Assessing the impact of input choices within Overseer (v6) on the modelled N losses to water for
Lincoln University Dairy Farm (LUDF)

Ron Pellow1, Steve Lee2, Alister Metherell3, Roy McCallum4, Jim Moir5, Ants Roberts6, David Wheeler7.

1South Island Dairying Development Centre, Lincoln University, NZ
2DairyNZ Ltd, Lincoln, NZ
3Ravensdown Fertiliser Ltd, Christchurch, NZ
4Grazing Systems Ltd, Palmerston North, NZ
5Lincoln University, Lincoln, NZ
6Ravensdown Fertiliser Ltd, Pukekohe, NZ
7Agresearch, Hamilton, NZ
ron.pellow@siddc.org.nz

Introduction:

Lincoln University Dairy Farm is a 160 ha dairy farm milking approximately 4 cows/ha on 160 ha irrigated pasture. The farm is focussed on grass based milk production, typically supplemented with small amounts of bought in grass silage and the majority of the cows wintered off.

The farm system was ‘primarily the same’ from 2004 through till 2010/11 when a system change was implemented to increase profitability through increasing productivity without increasing the farms total environmental footprint.

Production typically ranged from 1630 – 1720 kgMS/ha and around 400kgMS/cow up till and including 2010/11. Benchmarking of the farms profitability (operating surplus per ha) positioned the farm within the top 2-5% of Canterbury and NZ dairy farms. The system change in 2011/12 increased production 12.5% and profitability on a like for like basis (same payout) by 15%. Production rose to 470kg/cow and 1860kgMS/ha from 3.95 cows/ha. As the farms total environmental footprint includes its support land (grass silage, replacements and wintering), this exercise refers only to the milking platform.

paper1-1paper1-2

Data Collection:

LUDF is managed by the South Island Dairying Development Centre (SIDDC) on behalf of Lincoln University as a commercial demonstration farm. Daily and weekly data collection and transparency of inputs, strategies and results has created a substantial database of the farms performance over time.

Overseer 6.0

Recognising Overseer is a long term equilibrium model, intended to predict long term annual average outputs, the Lincoln University Dairy Farm provides a multi-year dataset to compare the effect over time of varying input choices, within a consistent framework.

LUDF’s data from an individual year reflects the tactical and management responses to seasonal conditions as they evolve. Inputting actual data over multiple years with a consistent methodology (see decision rules below) provides one means of assessing the effect of a set of decision rules on the outcome over time. The combined data set from consecutive years is likely to more closely resemble a long term average than any individual year.

N Loss to Water:

Only the N-losses to water are considered in this analysis, in part due to the greater emphasis currently placed on this output. Other losses of nitrogen remain of interest, both agronomically, in relation to the lost opportunity, and in the case of losses to the atmosphere, losses that may in the future form part of more specific on farm accounting for Nitrous Oxide emissions.

Decision Rules:

The phrase ‘Decision rules’ is used in this paper to describe the specific set of input conditions used across a range of years, to consider the impact of those inputs on the predicted loss of N from the rooting zone at LUDF. Multiple variations of decision rules could be used to drive a particular range of outputs if the user desired. In this set of examples, the default values were chosen, then largely, a single change or set of like changes were made in one area, using the actual LUDF data across years, while all other changes were left the same. Thus the results endeavour to explore the sensitivity of Overseer to various input fields but do not attempt to define a set of variables or combination of variables to deliver a specific outcome. The use of actual data from LUDF over time, rather than simple ‘what-if’ analysis enables the results to more closely resemble the intent of a long term average annual model.

Disclaimer:

The LUDF data used in the creation of the following Overseer scenarios is widely available and has generally been discussed in the context of the farms performance in many different forums. Other users of Overseer theoretically should be able to recreate the same scenarios and therefore arrive at the same conclusions. Nevertheless, errors may have occurred in data entry, or assumptions made, and, importantly, the results remain specific to LUDF and the input data used. Whilst the results show significance or otherwise of various decision rules on the predicted N loss for LUDF, this should not be taken to infer the same effect or significance will occur on other farms.

Standard Inputs:

Actual milk production per year, stocking rate, off farm winter grazing days, N fertiliser used and supplements purchased reflected actual farm practice each year. Eco-n (DCD) (as used each year) was included within the default values. Cow numbers were inputted based on peak cows milked, except for the comparison with monthly cow numbers. N-loss to water was then averaged across years for each set of decision rules.

Irrigation was defined as ‘Centre Pivot’ with Overseer calculating the irrigation volume and months of application (using deficit irrigation) as the default or standard input. Irrigation was available for application from September through to (and including) April.

Input choices considered:

The following input choices were considered in creating the decision rules for various scenarios using the LUDF dataset. These inputs were selected based on general discussions regarding the availability of accurate data on typical farms, expectation of effect on results, need to report for other purposes, and perception of uniqueness to farms like LUDF, compared to an average farm. For example the protein percentage of milk supplied is a little higher, suggesting more N may be exported than an ‘average farm’.

