Current Early Season Survey Methods
Terra Nova Trading has been providing an early-season crop estimate for the almond industry for 27 years. The idea transpired over three decades ago because there was a clear void of information between bloom time to USDA’s subjective estimate in May.
While Terra Nova Trading’s estimates are based on predictions, a lot goes into creating the seven-page report, said Jerry [JJ] Magdaleno, partner at Terra Nova Trading.
In a matter of days, over 500 orchards are assessed throughout the valley, and the estimators are walking anywhere from eight to ten miles each day. Even though there are only a few months in between the early estimate to harvest, changes like the first and second drop, and weather conditions can drastically impact Terra Nova’s numbers.
Human bias, another inevitable factor, can impact the estimate as well. “No matter how much you try and take away your own belief before you go out [to estimate], you have some kind of bias in your mind,” Magdaleno explained.
Even the market can dictate the accessor’s views on the crop size. “For instance, on a very bullish market, people’s views of the crop get smaller; on a bearish market, people’s views on the crop actually get larger,” Magdaleno said.
Brian Ezzell of Setton Farms also volunteers on a team of 16-20 participants who carry out early season estimates. To mitigate bias and develop the most accurate analysis, their group is equipped with over 24 years of historical data which includes acreage and yield by county. The accessors can compare peak areas, regions and varieties.
One of the partners involved in this process even counts the number of nuts on over 100 trees so they can determine a baseline. It’s especially useful because nut size and canopy can be deceiving, making some orchards look more productive than they actually are.
“By using this chart and knowing the footprint of the orchard, the spacing and the nut size, we can determine what the yield is and how many pounds per acre,” he explained.
Instead of comparing averages, the team also looks at standard deviations. It helps the industry develop budgets for the next year, create a sales strategy and helps processors prepare supplies and equipment especially if large yields are predicted.
Exploring the Drivers of Variability
Although early season estimates are subject to change, for the past several years, they have provided the almond industry with a glimpse of what could be coming its way.
“There’s the marketing, the planning, the logistics and orchard optimization,” said Patrick Brown, professor of plant sciences at UC Davis. “[It helps growers decide] 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.”
That said, figuring out what drives variation from year to year, county to county, and even orchard to orchard, is a topic ABC has been actively researching.
Despite how most growers manage their operation, Brown said measuring variability on a per tree basis is actually most beneficial. The reasoning is because that data provides information that can “help us understand what is causing variability, and it can also be used to feed information into our whole block, farm, or county level estimations,” he said.
To gather those insights, ABC funded a two-part project to assess yield variability by county based on the knowledge of weather, crop canopy, last year’s yield and several other variables. The second half of the project examined within-field variability to explain what is causing variation among orchards and county-wide. “When you put those two pieces together, they enrich each other and give you a better and stronger algorithm,” Brown stated.
The project used a yield monitor system installed on a TOL harvester to gather single-tree resolution yield data at field harvest speeds. The findings showed factors influencing large-scale variation include age, canopy volume, long-term spring temperature, previous year summer temperature and March precipitation.
On the other hand, a few factors that influence small-scale variability are things such as trunk circumference and growth, canopy volume, fruit set, nutrition, soil type, carbohydrate utilization by winter twig samples, and historical damage. “Knowledge of those variables will help inform the other surveys [that JJ and the others are doing] to attempt to explain when their estimations aren’t accurate enough,” Brown said.
Forecasting with Remote Sensing
While human estimations paired with historical data and research are a good starting point, the Almond Board saw an opportunity to utilize new artificial intelligence technology to provide more yield details beyond what meets the eye.
“The question is, how can we predict the almond yield at each individual block level in a more quantitative, objective and also cost-effective way?” said Yufang Jin, professor of remote sensing and ecosystem change at UC Davis.
Jin and her team at UC Davis worked with ABC to answer this question. Their initial hypothesis predicted that for any orchard within a given year, the production is likely governed by the history of the orchard growth, most likely to be determined by either long-term climate or the soil properties and regulated by short-term weather conditions, like March precipitation or hot summers.
The team took a deeper dive into this hypothesis using remote sensing technology and machine learning models. They learned that its monitoring capabilities can look at what’s happening at the individual tree and orchard level.
This technology can also assess things that we might not be able to see with the human eye, like color and the tree’s stress conditions. Most importantly, these sensors are constantly collecting imagery, which allows for repeated observation in a more efficient way.
Jin explained that with ongoing research, the goal is to take advantage of all this imagery and “integrate them with all the climate data, weather information, and orchard level characteristics such as cultivar and age to predict yield early in the season.”
Based on their first round of modeling work for individual blocks, the team identified the key predictors to be age; cultivar composition; weather and climate; fractional canopy cover and vegetation indices.
Their framework at the individual tree-level is still in the works, though Jin is confident that with machine-learning capabilities, historical data, and funding to analyze more orchards, more robust models will be developed to forecast yields.
Harnessing The Power of Data
Whether it be through early-season estimates from industry professionals, variability insights gathered from in-field monitors, or remote sensing imagery, there is great opportunity to equip growers with information to optimize their orchards. The use of technology will continue to make this process easier, and the Almond Board is proactive in making sure there is well-funded research and data to support this.
