percentile level modelled using climate data from 1890 to 2007. (Source: AusFarm based on Bureau of Meteorology daily climate records for Armidale, NSW.)

The cumulative number of SMSDs per month, shown in Fig. 5, compares the number of SMSDs over the Cicerone experimental period with the long-term record. It shows that the cumulative number of SMSDs exceeded the cumulative long-term median values, especially from 2002–03 onwards. The divergence from median conditions demonstrates that the levels of moisture stress imposed on pastures were clearly much greater than would be expected if average conditions prevailed over the duration of the experiment. This was especially noticeable during periods when the divergence increased during the period 2002–04.

Fig. 5.  Cumulative number of soil moisture stress days (SMSDs) to 300 mm soil depth from January 2000 to December 2006 compared with the cumulative average number of SMSDs over the long-term (1890–2007).

Figure 6 shows the annual sum of the number of SMSDs per year over the period 1890–2007. Some years that are widely recognised historically in eastern Australia as having been dry (e.g. 1965, 1982, and 2002) or wet (e.g. 1959, 1973, and 1983) are identified as reference points. It is clear that, over the experimental period (shaded area), the annual number of SMSDs was higher than the long-term mean of 60 days for 4 of the 6 years, with an average from 2000 to 2006 of 75 days/year.

Fig. 6.  Yearly number of soil moisture stress days (SMSDs) to 300 mm soil depth from 1890 to 2007. Horizontal line is the mean number of SMSDs (to 300 mm soil depth) per year from 1890 to 2007. Grey bar shows the experimental period 2000–06.

Another measure of the climate experienced over the experimental period is provided in Fig. 7, which shows the SOI relating to sea surface temperatures for each month of the period 1890–2007, including the 6.5-year trial period. Over the experimental period, it is clear that the value of the index was negative (associated with lower rainfall in eastern Australia) in more months than it was positive, especially in 2002, 2004, and 2006.

Fig. 7.  Monthly Southern Oscillation Index (SOI) from 1890 to 2007 (source: Bureau of Meteorology) with experimental period (2000–06) shown by grey bar.

Figure 8 shows the annual pattern of changes in indices of light (LI), temperature (TI), soil moisture (MI), and growth (GI) calculated using the procedures of Fitzpatrick and Nix (1975). The seasonal trends in average growth indices indicate that, over the period April–September, temperature tends to be the most limiting factor governing pasture growth. For the remainder of the year, from October through to the next March, soil moisture, on average, is the most limiting factor. It is also noteworthy that conditions for growth are climatically constrained throughout the ‘average’ year, with the growth index never rising above 0.45.

Fig. 8.  Average monthly light, temperature, soil moisture, and growth indices for Armidale, NSW, based on long-term climatic data and modelled soil water derived from the AusFarm model.

The long-term average trends in monthly temperature index are compared with the actual temperature index calculated for the experimental years (2000–06) in Fig. 9. Clearly, the values over the experimental years are closely aligned to the long-term average, with few aberrations. In contrast to the temperature index, changes in the soil moisture index (Fig. 9) show substantial deviations, both above but especially below the long-term average line. Thus, the monthly soil moisture index was above the 70-percentile region in 13 months, whereas it was below the 30-percentile boundary in 23 months. Figure 9 also shows a comparison of the long-term growth index with that calculated for each month of the experimental years. Although occasional peak growth periods are evident (e.g. during late summer in 2002, 2004, and 2005), it is also clear that there were many months when the growth index was at or near zero, especially during the particularly dry winters of 2002, 2004, and 2006 when low levels of soil moisture index were experienced. Thus, these climatic indices show that below-average soil moisture was the feature that most affected the potential for growth over the duration of the field experiment.

Fig. 9.  Long-term monthly average (—) compared with the actual monthly (●) values of temperature index, soil moisture index (with the 70–30-percentile region shaded), and growth index calculated for the experimental years (2000–06).

In this paper, we have compared the climatic conditions experienced by the Cicerone farmlet experiment with the long-term climate record for the region. This has been achieved through the comparison of local climatic data during the experimental period (2000–06) against the long-term record (1890–2007), and through the calculation and assessment of indices known to reflect pasture production. The multi-pronged approach taken has explored differences between short- and long-term records of temperature, rainfall, soil moisture either as an index or as SMSDs, SOI, and calculated growth indices.