  1. Clover levels – using documented clover content on farm (see LUDF Focus Day handouts, including February 2010 and February 2011).
  2. Pasture quality (MJME/kgDM) and N concentration in pasture using feed quality analysis from each year
  3. On farm irrigation volume records
  4. Livestock numbers (specifying monthly livestock rather than enabling Overseer to calculate stock numbers from peak cows milked)
  5. Milk components (milk fat and protein percentage)
  6. Grouping some of the soil types together to reduce complexity of data entry
  7. Standardising the effect of cows wintered off farm
  8. Increasing the effluent area to the whole farm
  9. Changing from Centre Pivot irrigation (using Overseer calculated ‘deficit’ irrigation) to actively managed irrigation*
  10. Increasing average annual rainfall from the NIWA local grid prediction to the highest recorded annual rainfall over the past 6 years

* Overseer defines actively managed irrigation as the application method and management that results in no direct additional drainage from the irrigation application (ie no leakages, overlaps etc) and presumes no rain within 5 days after application.

Effects of Decision Rules at LUDF – averaged over 5 seasons:

Decision rules resulting in limited change in predicted N losses to water include:

  • standardising wintering off,
  • increasing the size of the farms effluent area to the whole milking platform and
  • changing the milk composition to reflect the slightly higher protein content of milk supplied by LUDF.

paper1-3

Decision rules resulting in decreased predicted N losses to water include:

  • using farm specific data for as many inputs as possible, except irrigation volume and timing. Farm specific inputs included pasture quality, clover content, replacement rates, cow liveweight, milk composition etc
  • active management of irrigation also resulted in decreasing the predicted N loss

paper1-4

Decision rules resulting in increases in the predicted N losses to water at LUDF include:

  • reducing the number of blocks (effectively soil type variations)
  • using monthly cow numbers rather than peak cow numbers
  • adding actual irrigation volume and timing of irrigation
  • increasing annual rainfall to the actual rainfall as occurred in 2008/09

paper1-5

Effects of Decision Rules on year to year variation:

The LUDF data set also enabled a comparison of the year to year variation in predicted N losses for a given set of decision rules. Most decision rules responded in a similar manner, typically showing an increase in losses year on year, however some decision rules resulted in a downward trend in predicted N losses year on year while others showed year to year variation but no trend.

Overall the variation in predicted N losses ranged from a minimum of 19 kgN/ha/yr to a maximum of 24kgN/ha/yr using farm specific data with default irrigation (a range of 26%) to 44 to 69 kgN/ha/yr (a range of 55%) for farm specific data with actual irrigation volume and timing. Measured losses (reported elsewhere) have shown similar ranges in year to year N-loss to water, reinforcing the need to consider long term average effects, not an individual years results (whether measured or modelled).

paper1-6

The predicted loss for LUDF in the years analysed in this exercise demonstrate average N loss to water could range from 19 to 74 kgN/ha/year. This shows the importance of the input decisions, particularly when the Overseer data will be used to compare with other farms or over multiple seasons. The range in predicted N losses between years for some decision rules is greater than the effect of some of the choices contributing to a set of decision rules.

Impact of farm specific data

Farm specific data without actual irrigation volume generally resulted in the lowest predicted N-losses to water, while adding actual irrigation volume and timing of irrigation resulted in higher predicted losses. The decision regarding actual irrigation volume versus the calculated deficit irrigation is therefore significant in estimating the likely scale of predicted N-losses. Neither option is necessarily ideal, one ignores actual data which, along with actual N-fertiliser applied, supplements used and milk production reflects the management decisions and outcome of a specific season; the other endeavours to compare actual irrigation against long term average rainfall. Ignoring actual irrigation volume may distort comparisons between farms if some farmers with high irrigation reliability irrigate more while others buy in more supplements or change other management practices (including drying off early).

Pasture quality and clover content are two important contributors to the lower N losses that occur within the farm specific (without irrigation volume) set of decision rules. The documented loss of clover in 2009/10, correlates with Overseer’s prediction of less clover fixed N in that latter years of comparison. The reduction in clover fixed N reduces the total available N and therefore surplus N.

High pasture quality across the year at LUDF is a function of irrigation, pasture species and pasture management, all combining to lift milk production. It is likely that Overseer estimates higher intakes would be required when using the default pasture parameters, compared to the actual pasture quality at LUDF. Lower intakes would result in lower N excretion and hence lower leaching when using the actual pasture quality data at LUDF.

Discussion:

Using Overseer to predict on-farm N losses from a given set of inputs, management strategies and production outcomes can contribute to a greater understanding of part of the environmental footprint for a specific farm. Consideration is required however of the intended use of the Overseer output which could range from compliance with supplier protocols or regional council requirements to internal understanding, benchmarking and / or seeking on-farm efficiencies.

In the absence of greater clarity (and agreement) on the use of farm specific input data, and the effect of this on the predicted N-losses, these potential alternative uses may require, or benefit from different decision rules regarding input data. In all cases the consistent use of a set of decision rules for a specific purpose, year to year will provide more meaningful data. This is of increased significance if the individuals creating Overseer files for a farm change over time.

