Pasture production from a long-term fertiliser trial under irrigation

Abstract

New Zealand pastoral farming has benefited greatly from the application of phosphorus (P) and sulphur (S) fertilisers supplied in the main by superphosphate (SSP). The long-term fertiliser trial at Winchmore, mid-Canterbury, New Zealand was set up in 1952 and has yielded a wealth of data on the effect of fertiliser, grazing by sheep and flood irrigation on pasture production. The trial was initially (1952–1958) designed to measure the effect of no fertiliser and SSP applied each year at 188, 376 and 564 kg ha⁻¹. From 1958 to 1980, SSP applications were stopped to the 564 and one half of the 376 kg SSP y⁻¹. The cessation of fertiliser decreased clover content and increased the proportion of weeds and low-fertility grasses. The decline in production once fertiliser application ceased followed a curvilinear pattern, but never reached the low production of the no fertiliser treatments even after 20 years. In 1980, the residual treatments were changed to compare a reactive phosphate rock (RPR)/S treatment as well as an intermediate (250 kg ha⁻¹) rate of superphosphate. The 0, 188 and 376 kg SSP ha⁻¹ have now been unchanged for 60 years, while the 250 kg SSP or equivalent in RPR/elemental S have remained unchanged for 30 years. The 188 and 250 kg SSP ha⁻¹ y⁻¹ treatments have shown that without any nitrogen fertiliser, ryegrass and clover will persist in irrigated pastures and result in high levels of pasture production (11–12 t ha⁻¹) for up to 60 years.

Keywords:

• long-term
• phosphorus
• sulphur
• pasture production
• pasture quality
• grazing
• irrigated pasture

Introduction

Phosphorus (P) and sulphur (S) have long been regarded as the key nutrients for the successful growth of pastures in New Zealand pastoral farming systems. In supplying this requirement, the main phosphate fertiliser form used over many years has traditionally been superphosphate (SSP), which has the added bonus of also containing sulphur (S) and calcium (Ca) (During 1984, McLaren & Cameron 1990). Both P and S are required by the grass–clover system used in New Zealand for growth, especially P, which boosts clover production and in turn facilitates nitrogen (N) fixation (During 1984). Over time, the emphasis has shifted from the development of pasture with high fertiliser inputs to a maintenance regime where fertiliser inputs are adjusted to match losses from the soil–plant–animal system (Cornforth & Sinclair 1982). In addition, the high rates of fertiliser used in some areas have resulted in a build up of soil reserves of P and S which, combined with increasing fertiliser cost, has seen a lessening of fertiliser use or in some cases seen a complete cessation of fertiliser application (Quin & Scobie 1985). Another shift has seen, especially in the early 1980s, a move by some farmers away from SSP towards the use of reactive phosphate rock (RPR), plus sulphur as required (Quin & Scobie 1985).

Much research has been carried out to ensure that current P fertiliser application rates are sufficient to maintain high pasture production via an optimal Olsen P concentration (Cornforth & Sinclair 1982, Sinclair et al. 1997, Roberts & Morton 1999, Morton & Roberts 2010, Edmeades et al. 2006). The majority of these field trials were conducted under mowing with stock excluded (e.g. Sinclair et al. 1997) and ran for a maximum of 4–6 years. The relationship between P fertiliser and pasture production has been found to be similar under grazing to that found under mowing, despite some differences in absolute pasture production between the two trial techniques (Morton et al. 1995, Morton & Roberts 2001). In addition, research has been carried out to measure the effectiveness of RPRs compared with SSP, again with most trials running for 3–6 years (Mackay 1990, Sinclair 1990, Smith et al. 1990, Sinclair et al. 1993, Sinclair et al. 1998).

