Artificial Intelligence and Turf Management

OK, this isn’t something you normally see in the same sentence but I’m having a brief thinking session on the potential for using artificial intelligence – AI – (well, this depends upon its definition) to help aid manage turfgrass surfaces more effectively. Essentially this is developing the turf management model concept from earlier postings in March and April 2017.

Artificial Intelligence?

Artificial Intelligence?

There are many variables within turf management and many of these are poorly understood, especially regards how one variable might interact with one or more other variables. However, this shouldn’t stop us looking into how we can attempt to solve problems and issues, or at least make ourselves better informed into likely outcomes.

Currently the main approach to turf management is more along the formal Boolean logic lines, such as : “If a certain quantity of nitrogen is applied to a turf surface then this will typically increase sward density, encourage grass growth and increase mowing requirements, amongst other things.” This is quite a well informed approach to take and produces some excellent surfaces, at least on the face of it. However, how are the results framed within a wider, sustainable, turf management system? This is rarely explained.

Examples of this logic 

In computing terms we would say IF the condition is met (in this example, the application of nitrogen fertiliser) THEN the outcome (or resulting action) is as stated (in this example, increased growth etc.).

This approach can be related to all turf maintenance activities, for example:

  • IF a light irrigation is applied to an established sward THEN this will encourage shallow rooting.
  • IF the grass is cut at 25mm height THEN the sward will be uniform height of 25mm straight after cutting.
  • IF a cricket square is rolled THEN this will consolidate (compact) the soil profile.
  • IF a surface is slit-tined using 200mm deep slit tines THEN the rootzone will be aerated to a depth of 200mm, etc.

In general terms the above statements are relatively accurate and are typically representative of many turf management practices where a manager will correctly identify what is needed and the necessary action taken. However, you will have noticed that there are a number of issues with the statements. In particular, the statements haven’t considered any variables which might impact on the outcome.

What we also need to consider is ‘conditional probability’ and how this might manifest itself within an AI system.

Height of cut example

If, for example, a cylinder mower is set to cut at 25mm then the assumption is that this is what the finished cut will be. But we know that this is not necessarily the case.

For example, some variables we might consider and could apply a probability factor to in influencing the action (i.e. the THEN) aspect are:

  • What impact will a moist surface have on the mower sinking into the surface?
  • What impact will thatch, and the depth of thatch have on the actual cut height?
  • How accurate is the equipment used for setting the height of cut?
  • How competent is the person setting the height of cut?
  • Will the bench setting be correctly adjusted to produce a desired effective cutting height?

IF the grass is cut at 25mm height (we will call this event A) THEN the sward will be uniform height of 25mm straight after cutting WHERE soil moisture content is <20% AND Thatch depth is <12mm (we will call these a composite condition B for this example).

In conditional probability we would apply a figure for the probability of this occurring and this might therefore be represented as follows: P(A/B) = 80%, so we aren’t certain that the height would be 25mm but there is a strong likelihood it would be.

We could also enhance this further by being more specific by adding an OR into the statement such as: OR WHERE soil moisture <15% AND thatch depth < 18mm.

We could refine this further by also adding in variables of ‘competence’ and ‘equipment accuracy’ and this might increase the probability of achieving event A (of a correct 25mm height of cut) to let us say 95%.

We would utilise sensor technology to provide environmental data which is fed into the AI system, as well as other data inputs, to provide a range of data that the AI system would be analysing to determine various outcome probabilities for the many interconnected threads of managing a turfgrass surface.

The aim being to identify – within different time scales and time periods – when variables align to increase the probability of a successful achievement (however this might be measured) of the desired event.

The AI system would be using real-time data, with positive and negative feedback mechanisms, in providing more accurate and relevant information to help a turf manager make even more effective decisions in managing a turfgrass – natural or synthetic – surface.

It doesn’t matter about the various figures given above, it’s just the concept of gathering a wide range of data and information on which to make much better informed decisions and more effective use of often limited resources.

Another good example for the use of AI would be in more effective and responsible use of pesticides. If environmental conditions are such that a disease attack is highly likely does this mean a fungicide application is actually necessary? Typically this would be the case, however, an AI system might identify that the disease attack, whilst being highly likely, will most likely cause very limited damage to a turfgrass surface due to a range of factors it is able to take into account and compute. For example, most of the grasses present in the sward are actually relatively resistant to the disease; the maintenance regime has kept stress impact on the grass plants to a minimum during this period, no inappropriate application of Nitrogen had recently been given, the rootzone is relatively free draining etc.

These are all issues a good grounds manager will be thinking about anyway, but an AI system would be able to provide an increased level of confidence in helping a turf manager in making a more effective decision – this may actually be to not spray (saving otherwise considerable financial costs) and accept a minor blemish from a disease attack, which will actually build up resistance within the sward and reduce the need for fungicide applications in the future, subject to good turf management practices being carried out.

Anyway, lots to think about, with plenty of scope for further development.

Chris Gray, 10th January 2018