Further there is the concept of long term average effects, and the decisions regarding the amalgamation of individual years inputs and production, modelled against long term average climatic data. Is a long term average of input data the effective means of ensuring relevancy and consistency against a long term climatic data set; and if so, is the reporting and averaging of individual years results more appropriate, or the averaging of multi-year input data to create an ongoing rolling average output?

Valid arguments can also be made to simplify as much of the data input as possible, relying largely on default values and perhaps seeking decision rules that either minimise the predicted N loss, or the year to year variation. Alternatively, using more farm specific data has the potential to increase the relevance of the output to an individual farm operation, and thus drive greater nutrient efficiency as the significance of changes in farm practice are reflected in Overseer outputs. It is also probable that future versions of Overseer will consider more detail and thus utilising the advanced input options (farm specific details) available in Overseer 6 may provide more relevant comparisons from current farm practice / predicted losses to future farm practice / predicted losses with future versions of Overseer.

Farmers, regulators, individuals generating Overseer reports and industry participants have the opportunity to influence the decision rules regarding the use of input choices within Overseer 6. Questions of the intended use of the output need to be considered along with the desire for a predicted N-loss to water that is relevant to the specific farm and consistent with other properties and targets when benchmarked.

Analysis of the input data from five consecutive and broadly similar years at LUDF shows it is difficult (and possibly inappropriate) to pinpoint an exact N-loss for this farm. A range of losses or average loss over time will more correctly describe the impact of LUDF’s management on the predicted loss of nitrogen from the root zone.

Appendix 1: Overview of Soil Details and Rainfall Records

paper1-7

The comparison using ‘fewer blocks’ ignored the Eyre soils (adding this area to the Templeton soils), added the Wakanui effluent block to the Templeton effluent block to result in a single effluent block only, and combined the remaining Wakanui and Temuka non-effluent blocks, choosing ‘Wakanui’ as it dominates this combined block (44 vs 32 ha).

Rainfall

paper1-8

Appendix 2: Nitrogen Application and Irrigation Volume

Nitrogen Applications – Total kg N applied per ha per month (Separated into Effluent and non-Effluent Areas in years where Nitrogen fertiliser was applied to effluent blocks)

paper1-9

Total Irrigation Volume (mm/ha) applied each month*

paper1-10

* Note: Irrigation volumes are ‘as recorded’ from water meters at LUDF. The water meters were upgraded prior to the 2012/ 13 season to accommodate the requirements of the National Regulations and Reporting Water Takes 2010.

Appendix 3: Clover content / Pasture Quality Parameters and Wintering off details

paper1-11

Notes:

  1. Clover levels interpreted into ‘medium’. ‘low’, or ‘very low’ based on description of clover in page 7, LUDF focus day notes, February 2010.
    1. 2008-9 season – Clover root weevil present in many pastures, not considered an economic problem
    2. 2009-10 season – Clover root weevil obvious in much greater numbers. Now considered to be making a negative impact on quantity and quality of pasture.
  2. Pasture utilisation left as default values
  3. Pasture samples are collected on a regular basis and analysed for quality. The sample is representative of the ‘as grazed’ pasture – ie harvested to the same height / pasture residual as the cows were expected to graze to.

paper1-12

Appendix 4 – Other assumptions in the Overseer Model for LUDF

2 ha Native Trees established in 07/08 and therefore farm area changes at the beginning of 2008/09 Season. 1 ha each from Templton and Wakanui taken for trees

Purchased silage storage conditions – excellent, type – baled silage, well stored
Silage utilisation – left as average (default)
Eco-n (DCD) – applied April and July – 25 day rotation in April, 70 in July, no N within 7 days. Three applications occurred in 2011/12, with the first in March, also on a 25 day rotation
Mature Cow weight – Default (439kg)
Breeding replacement rate – Default (23%)
Calving time not specified
Milk solids entered – not separated by milk fat and protein
Mg, Salt, Limeflour not added

Cultivated in last 5 years – not ticked
Dist from sea – 30km
Annual rainfall – 593
Rainfall variability – low (As per Overseer data file / map)
Temperature – 11
PET 685
Potential ET – moderate (As per Overseer data file / map)

Soil described by ‘type’
Average Soil test results used 10/11 soil test data used for all years
Default ASC used
Default TBK used

Effluent – Liquid, Spray from sump, <12mm, actively managed
(All Eff option – Spray from sump, Low application, actively managed).

Irrigation – Assumed Overseer calculates required volume in all months – Sept to April inclusive
Pasture Quality – Default
Clover content – Default

OAD Never (except specific cow number option below)

paper1-13

Appendix 5: Multi-year Summary of Nitrogen Inputs and Outputs:

paper1-14

Appendix 6: Predicted Farm N Loss to water – Individual Year Results

paper1-15

Extract: Assessing the impact of input choices within Overseer

Recognising Overseer is a long term equilibrium model, intended to predict long term annual average outputs, the Lincoln University Dairy Farm provides a multi-year dataset to compare the effect over time of varying input choices, within a consistent framework.


Login