Predicting almond yields before harvest is crucial to a grower’s ability to manage resources effectively. By combining practices such as field surveys with technologies like satellite imagery and machine learning, growers and researchers are pushing the boundaries of what’s possible in yield prediction. While the process can be a significant challenge and far from an exact science, recent advancements by independent consulting firms and researchers have made it more accurate and effective.

Researchers at UC Davis have developed machine learning models that combine historical yield data, weather patterns and remote sensing imagery to improve almond yield forecasts. These models account for factors such as orchard age, canopy size and climate, providing growers with precise tools to optimize management and productivity. 

Yet, as Almond Board of California Associate Director of Agricultural Research Sebastian Saa pointed out at The Almond Conference (TAC) in December, you can’t improve what you don’t measure.

“I think one of the key variables that we usually have a hard time measuring at a high resolution and understanding the variability that may happen, or plots early in the season, is yield,” Saa said during a session he moderated on early season yield prediction.

Researchers are leveraging agronomic and technical data to enhance early yield prediction, incorporating factors like field conditions, historical trends and advanced modeling techniques. These tools provide experts with valuable insights that inform both on-farm management decisions and broader marketing and inventory strategies. 

Both preharvest yield prediction and yield quantification at harvest are being studied through a comprehensive, multi-year project. Now in its third year, the initiative, led by UC Davis Professor of Plant Science Patrick Brown, is making substantial strides in refining predictive models and measurement techniques, paving the way for more precise, data-driven orchard management. 

“I think we have made significant progress in this area,” Saa said.

Challenges in Current Yield Estimation Methods
Traditional yield estimation methods face limitations due to variability within orchards, regional differences and unpredictable factors like weather. Human bias and the gap between early observations and harvest outcomes can also further complicate accurate yield forecasting. 

Speaking on the same panel at TAC, JJ Magdaleno, partner at Terra Nova Trading Inc., which has published annual crop estimates for the past 27 years, said predicting the future is no easy task. 

“It’s a humorous comment we make before we go out and do our estimate that I think all four of us on the panel get to look at and say, ‘Really, what we’re doing is trying to predict the future, which more or less is somewhat impossible,’” he said. 

Because of the unpredictable nature of the growing season, Magdaleno said, a key challenge in yield prediction is the significant differences between what’s observed in April and the actual harvest results in August. 

While Terra Nova has earned a reputation for accuracy with what he called a well-intentioned estimate, Magdaleno acknowledged they can’t please everyone. 

“I always joke with my mom, no matter how accurate we can be, whenever we do give the estimate, 50% of the people that we talk to will end up being upset at us,” he said. “Either it’s going to be too low or too high, but inevitably, somebody’s going to be upset, which kind of makes it the Sisyphean test that we have.” 

The report serves as a comprehensive yet concise snapshot of the season’s conditions across the growing region. Typically spanning five pages, Magdaleno explained that it provides an overview of key factors influencing the crop, including water availability, orchard removals and varietal shifts. While the report includes an estimated yield figure, its primary focus is on observations made over a nine-day assessment period, offering insights into emerging trends and potential challenges.

“Basically, it’s an overview of what we have seen in the nine days that we’re doing it,” he said. “Obviously, it comes with an estimate, but our intention is to just give a brief overview of what we’re seeing.” 

The team at Terra Nova examines more than 500 orchards during their surveys, gathering over 1,000 data points to obtain accurate industry trends. Their latest survey highlighted the challenges of predicting yield due to unpredictable weather and natural crop drops between April observations and the August harvest. While localized weather events like hailstorms have minimal statewide impact, 2024’s extreme heat wave may have reduced the crop by 3% to 5%, Magdaleno said. Additionally, human bias, whether from growers’ optimism or market sentiment, can distort expectations, making objective assessments critical. 

“I think people’s views of seeing the market actually have the effects on their actual views of the crop size themselves,” he said. 

Technological Innovations and Research
Brown is collaborating with four other UC Davis researchers on a project to refine forecasting models using remote sensing and machine learning. Their research enables high-resolution monitoring at both orchard and tree levels, improving accuracy and scalability. 

It’s not just market planning and logistics that are important in early yield prediction though, orchard optimization is another crucial aspect, Brown explained. 

“How will you choose to manage your orchard if you have a good vision of what the yield of that orchard will be in the coming year?” he said, noting that understanding variability in yield provides valuable insights into why trees produce as they do.

Their research is refining almond yield predictions by analyzing variability at multiple scales, from countywide trends to individual tree performance. A key finding is that significant variation exists even within single orchards, with yields ranging from as low as 1,000 pounds per acre to as high as 3,000 pounds. 

“And I know most growers will say, ‘I would never manage at a single yield, a single tree,’” Brown said. “But it provides us a lot of information that can help us understand what is causing this variability.” 

This data can also be used to inform broader estimates at the block, farm and county levels, helping growers make more informed management decisions, he added. 

The other part of the project, Brown explained, examines the factors driving yield variability within orchards, between orchards and across counties. While some influences like age and canopy volume are apparent, others, such as long-term spring temperatures, the previous year’s summer conditions, past yields and March precipitation, may also play a significant role. 

“Knowledge of those variables will help inform the other surveys that JJ and the others are doing to attempt to explain when their estimations are not accurate enough,” Brown said.

Beyond prediction, the research highlights the importance of precise resource management, particularly in fertilization. Because high-yielding areas require significantly more nitrogen than lower-yielding zones, mismanagement can lead to unnecessary nutrient loss and inefficiencies. 

Future Directions for Yield Prediction
As the industry moves toward more comprehensive and accessible tools that incorporate historical data, bloom dynamics and new imaging technologies, continued collaboration between researchers and growers will be critical for scaling these innovations and driving long-term sustainability. 

Brian Ezell, senior director of operations for Setton Farms and former vice president of the Almond Division at Wonderful Pistachios & Almonds, believes improving crop estimates will help stabilize the almond market by reducing speculation and bias from growers, sellers and buyers. 

“There is a ton of bias, not just at the grower level, but at the seller level and at the buyer level,” he said during the predicting early yield session at TAC. “Our hope was if we could put out a good product, it’s fairly accurate and gives more confidence around the other estimates as well.” 

Growers can also play a larger role, Ezell noted, highlighting the importance of participating in industry surveys to improve crop estimates and planning. He urged growers to complete acreage surveys from the National Agricultural Statistics Service (NASS), noting while NASS publishes an acreage report annually, limited participation impacts its accuracy. Accurate acreage data, he added, allows industry leaders to refine forecasts, set marketing strategies and allocate resources more effectively. 

“If you guys want better estimates, when you get that little postcard or email to fill out, please fill that out,” Ezell said. 

Looking ahead, he hopes that refining prediction models with more accurate and comprehensive data, such as tree-level yield assessments and regional variability, will enhance growers’ ability to plan for harvests, adjust input use and navigate market conditions with greater confidence.