While there are a few fertiliser trials in New Zealand that have run longer than 4–6 years (e.g. at Whatawhata, Ballantrae and Winchmore), the longest running trial by far has been at Winchmore (Nguyen et al. 1989). There are several advantages of such long-term trials, including the ability to highlight the effectiveness of slow-release fertiliser products, long-term changes in C, N, S and P cycling (Nguyen et al. 1989, Murata et al. 1995, Blake et al. 2000), P accumulation and losses (Condron & Goh 1990, McDowell et al. 2003), losses and cycling of fertiliser contaminants such as Cd (McDowell 2010), and the ability to account for nutrients in the soil–plant–animal system via a mass balance approach (Williams & Haynes 1992). Datasets from long-term trials such as that at Winchmore have also been used successfully to construct, modify and validate both nutrient cycling and pasture growth models (Metherell et al. 1995, Romera et al. 2009).

In writing up the first 30 years of this, trial Nguyen et al. (1989) highlighted several key production changes that occurred with the cessation of P application. One was that the accumulation of S and P reserves made a significant contribution to ongoing pasture productivity. This paper reviews and expands on this data, examining changes in pasture production over the 60 years of the trial.

Methods

Trial site

The trial has been previously described in detail by Rickard (1968), Rickard & McBride (1987) and Nguyen et al. (1989). Its history is given by Rickard & Moss (2012). Briefly, the trial is located on a Lismore stony silt loam soil (Pallic Orthic Brown soil; Hewitt 1998) situated on the Winchmore Irrigation Research Station (171°48′E; 43°47′S) in mid-Canterbury, New Zealand. Winchmore has a mean annual rainfall of 745 mm relatively uniformly distributed throughout the year. The trial was initiated in 1952 to investigate the effect of different rates of SSP on grazed pasture irrigated by the border-strip method (Taylor 1981). Details of the cultivation and pasture establishment are given elsewhere (Rickard & Moss 2012)

Treatments and trial design

The trial design included four replicates of five treatments in a randomised block design. Plots (0.09 ha) within a treatment were grazed by separate mobs of sheep that rotated between the replicates during the September–May growing season. The stocking rates were set and adjusted so that pasture utilisation for each treatment was 80%. To avoid fertility transfer onto the trial, all stock were fasted for 30 h prior to entering the trial. All treatments within a replicate were grazed simultaneously.

The trial was flood irrigated by the border strip system (Taylor 1981) where irrigation water flowed down the strip at a mean application rate of 80–90 mm per application. Water was applied whenever gravimetric soil moisture content (w/w) in the top 10 cm depth was approx 15% w/w until the winter of 1996, and at 20% w/w following that date, the change being made to better reflect typical farm practice. There were, on average, five irrigations per year. Wilting point and field capacity were 10 and 30% w/w soil moisture respectively. There were several occasions over the duration of the trial when this irrigation scheduling system was not applied, for a variety of reasons. These occurred in 1954 and 1984 (Rickard & McBride 1987) and latterly in 2007 and 2008.

Between 1952 and 1958, the trial evaluated the effects of five different SSP treatments (Rickard & Moss 2012) initially applied in autumn, but in later years applied in late winter (Rickard and McBride 1987; Nguyen et al. 1989). From 1958 to 1980, SSP applications were stopped to the 564 and one half of the 376 kg SSP ha⁻¹ y⁻¹ treatments. These are denoted as residual (R) treatments. The control, 188 kg ha⁻¹ and 376 kg ha⁻¹ SSP treatments have remained unchanged since 1952. From 1980, the residual fertiliser treatments were replaced by a 250 kg ha⁻¹ SSP treatment and a Sechura RPR/sulphur blend applied at rates of P and S equivalent to the 250 kg ha⁻¹ SSP treatment (McBride 1992, Rickard & Moss 2012). Superphosphate quality has varied over the years, but has typically been 9.0% P and 11.0% S. However, from the late 1960s to the late 1970s, the product applied was ‘Flowmaster Super’ with a registered total P content of 8% (Rickard & Moss 2012).

Measurements and analysis

Pasture production was measured using exclusion cages (2.75 m×0.61 m), with two cages per plot (Lynch 1966). Soil samples for analysis were taken annually to 7.5 cm depth. Details of these samples and the main results from these soil samples are presented elsewhere (McDowell 2012; McDowell & Condron 2012).

