One of the best things food safety research projects can give isn’t certainty. It’s narrower uncertainty. 

Think about our agricultural systems. Food safety in agriculture operates in an inherently variable, uncertain and dynamic environment. Weather changes. Water moves. Animals move. Microbial populations fluctuate. Agricultural activities change throughout the season. Conditions upstream and adjacent to a farm may change without the grower controlling, or even immediately knowing about them.  

Growers manage this complexity every day, adjusting production decisions to deliver safe, nutritious fresh produce. However, it’s also very unlikely that we can learn enough to control the complexity and dynamic nature of this system. 

Research alone will never provide certainty across every fresh produce environment.  

First, variability vs uncertainty. 

I often see variability and uncertainty used interchangeably, yet they are very different things.  

• Variability refers to the inherent diversity of data. For example, with population height there will be shorter and taller people. Variability cannot be reduced (yet it can be better characterized). 

• Uncertainty refers to the lack of data or understanding of the context around a decision. Uncertainty can be reduced with better data and information. 

Research increasingly points us toward the need for site-specific risk management systems that use local conditions, data and knowledge to continuously optimize food safety outcomes. 

Dynamic risk management is the real-world intersection of what we know from research about our fresh produce systems, and the design of functional tools to help manage inherent variability, risks and uncertainty. 

In these agricultural environments, we cannot expect to be able to eliminate inherent variability or expect to remove all uncertainty. What we can do in these systems is get much better at separating, characterizing and visualizing them. 

Imagine that we begin with a very large list labeled “We don’t know.” Research, surveillance and operational data should be designed to progressively make that list smaller and smaller. 

From that effort, data and research can help us understand when and under what environmental conditions a hazard is more likely to become a meaningful risk. More plainly, we can identify pathways that matter more during certain parts of the season because of rainfall, hydrology, temperature, animal activity or agricultural practices. 

Most of these factors may be unremarkable on their own. But when certain factors occur together, they can create conditions that meaningfully increase risk. That is an important distinction: the goal should not be to manage individual factors in isolation, but to recognize, monitor and manage the combinations and relationships that create meaningful risk pathways. 

With systems capable of monitoring these relationships, we gain something we didn’t have before: the opportunity to change what we do when, and only when, the risk changes. 

That is the value of on-going operational risk management. 

Instead of treating every acre, every day, every water source and every hazard equally, a risk-management system continuously asks: What is happening right now, and does it meaningfully change risk? 

When conditions change, monitoring can change. Sampling can increase or move. A water source can receive additional scrutiny or treatment, but only when it needs it. Harvest decisions can incorporate new information and be delayed or modified to reduce risk. And when conditions return toward baseline, resources can be redirected elsewhere.  

That is fundamentally different from a static checklist. It gives food safety systems the ability to adapt as conditions change.   

Checklists ask whether something was done. Risk-management systems ask whether what we are doing still makes sense given what we currently know. 

This also changes how we should think about data, the type we collect, and how we do it.  

Sometimes it may feel like everyone just keeps asking for more data and more information. Ultimately, that can often just feel like more work. But the goal is not simply to collect more of it. More samples, sensors and tests do not automatically create a safer food system. We need the right type of data, not any type.  

The right type of data helps characterize normal variation in our fresh produce environments and reduce uncertainty about where observed change may indicate a meaningful pathway(s) of risk. The systems we need to build must be capable of translating that information into decisions – if not, it’s just another exercise in data collection.  

Agriculture has already begun to embrace this concept in other areas. Precision agriculture uses increasingly sophisticated information about soil, weather, water, plant health and other variables to optimize decisions rather than treating an entire production system as uniform. The opportunity is that much of this same information may also help us understand food safety risks. We do not necessarily need to build an entirely separate system. We just need to expand how we use the information agricultural systems are already beginning to collect. 

Precision agriculture should not only help us optimize yield, water use and crop health. It can also help us understand when conditions are changing in ways that matter for food safety. 

We will probably never have perfect information about every hazard, pathway or field. We also will never have one research study that answers all our questions. But the good news is we do not need to wait for them either. 

The future of food safety isn’t knowing everything. It’s getting much better at knowing when, where and why risk changes—and knowing what to do when it does. 

Today, I want to write about baseline measures: what they are, why they are useful, and why I think we sometimes get them slightly wrong.

In 2022, Western Growers launched GreenLink®, a private data-sharing platform that helps the fresh produce industry learn from food safety data. Much of our work since then has come down to a few deceptively simple questions: What is normal? What is changing? What deserves a closer look? Answering those questions begins with defining a baseline. 

So, what is the baseline? 

