Categories
Blog

How models can improve your farming decisions

Not many careers carry the same challenges of complexity as farming. Politics requires substantial adaptability. Consumer marketing and the share market both have challenges of trying to forecast what is erratic and volatile. Manufacturing requires controls over process and product quality. Farming combines all of these, and requires complex, multi-factor decisions that deal with climate and weather, market demand and pricing, as well as the sometimes-hidden feedback loops of biological systems.

The Deloitte Red Meat Sector Strategy report estimated that there was potential to achieve $3.4bn of additional profit in the New Zealand red meat industry by 2025. The best performing farms are achieving excellent levels of profitability, but there is a substantial gap between the top 20% of farms (profit in excess of $500/ha) and what “average” farms earn ($169/ha): a gap that is primarily explainable by farm decision making.

The ANZ Red Meat Key Insights report showed that high performing red meat sector farms:

  • put an emphasis on good governance;
  • mitigate key risks;
  • use financial management tools;
  • invest to improve pastures;
  • focus on genetic selection;
  • use crops and supplements;
  • use feed budgeting and regularly weigh stock; and
  • operate flexible stock policies and regular stock rotations.

All of the above require farmers to make decisions with significantly more flexibility than “we’ve always done it this way”, faster than can be achieved through annual analyses, and with more precision than is possible on the back of an envelope.

At Rezare Systems, we believe that improving the flow of data to support decision making must be complemented by forward-looking predictive models. Mathematical models allow farmers and advisors to combine data from various sources (and sometimes at different scales) and to project that data forward (to see what might happen and how different options might work).

Models also allow farmers and advisors to visualise what cannot otherwise be seen. Want to know how much nitrate your farm is losing to the environment? Fully instrumenting your farm is expensive and infeasible, and the process of installing monitoring devices may well upset the results. Models may not be as precise, but allow you to forecast what is happening “under the hood”.

Inside the Models

There are different approaches to building models:

Mechanistic models attempt to use mathematics and first-principles logic to represent “what is happening” in a system (how animals or plants grow, or chemicals transform in the soil). They are often based on many years of discrete scientific experiments around the world, as researchers build an understanding of the working of these systems.

Although they are often simplified for practicality, mechanistic models often require many inputs. A strength of these models is that they may make useful predictions about circumstances that we have not yet seen.

Statistical or Regression models use the observed relationships between parameters we can observe. These models can be helpful where we don’t completely understand all the details of underlying systems, but where the relationships are statistically significant and useful. They are also useful where the absolute values of inputs can’t be precisely predicted because there is a degree of “randomness”. For this reason, these models are sometimes called stochastic models. A weakness of statistical models is that we don’t always know all the circumstances where the relationship may not hold true.

Models in practice

In reality, the software that we use for analysis and prediction at farm scale often combines many models – some mechanistic, some statistical. Models empower, but they don’t eliminate the value of competence and experience in decision making; and of course they require good quality input data to be effective.

Wise farmers and advisors spend time “calibrating” the tools for their circumstances so they have confidence that they know where it does and does not work. They use the model for forecasting and planning, but also put in place monitoring regimes so they see when actual differs from plan.

It was George Box who famously said “The most that can be expected from any model is that it can supply a useful approximation to reality: All models are wrong; some models are useful.”