Data from the trial for the early years (1952–1989) was obtained from the database of P, S and K research trials set up in 1990 (Feyter 1993) and described by Edmeades et al. (2006). The data was in the form of individual treatment means plus the standard error (SE) for the individual harvests as well as annual totals. From 1990 until the present, all data were analysed by analysis of variance (ANOVA). Annual production figures presented in graphical form were calculated from the sum of individual harvests, which in some cases meant that the growth covered a period slightly longer or shorter than 12 months, depending on harvest dates.

Monthly growth rates (kg DM ha⁻¹ day⁻¹) from 1980 to 2010 were calculated from daily growth rates for the period between each harvest date, allocated to the appropriate month and compared between treatments using a repeated measurement model (REML).

The relationship between pasture yield and Olsen P concentration at the end of each year (McDowell & Condron 2012) was fitted for each SSP treatment annually from 1980 to 2010 using a least-squares fit Mitscherlich relationship. While Mitscherlich curves could be fitted for all years, the SE of the asymptote was, in some cases, quite large. Large SEs in asymptotes can give large SEs in relative yield (RY), and such data are unsuitable for examining the Olsen P–RY relationship (Sinclair et al. 1997). In this work, any datasets where the SE of the asymptote was greater than 10% were thus omitted. This excluded eight of the datasets, leaving 22 datasets to be used.

Results and discussion

Pasture growth pattern 1980–2011

The growth pattern was characterised by low winter (May–August) production followed by a rapid increase in growth rates over spring (Fig. 1). Production peaked in October and November at approximately 60 kg DM ha⁻¹ day⁻¹ before a small decline in December to c. 50 kg DM ha⁻¹ day⁻¹ that was maintained over the summer period (December–February) before a steady decline over the autumn. About 74% of annual growth occurred in spring and summer, with 17% in the autumn and only 8% in the winter (Table 1). This production pattern was similar to that previously measured on irrigated pastures at Winchmore (Nguyen et al. 1989) and other sites in Canterbury (Rickard & Radcliffe 1976) and North Otago (Greenwood & Sheath 1981, McNamara 1992). While there are some relatively small variations in seasonal production between these reported results and the current data, particularly over the spring, these can be accounted for in seasonal temperature variations, differing measurement intervals (Rickard & Radcliffe 1976) and the duration of the trial measurements.

Figure 1  Effect of P fertiliser treatments on the distribution of daily DM production for 1980–2011. Bars indicate the standard error of the difference between treatment means.

Table 1 Effects of long-term fertiliser application (kg ha⁻¹) on seasonal and annual pasture production (kg ha⁻¹) for the 1980–2011 growing seasons (July–July). The standard error of difference (SED) is given along with the F-statistic for treatment comparison (bold if significant).

Treatment	Spring	Summer	Autumn	Winter	Total
Control	1959	2092	861	406	5289
188 Super	4280	4289	1917	998	11428
250 Super	4333	4490	1976	1034	11798
175 RPR	4349	4270	1904	1066	11497
376 Super	4448	4738	2064	1027	12242
SED	405	495	491	219	1364
F-statistic	<0.001	0.002	0.202	0.030	0.006

Fertiliser application significantly increased growth rates for most months, with the control plots consistently producing less than half that of the fertilised plots (43–46%; see Table 1). This difference in production is similar that measured at the long-term Park Grass continuous hay experiment (Jenkinson et al. 1994) where production on the non-fertilised plots was only 40–50% of that measured on the fertilised plots, despite the nutrient cycling under grazing that occurred at Winchmore but not Park Grass. This clearly shows the benefit of continuing fertiliser applications and the change in sward to lower producing/low-fertility pasture species (e.g. hair grass Vulpia bromoides and browntop Agrostis capillaris) that occurred in the control (no fertiliser) plots (Fraser et al. 2011). There were no significant differences in pasture growth rates between fertiliser rates or forms (Fig. 1; Table 1). The combination of P and S in the fertilisers used in this study is important as Nguyen et al. (1989) noted the increased pasture production on the fertilised plots was due to both the P and S in the fertilisers applied. In particular, growth over spring and summer is very sensitive to both S and P inputs (Saunders et al. 1963, O'Connor et al. 1985). There is some suggestion that, for peak spring production, winter/spring applied S is possibly more important than P (Nguyen et al. 1989), particularly for soils with a low ability to retain S such as the Lismore stony silt loam (Sinclair & Saunders 1984). Soil sulphate levels measured in winter prior to fertiliser application have typically been in the range of 4 to 10 in all SSP treatments indicating continuing sulphur responsiveness.