When we study a population, operation, or process, one of the first things we want to understand is what is already happening. In produce safety, we may want to know how frequently positive findings occur, whether results change over time, and what factors may influence those changes. 

Together, this information helps us establish a baseline: a description of how a defined process has behaved under specific conditions and over a specific period. It gives us a reference point for comparing future results and determining whether something may have changed. 

A useful baseline considers more than an average. It also accounts for how many results normally vary, how they are distributed, whether there are trends or seasonal patterns, and how confident we can be based on the number and representativeness of the samples. 

This matters because two operations can have the same average result while behaving very differently. One may produce remarkably consistent results, while the other moves between very high and very low values. Their averages may be identical, but their underlying processes, and the questions we should ask about them, are not. You can read more about this in our article on understanding summary statistics. 

There is also an important limitation  

A baseline describes how a process has behaved, not how it should behave. If a process has consistently performed poorly, its baseline will reflect that history. The baseline is a reference point, not a measure of good orbad performance.  

So instead, I like to think that baselines are personal 

So, what do I mean by this? Well baseline are personal to each operation and process. It separates routine variation from meaningful change and helps us recognize when something is behaving differently, even before we understand why. 

Think about your usual commute. Perhaps it normally takes between 25 and 35 minutes. One morning, it takes 38 minutes. That is slightly longer than usual, but it could easily be explained by a slow traffic light. If the same commute suddenly takes 65 minutes, however, you immediately begin asking what changed. The longer commute does not tell you what happened. There may have been an accident, road construction, or bad weather. What it tells you is that the experience was different enough from normal to deserve a closer look. That is what a baseline is designed to do. And yes, in the long term with good information we would know what was the reason for the delay and learn from it.  

Now imagine that you are moving to a new home. Your previous commute may remain unchanged for whoever still drives that route, but it is no longer your baseline. You now have a different starting point, route, distance, and set of conditions. 

The same idea applies to food safety data. A baseline does not belong to a measurement alone. It belongs to a measurement collected from a particular process under conditions. If the operation, location, commodity, season, equipment, sampling method, or laboratory method changes, the baseline may also need to change. This is why I like to say that baselines are personal, my operation’s baseline may not be the same as my neighbor’s baseline.  

How do baselines help us recognize change? 

A meaningful change is not limited to one extreme result. It may appear as several consecutive results moving in the same direction, greater variability, more frequent positive findings, repeated findings in one location, or an unusual cluster associated with a season or process. An individual result may not look concerning by itself; the pattern may be what tells us that something has changed. 

Being outside the baseline is therefore a signal, not a diagnosis. It tells us that something may have changed, not why it has changed or that food is necessarily unsafe. The next step is to verify the data, review the surrounding conditions, and determine whether the result is isolated or part of a larger pattern. Here is a quick guide by NIST on control patter and understanding “Change”.  

Why baselines are needed in produce safety and why they are important 

Fresh produce is grown and handled in complex, changing environments, that depend in an infinite number of variables. Operations differ in their commodities, locations, seasons, water sources, equipment, weather conditions, and sampling programs. Even within one operation, conditions can vary significantly among fields, facilities, and seasons. 

Yet, we often interpret food safety data using one number or expectation. We ask whether the result is high or low (in case of prevalence, or water test results) without first asking: High or low compared with what? 

Baselines provide the context needed to answer that question. They can help us move away from a completely one-size-fits-all approach to interpreting data. Common food safety expectations may still apply across the industry, but the way we prioritize risks and implement preventive practices should reflect the conditions of an individual operation. 

Baselines can also help improve food safety metrics and standards; one good example is our food safety guidelines or best practices. Defining a baseline is extremely important to know how to set the metrics, if something may be too stringent, or if something provides an adequate level of protection.  Baseline measures can show us which defined or updated metrics have provided food safety value, where additional guidance is needed, and whether a metric continues to work as intended.  

This does not mean lowering expectations for operations with historically poor performance. It means using evidence to make requirements more focused, practical, and responsive to risk. 

They are equally important for predictive models. A model must understand what is typical before it can identify what is unusual. Strong, representative baselines allow models to consider factors such as location, commodity, season, weather, sampling stage, and analytical method. Without that context, a model may flag routine variation as a problem or overlook a change that is meaningful for a particular operation. 

Ultimately, baselines help us focus our attention on the changes that matter. Their purpose is not to assign blame or declare that something has gone wrong based on one unusual result. Their purpose is to help us understand how a process behaves, recognize meaningful changes early, ask better questions, and act before a developing issue becomes a larger food safety problem. 

That is the beauty of a baseline: it turns data into context, and context into better decisions. Let’s keep working towards defining our baselines.