Limitations of modelling

• Scale

Scale is an important concept in modelling, and the term ‘scale’ itself can be easily misinterpreted because it can refer to both large and small areas of land, for example, a large scale development refers to a large land area, whereas, a large scale map refers to a small land area (Addiscott, 2003). Also, government organisations and policymakers require information from models at national or regional scales, whereas many models are developed at a plot or paddock scale. Up-scaling of models reduces the accuracy of inputs and increases heterogeneity and down-scaling requires increased accuracy

(Addiscott, 2003). Validation and error propagation are also potential issues arising from up or down scaling models (Addiscott, 2003).

• Decoherence

Decoherence is also associated with scale and is a term used to explain the loss of indeterminacy that occurs as smaller systems are amassed to make up larger ones. An analogy for this is the nature of physics at large and small scales. Historically, physics was based on large scale observations (e.g. Newton’s Law), however more recent quantum physics is associated with very small scales and is largely indeterminate

(Addiscott, 2003). Relating this to agriculture, processes occurring at a large paddock or catchment scale should be more determinate than the same process at a small soil particle

scale, e.g. considerable N2O emissions from soil occur from randomly located microsites, where conditions are favourable for denitrification (Parkin, 1987). The exact location and the processes occurring in these sites are unpredictable, and very difficult to model at the plot or field scale. However, the establishment of predictive models for denitrification have been easier at the larger catchment scale (Groffman and Tiedje, 1989; Corre et al., 1996) due to the fact the soil moisture (which has a significant effect on denitrification) can be estimated based on topographical features in the landscape. This information, combined with other factors, such as soil and climate, can then be used to create denitrification models (Addiscott, 2003).

• Error

As described earlier, models cannot conceptually be 100% accurate. In a modelling context, error refers to the disparity between the modelled representation of a system, and our scientific understanding of the reality of the system (Heuvelink, 1998). There are three primary sources of model error:

(a) Input error: Model parameters such as soil properties and weather and/or climatic data always contain a degree of error. Some of these may be “human error” or mistakes, and although it is important to minimise this sort of error, little can be done to avoid it. What is of greater concern is statistical error which arises from either natural variation, or error introduced from measurements or estimation (Addiscott, 2003);

(b) Model error: A fault in the model itself can arise from “concept error”, i.e. an error in understanding, or deliberate simplification by the modeller of the system being modelled. There is no diagnostic test for this kind of error, however they may be exposed by

sensitivity analysis and review critique, and should certainly be uncovered if the model is validated against observed data (Addiscott, 2003). Corrective action against model error is to simply change the model. As noted earlier, changes in scale may also require

changes to a model (Addiscott, 2003). Another possibility is “error in translation”, where error occurs during the process of converting the concept or theory into a set of

mathematical equations and computer code. Translation errors are usually revealed during model verification, and remediation will depend on the error (Addiscott, 2003); (c) Output error: This can be a result of input error, model error or both. The relation between input and output error (in terms of variation) is the essence of error propagation.

The majority of variables measured for use as parameters in models have a certain amount of error, as do parameters that are inferred rather than measured (Addiscott, 2003). If a given model is non-linear, the error in the input contributes to the value of the mean of the output, and can significantly increase the output error. Error in multiple parameters or equations can also cancel each other out.

• Units and conversion factors

The choice of units can often cause problems and confusion in modelling. The mixture of units for mass (e.g. milligrams and kilograms), or area (e.g. square kilometres and

hectares) within one model increases the potential for error and confusion. It is therefore important to have a consistent set of units throughout a model. If unit conversions are deemed necessary, these should be undertaken outside of the model, leaving it

independent of conversion factors. Many modellers use the universally accepted

International System of Units (SI) (Royal Society, 1975) where the basic units for mass, length and time are kilograms, metres and seconds respectively.

2.8 Summary

Increases in the industrial scale production of N fertilisers for agriculture has allowed for increases in the intensification and profitability of this sector. However, the loss of reactive N from agricultural systems to the wider environment, namely via leaching and gaseous emissions, contributes to some large present day environmental problems, as well as representing a waste and loss of potential productivity.

In pastoral systems, urine patches are considered the primary source of N loss, however, this review identified there is little quantification of the area actually affected by a urine patch (the ‘effective area’). This is important in determining how urinary N is partitioned between plant uptake and losses, and has important implications for the ability of

predictive models and nutrient budgeting software to accurately estimate N losses. In addition to this, the application of fertiliser over grazed pasture is an added

contributing factor to N losses from urine patches; however, very little is known about the fate and dynamics of N loss when the two are applied concurrently. Also, while some research has quantified the fate of urinary N, in the presence of fertiliser, little attention has been paid to quantifying the fate of the fertiliser component, rather, it has been assumed to simply have an additive effect on N loss. Much research and development

into precision fertiliser technology has occurred (in particular urine/dung patch avoidance technology) without any quantification of its environmental implications for N loss reduction, making this an important research gap.

Finally, process-based models to aid in decision support and predictive estimates of N loss from pastoral systems are increasingly useful tools for farmers, councils and consultants; however, continued evaluation and validation of modelled outputs with experimental data can be an essential element of quality control. The employment of models to aid in policy making and farm management decisions will likely continue to increase with increasing reliance. While the aim of many of these decisions/policies is to improve environmental health, they also have the potential to have widespread social and economic impacts on individuals, communities and the pastoral agricultural sector in NZ as a whole. Therefore, it is important that models replicating pastoral systems are evaluated and validated against relevant experimental data where possible. The review of literature has identified some key knowledge gaps. These include:

• A lack of understanding of the fate of fertiliser N when it is deposited on a urine patch;

• Quantification of the ‘effective area’ of a urine patch; and

• An ongoing requirement for the accurate prediction of the fate of N in pastoral systems using models, to (a) increase confidence in their use as decision support mechanisms, and (b) add value to and/or extrapolate on experimental data to answer scientific questions and increase understanding of the whole system behaviour.

Chapter 3

Urine patch and fertiliser N interaction: effects of