1952–1980

There were considerable fluctuations in annual pasture yields from 1952 to 1980 irrespective of SSP treatment (Fig. 2). Such annual fluctuations have been noted in other long-term experiments (e.g. the Park Grass continuous hay experiment; Jenkinson et al. 1994) where they have been attributed in part to changes in annual climatic (rainfall) conditions (Silvertown et al. 1994). Some of the variance in our annual yields could be similarly explained by insufficient water for irrigation in dry years, such as 1954 (Rickard & McBride 1987). However, some of the fluctuations in latter years could also be attributed to a combination of grass grub (Costelytra zealandica) infestation (Nguyen et al. 1989) and the quality of the SSP available during the 1970s (Rickard & Moss 2012). Despite these fluctuations, the application of SSP resulted in significantly greater annual pasture production compared to the control. The lack of significant differences between the 188 and 376 kg ha⁻¹ SSP treatments suggests that 188 SSP is sufficient to maintain at least 90% of maximum pasture production under these management conditions (Fig. 2; Nguyen et al. 1989). This supports the findings of Morton & Roberts (2010) who suggested that for a mid-Canterbury sheep farm under irrigation, a P fertiliser rate of 23 kg P ha⁻¹ y⁻¹ (250 kg SSP) is sufficient for maintenance of both pasture production and soil Olsen P concentration. Under the experimental management regime used for this trial there was minimal loss of P in animal products (McDowell & Condron 2012), but there was considerable lateral transfer of P in animal dung from the centre to the sides or crutch of the irrigation border strip (Saville et al. 1997) and longitudinally in irrigation outwash (McDowell & Rowley 2008).

Figure 2  Effect of SSP application on annual DM production (t ha⁻¹) for the period 1952–1980. Bars indicate the standard error of the difference between treatment means.

The residual effect of the initial 6 years of fertiliser application (1952–1958) on pasture production was still evident 22 years later (Fig. 3; Nguyen et al. 1989). The decline in production was steady and commenced in the first year that fertiliser application ceased. However, the rate of fertiliser application prior to cessation had considerable bearing on the rate of pasture production decline (Fig. 3), as Nguyen et al. (1989) noted that the rate of yield decline was less in the 564R than the 376R treatments. The pattern of decline was approximately linear for the first 10 years (1958–1968) but remaining relatively steady thereafter (Fig. 3). Nguyen et al. (1989) expressed the first 10 years of decline in pasture production by linear regression equations

Figure 3  Annual DM production in the unfertilised control, 188 kg ha⁻¹ annually applied superphosphate and two residual superphosphate (376 and 564 kg ha⁻¹) treatments relative to the 376 annual applied superphosphate treatment (from Nguyen et al. 1989).

where 376R and 564R are the percentage declines in yield from residual treatments compared with the 376 SSP treatment and YR is the number of years after the cessation of SSP. They also expressed the percentage yield decline for 1968–1983 as

The higher application rate resulted in a longer period before the pasture production decline became significant (Fig. 3). The short time (6 years) that fertiliser had been applied before applications ceased is also likely to have had an effect on the rate of pasture growth decline affecting soil P forms (McDowell & Condron 2012).

Although the above regressions were fitted for the decline in pasture production after 1968 there was still considerable annual variation. In 1968, the residual plots and the control showed a marked production decrease (Figs 2 and 3) whereas there was no such production decline in the SSP treatments. Decreases also occurred in 1971 and 1975, but here all treatments had a decline in pasture production. In the year after these decreases, most treatments saw an increase in pasture production back to previous levels, although for the latter events this took more than one season. If one assumes that, as indicated by Nguyen et al. (1989), these decreases in production are a result of grass grub infestation, and therefore can be ignored, the relationships are

