The role of predictive modelling in agriculture
Predicting the future is a skill many farmers wish they could have. After all, if one could predict the maize price six months from now or how much rain will fall within the next six weeks, one might find that farming would be infinitely easier. Of course, this is not possible to such a precise extent, but there is an obvious and urgent need for predictive models that can be used to increase our scientific understanding of variation in the natural environment.
According to Van Rijmenam (2013) the three areas in agriculture that will be affected the most by the opportunities of predictive modelling and the use of big data are:
- improved efficiency and reduced costs of all machines operated while farming;
- improved productivity and efficiency of crops and animals; and
- to mitigate weather conditions and to optimise pricing for products.
Prediction is defined as organised thinking about what is possible, whereas predictive modelling is a form of business intelligence that focuses on combining existing information to predict future outcomes or trends. It uses dynamic models and algorithms that integrate data from different sources with basic knowledge on the physical, chemical and physiological processes that underlie crop growth and agricultural production in order to predict certain outcomes such as crop yield (Figure 1).
Figure 1: Graphic representation of the predictive modelling process
There are three general approaches to predictive modelling:
- The traditional approach
Modelling begins with the definition of a theory or model. Model building involves fitting models to data using traditional methods, such as linear regression and logistic regression, to estimate parameters for linear predictors.
- Data-adaptive approach
This approach only searches for useful predictors within a given dataset. It gives little thought to theories or hypotheses prior to running the analysis and is sometimes called statistical learning or data mining.
- Model-dependent research
Modelling begins with the calculation of an algorithm, which it uses to generate data, predictions, or recommendations. This approach integrates all aspects of crop growth through simulations and mathematical programming and is constantly improved by comparing generated data with real data.
Omnia’s Agronomy Support division is currently busy with a project to develop a predictive model that will provide farmers with real-time soil water content information and real-time yield potential for certain crops, under different environmental and rainfall scenarios. The objective is to assist agronomists and farmers in their decision-making process in order to optimise yield and to reduce risk.
This model is based on the model-dependent research approach and simulates soil water content (SWC) and yield potential by making use of soil physical and soil chemical data, real-time climate data, as well as crop parameters at cultivar level, as model inputs. An example of the simulated SWC during the growing season of maize, expressed as plant available water (PAW) and the associated grain yield potential (Ypot), is illustrated in Figure 2.
Figure 2: Simulated soil water content
To understand soil conditions and to improve the SWC outputs even further, on site and real-time soil sensors equipped with GPRS (general packet radio service) are implemented at strategic locations. At the end of each growing season the simulated yield potential for different cultivars and crops is compared against actual yield data generated by properly calibrated yield monitors for model improvement and calibration. Figure 3 illustrates that a simulation accuracy of 75% was obtained when compared to the actual maize yield measured during the course of this research project.
Figure 3: Relationship between simulated and actual maize yield
An ever-expanding sea of data surrounds us and data gathered from people and sensors is constantly transforming our world. The Strategic Agricultural Services department within Omnia Fertilizer is hard at work building predictive solutions and when combined with clever analytical techniques, this will create the opportunity for farmers and agronomists to take action in real time. Timely and accurate crop yield forecasts are essential for managing the risk associated with crop production, as well as for other activities such as marketing, storage, and transportation.
Van Rijmenam, M., 2013. From Machines to Crops to Animals: Big Data Turns Traditional Farming Upside Down. www.smartdatacollective.com.
By Dr Louis Ehlers: Manager - Agricultural Services