Comparison of climatic data revealed that, during the experimental period, the Cicerone farmlets experienced colder and drier winters than the average conditions over the long-term climate record, with more severe frosts occurring over most winters, with the exception of 2003. Both rainfall and available soil water were found to be significantly lower than the long-term climate record over most of the experimental period. How these climatic conditions combine to influence pasture production was assessed through the use of a biophysical model to derive SMSDs, a development from the concept proposed by McCaskill and Blair (1988), and growth indices described by Fitzpatrick and Nix (1975).

These analyses enabled the results of the farmlet experiment to be put into the context of longer term climate variability and the effects of these climatic constraints on pasture production and subsequent farmlet performance. This approach of using models to understand the variability of soil water content has also been used by other researchers, such as Stafford Smith and McKeon (1998) and Lodge and Johnson (2008b). Such studies describe the relationship between soil water content and rainfall over previous consecutive months to define drought or marginal conditions for grassland growth. It is clear that some of the extended periods of several consecutive months of low soil moisture conditions observed in this experiment qualify as ‘drought’ conditions as specified by Lodge and Johnson (2008a).

Although modelling is valuable, as pointed out by Specht (1981) and Bryant and Snow (2008) there is a need for models to adequately account for soil fertility in combination with climatic constraints. It is noteworthy that although livestock producers may be unwilling to adopt higher fertiliser strategies during dry periods (McCaskill and Blair 1988), significant responses to increased soil fertility were observed in the drier than average conditions over this field experiment (Guppy et al. 2013) in both pasture (Shakhane et al. 2013a) and livestock (Cottle et al. 2013; Hinch et al. 2013) production.

Measurement of soil water both within and below the root-zone, as carried out in an earlier trial on an adjacent experimental area (McLeod et al. 2006), would have been desirable; however, resource constraints meant that this was not feasible in the Cicerone farmlet experiment. In these circumstances, modelling has provided valuable estimates of soil water levels and changes over both the short- and long-term. The AusFarm modelling derived a soil moisture index profile of similar shape to that calculated for the same region by Smith and Johns (1975).

During the experimental period, the farmlets experienced few times of above-median conditions that would have allowed production responses to changes in species composition and soil fertility (spring 2005 was an exception). The summaries of modelled soil water over the experimental years show that the Cicerone farmlet experiment was always below cumulative median moisture conditions. In addition, the farmlets experienced comparatively low winter temperatures during the experimental period, with severe frosts occurring during 2001, 2002, 2004, 2005, and 2006. These constraints would no doubt have restricted the responses to pasture inputs and grazing management; evidence of this restriction can be seen in related papers dealing with herbage mass (Shakhane et al. 2013a), remote sensing of normalised difference vegetation index (NDVI), which is related to potential pasture growth (Donald et al. 2013), and livestock weights (Hinch et al. 2013). In the same region, Cook et al. (1976) reported that differences in tolerance to soil fertility and soil moisture constraints could lead to substantial shifts in pasture composition, especially between temperate (C₃) species such as Lolium perenne and tropical (C₄) species such as Bothriochloa macra. Over the duration of the experiment, large changes in the ratio of C₃ to C₄ species were observed among farmlets due to treatment effects (Shakhane et al. 2013b).

A comparison of SOI values indicates that they were negative more often than positive during 2002, 2004, and 2006. These dry years are consistent with the pattern of years with lower potential pasture growth found with NDVI images of the Cicerone farmlet experiment (Donald et al. 2013). In addition, climate specialists have noted that fluctuations in the Indian Ocean Dipole (IOD) index are linked to important changes in Australia’s climate, including the recent period since 1995, which they have described as the ‘Big Dry’ (Ummenhofer et al. 2009; Verdon-Kidd and Kiem 2009). This is consistent with the frequency of above- and below-average winter rainfall experienced in the region identified by Carberry et al. (2005).