where 376RY and 564RY are the percentage yield from residual treatments compared to the 376 SSP treatment and YR is the number of years after the cessation of SSP. These functions indicate that, even after 22 years, the 376R and 564R treatments still produced 46% and 59% of that from the 376 SSP treatment, whereas the control treatment produced 30–36%. This difference and the indication that the decline in production in the residual treatments appeared to have levelled off perhaps reflect the contribution of P and S via animals to pasture production. Nguyen et al. (1989) concluded that production was maintained more by increased soil P reserves particularly organic P, resulting from the previous above maintenance applications, than the direct faecal P contribution. However they showed that that cessation of fertiliser on some plots in 1958 resulted in an increase in weeds and grasses more suitable to the lower fertility and a decrease in clover content. It is likely that this species change, as well as the associated decrease in N fixation, contributed to the decline in pasture production in the residual treatments.

Table 2. Effects of long-term fertiliser application (kg/ha) on annual pasture production (kg/ha) for the fertiliser treatments that have been continually measured since 1952. The standard error of difference (SED) is given along with the F-statistic for treatment or year (1952–1980 vs. 1981–2011) comparison (bold if significant).

Treatment	1952–1980	1981–2011	Total 1952–2011	SED Year	F-Statistic year
Control	4492	5289	5017	569	0.186
188 Super	10028	11428	10918	1133	0.221
376 Super	11298	12242	11888	966	0.290
SED Treatment	694	1358	774
F-statistic Treatment	<0.001	0.021	<0.001

1980–2011

The sometimes large fluctuations in pasture growth between years continued from 1980 to 2010. However, overall there has virtually been no change in fertiliser response over 60 years from 1957–1980 compared to 1980–2010 (Table 2). Nevertheless, production appeared to fluctuate more and over a longer time period from 1980 to 1993 before levelling off to a more steady state until the drop in 2007 and 2008 (Fig. 4). Some of these fluctuations were caused by lack of water for irrigation (e.g. 1984; Rickard & McBride 1987) or, in the case of 2007 and 2008, water and irrigation scheduling issues over the critical summer months.

Figure 4  Effect of phosphate fertiliser application on annual DM production (t ha⁻¹) for the period 1980–2011. Bars indicate the standard error of the difference between treatment means.

Applications of SSP continued to result in, on average, 6 t DM ha⁻¹ greater pasture production than measured in the control plots (P<0.05). The period 1980–1990 saw a steady increase in pasture production from an average of 8 t DM ha⁻¹ to 12 t DM ha⁻¹, though the reasons for this increase are unclear. In contrast, production over the period from 1990 to 2007 was relatively static at 12 t DM ha⁻¹, in part attributable to more frequent irrigation since 1996. Production during the 2009–2010 and 2010–2011 seasons did not match that from 1990 to 2007, despite irrigation problems having been corrected. The reason is unclear, but may in part be due to change in clover content while irrigation was limited.

Following the capital application of P to residual treatments in 1980 (850 kg SSP ha⁻¹ or equivalent RPR; Metherell & Perrott 2001), the introduction of 250 kg ha⁻¹ SSP and RPR treatments took 1 (250 kg ha⁻¹ SSP) to 2 years (RPR) to reach production similar to that in long-term SSP treatments. This time lag in pasture production was not unexpected when one considers that, in 1979, pasture production on these two treatments averaged 50% of that from the 376 SSP treatments (Fig. 3). Indeed, Sinclair et al. (1993) showed that even where P had been applied at maintenance levels there was a significant drop in pasture production in year 1 of RPR use, requiring 3 years before plant available P and production was the same as that in plots receiving SSP continuously. However, it must be noted that at Winchmore the time lag was probably shorter due to the large initial capital P application as well as a favourable soil pH promoting the dissolution of apatite bound-P (i.e. 5.6 to 5.8; Sinclair et al. 1993) and adequate soil moisture via irrigation.