The analysis of growth indices showed that pasture growth rates in this region are, on average, severely climatically constrained, predominantly by temperature from April to September, and by soil moisture from October to March. In addition, the majority of above-average growth indices occurred during summer when C₄ species have an advantage over C₃ species. However, the drier than average conditions during spring would have been likely to restrict growth of the legume best adapted to the region, white clover (Hutchinson et al. 1995). This would have potentially constrained the response to phosphorus applications due to limited nitrogen fixation by the legume, which in turn would have limited the amount of nitrogen cycled, resulting in restrictions to grass growth (Sale and Blair 2005).

It is important that livestock producers on the Northern Tablelands appreciate these constraints and consider how they can be managed in spite of the complex interactions of climate with every aspect of a grazing enterprise. The variability of the response in the grazing system under both drier and wetter years is required to more fully understand the interactions between climate and the management strategies studied in the farmlet experiment, namely pasture renovation, fertiliser application and grazing management. The management of the farmlets in response to climatic uncertainty has been conceptually described as a complex decision problem that is familiar to graziers who commonly have to make decisions in spite of imperfect knowledge of the state of a farm’s resources and an uncertain climate (Behrendt et al. 2013a). Rational responses to these climatic constraints include employing deep-rooted perennial species that permit some growth over dry periods (Boschma and Scott 2000) and ensuring the retention of sufficient temperate C₃ species, which, especially when adequate soil nitrogen is available, are capable of some modest growth under cold winter temperatures (Newman et al. 1962).

Although severe frosting occurred during the experimental period, this was not evident in the comparison of the short-and long-term temperature index values (Fig. 8). This suggests that the ‘temperature averaging’ procedure used by Fitzpatrick and Nix (1975) may not adequately reflect the effect that severe frosting can have on pasture growth in this region. Similarly, in the GrazPlan suite of models, plant growth is governed by a surrogate for soil temperature, which is computed from mean temperatures during daylight hours (Moore et al. 1997). To enable calibration of the AusFarm model against Cicerone farmlet data, additional management scripts were required to suppress the winter growth of phalaris, the dominant species within the desirable sward (Behrendt 2008). This was necessary, as the temperature-limiting factor for phalaris insufficiently constrained growth compared with other C₃ species available in the GrazPlan suite of decision support software (Behrendt 2008).

The range of climate experienced over the experimental period was biased towards drier than average conditions. Ideally, given the marked influence climate has on pasture and livestock outcomes (Jones et al. 1995; O’Reagain et al. 2011), the length of the experimental period for farming systems research should be sufficient to allow adequate representation of the expected range of climatic conditions (Walker 1988). This would enable sufficient demonstration of treatment responses and would reduce the need for assumptions or modelling to extrapolate responses to climatic conditions for which there are no data to support calibration. The analyses reported here reinforce the case made by Jones et al. (1995) that funding organisations need to support dryland farming systems research trials for periods sufficient to allow for realistic appraisals of the systems under investigation.

Accepting the results of the experiment without consideration of potential outcomes under an alternative sequence of years could lead to erroneous conclusions. This paper has attempted to provide a climatic context within which the results presented in related papers, such as changes in botanical composition (Shakhane et al. 2013b), herbage mass and quality (Shakhane et al. 2013a), liveweights (Hinch et al. 2013), economic risk (Scott et al. 2013a), and integrated outcomes (Scott et al. 2013b) can be viewed. Because these results were obtained over a limited period, several other related papers in this Special Issue have focussed on system optimisation over the longer term (Behrendt et al. 2013a, 2013b; Scott et al. 2013a).

Through a comparison of climatic data over the experimental period, and the use of biophysical modelling and growth indices against the long-term record, this paper has demonstrated a method of assessing whether a dryland farming system experiment experienced a sufficiently representative climate for the region during the trial period. Key deviations from the expected long-term climate for the region were identified in the frequency of severe frosts during the abnormally dry winters of 2002, 2004, and 2006. In addition, the analysis of rainfall, available soil water data, and SMSDs showed that the Cicerone farmlet experiment experienced moisture-constrained conditions throughout most of the experimental period. This was supported by the comparison of growth indices against the long-term indices. The analysis suggests that the Cicerone farmlet experiment did not experience climatic conditions typical of the long-term record. This would have restricted the responses observed to pasture inputs and grazing management, and hence the results from the experiment should be interpreted with due regard to climatic expectations over the long-term.