P response curves

The relationship between relative yield (RY) and soil Olsen P concentrations for the period 1980 to 2010 saw only a small scatter of points at the lower Olsen P concentrations, but a large scatter at higher Olsen P concentrations (Fig. 5). The least-squares fit of a Mitscherlich relationship between RY and Olsen P test for the SSP treatments was

Figure 5  Phosphate response curve for the period 1980–2010:

, superphosphate treatments;

, RPR treatment.

accounting for 78.4% of the variation in RY. The relationship included the sulphur response, so may differ in curvature from relationships derived from experiments with basal S fertiliser as reported by Sinclair et al. (1997) and Edmeades et al. (2006). The relative yield of the RPR treated plots relative to the SSP asymptote are also presented in Fig. 5 and show that, in general, the RPR relative yield was similar to that of the same rate of P as SSP (i.e. not statistically different). It would appear that the optimum Olsen P concentration for this site for near-maximum pasture production (i.e. 97%RY) was c. 19 µg/ml. This Olsen P concentration is just below the average of 20 µg/ml derived for sedimentary soils (Morton & Roberts 2010) and less than the critical level for sedimentary soils of 26–32 µg/ml reported by Edmeades et al. (2006), but falls within the scatter of field trial results presented in their paper. Furthermore, work by Sinclair et al. (1997) and Gillingham et al. (2007, 2008) showed that there is little advantage in sheep pastures of Olsen P concentrations greater than 20 µg/ml.

Botanical composition

Nguyen et al. (1989) showed that SSP application resulted in high-producing ryegrass and clover pastures in the initial years, and that cessation of fertiliser on some plots in 1958 resulted in an increase in weeds and poor producing/lower fertility grasses and a decrease in clover content. From 1980 onwards, there was considerable fluctuation in summer clover content, which increased following a large decline in clover in 2000 and 2001 (Fig. 6). Summer clover content in the fertilised plots decreased slightly from the 40% in the early 1980s to 30% over the 1990s. For the control plots, the clover content over summer also decreased initially but tended to increase following the minimum in 2000. It was noticeable that the decrease in summer clover content over 2000 and 2001 was considerably greater in the fertilised treatments (>20%) than the control treatment (<10%) and that fertilised plots had recovered to previous levels by 2003 whereas recovery was slower for the control plots. There were similar drops in summer clover content for the 2008 and 2010 seasons. Other researchers (e.g. Edmeades et al. 1990) have noted that there can be considerable variation in botanical composition, particularly clover, between years on a dryland site, and that the size of the P response can be partly related to botanical composition. While the Winchmore site is irrigated, limited irrigation in 2007–2008 may explain why there was lower clover content in 2010 and 2011 and lower pasture production. This clearly shows that if there is sufficient clover to fix N, adequate fertiliser (i.e. ≥188 SSP at Winchmore) can maintain high annual yields.

Figure 6  Effect of phosphate fertiliser application on clover content in the summer (January) sward for the period 1980–2011.

Conclusion

Monitoring of pasture production, under grazing and irrigation, over a 60-year period has demonstrated that both P and S are required for the maintenance of a high-producing ryegrass/white clover pasture. High levels of pasture production (12–14 t ha⁻¹) and pasture quality in a sheep grazing system have been maintained by annual applications of 188 to 250 kg SSP or equivalent in RPR/elemental S without N fertiliser. Optimum production (97% relative yield) was obtained at an Olsen P concentration of 19 µg/ml, with little benefit shown from greater Olsen P concentrations. For those fertiliser treatments that have continued unchanged for the full 60 years there has been virtually no change in fertiliser response (Table 2). This confirms the sustainability of superphosphate in supplying the P and S required for such pastoral systems.

Stopping the application of fertiliser had a major effect on pasture composition (especially clover content, which decreased) but with regular annual fertiliser applications, ryegrass and clover will persist in irrigated pastures. The decline in pasture yield when fertiliser is withheld is dependent on initial fertility (as indicated by Olsen P) but the effect of previous fertiliser applications was still apparent after 20 years.

Acknowledgements

The authors wish to acknowledge the contribution of a considerable number of technical staff, particularly JH Baird, SD McBride, GN Green and BM Stuart, and scientists DS Rickard, BF Quin and ML Nguyen. Thanks to J Carson for farm management, P Johnstone for statistical advice and the New Zealand Fertiliser Manufacturer's Research Association for funding.