When Yield Predictions Crack: The 5 Worst Forecasting Failures (Nut Industry)
...and how AgTech can can help avoid failures in the future.
Nut yields are notoriously difficult to predict, as yields are highly variable across regions, orchards, and cultivars. Many of the causes of this variability are still unknown. But what is well know, is that there are always risks to yield in any given season.
Risks that influence in-season yield include: Weather impacts, pest and disease impacts, and other factors largely out of our control (or those that come at us out of the blue) like theft, mismanagement, social unrest, and unforeseen events like epidemics and legislature changes.
Importantly, these in-season risks impact yield predictions and their accuracy.
This is especially true where historical yield data is used to provide a projected figure for the future.
While past data can shine a guiding light on the general norms for a certain nut type or cultivar according to age of the orchard, growing region, and used for predictions if “all things being equal”… it is not an accurate method of predicting the harvest size of the year to come.
Life is seldom an “all things being equal” game, if ever. There are simply too many influencing factors and nuances when it comes to farming.
PART 1:
When nut harvest forecasts went horribly wrong
There have been several instances where yield and harvest forecasting in the nut industry went horribly wrong, leading to significant financial and operational disruptions. Here are a five examples:
1) The California Almond Crop Forecast Overestimation, 2014
What Happened? The USDA’s National Agricultural Statistics Service (NASS) overestimated California’s almond crop yield in 2014, predicting a bumper harvest.
Impact: Growers and traders anticipated a surplus, which initially depressed prices. However, the actual yield was significantly lower than expected due to drought conditions and pollination issues. This led to market volatility, supply chain disruptions, and financial losses for those who had priced their contracts based on the faulty forecast.
What Went Wrong? The forecasting models failed to account for water shortages and underestimated the impact of extreme weather on pollination and nut set.
2) The Australian Macadamia Crop Underestimation, 2016-2017
What Happened? Early industry estimates suggested a poor macadamia yield due to unfavourable weather conditions.
Impact: Many processors and exporters planned for limited supply, affecting contracts and pricing strategies. However, the final yield was significantly higher than expected, causing market corrections and pricing issues.
What Went Wrong? Forecasting models were heavily reliant on historical weather data but did not accurately incorporate real-time data on tree recovery and late-season flowering.
3) The U.S. Pecan Harvest Overestimation, 2018-2019
What Happened? The pecan industry relied on optimistic forecasts suggesting a strong U.S. crop in 2018-2019.
Impact: Traders and exporters made financial commitments based on these predictions. However, hurricanes (such as Hurricane Michael in Georgia) devastated orchards, reducing actual yields. This led to price spikes, contract failures, and supply chain struggles.
What Went Wrong? Forecasting did not adequately account for extreme weather risks, and the reliance on early-season indicators proved unreliable.
4) The South African Macadamia Crop Forecast Overestimation, 2020
What Happened? Industry bodies and analysts predicted a record macadamia harvest, leading processors and exporters to make supply chain and pricing commitments accordingly.
Impact: The actual harvest fell short, mainly due to unexpected late-season droughts and increased pest pressure. This created financial stress for processors who had locked in contracts based on inaccurate yield forecasts.
What Went Wrong? The models used for yield prediction underestimated the impact of climate variability and pest outbreaks.
5) The Chilean Walnut Harvest Shortfall, 2021
What Happened? Chile, a major global walnut exporter, predicted a strong harvest in 2021.
Impact: Due to unexpected weather fluctuations (frost and rain during flowering), actual yields were significantly lower. Many export contracts had to be renegotiated, and some buyers faced shortages.
What Went Wrong? The forecasting model relied on early-season flowering indicators but failed to account for unexpected frost damage.
Key Takeaways from These Failures:
Over-reliance on Historical Data:
Many forecasts rely too much on past trends and fail to adapt to emerging risks.Climate Variability:
Extreme weather events, such as droughts, hurricanes, and frosts, can severely impact nut yields.Pollination Uncertainties:
For tree nuts like almonds and macadamias, successful pollination is crucial but often unpredictable.Pest and Disease Outbreaks:
Unexpected infestations can rapidly reduce yields but are not always included in predictive models, or it’s difficult to know the extent of the damage.Lack of Real-Time Data Integrations:
Many forecasting models fail to incorporate up-to-date satellite imagery or drone data, soil moisture data, and AI-driven analytics.
PART 2:
Counteractive Measures: Proactive Sampling
One way to counter all the uncertainty is to know what the crop status is in real time. Or as near to real-time as possible. A massive challenge in large scale orchard scenarios.
There are simply too many trees and too little time (and oftentimes manpower or budget) to go around to every single tree and record where the crop is at, and to collate the data.
And what’s more… to do it all over again after an extreme weather event, pest infestation, or another unforeseen and destructive force.
In-field technologies like Aerobotics’ TrueFruit app are a new way of measuring yield in near real-time for review by all stakeholders. The best part? Sampling is done either regularly, or strategically after any unforeseen orchard changes.
We can now get an updated and accurate view of the amount of crop left on the trees.
In the context of sampling for yield projection - there are two major AI-use instances:
Sample tree generation
Weighted extrapolations
Herein lies the power of pairing drone insights with a world-class mobile app...
Drone data outputs are used to AI-generate the most representative trees to visit for nut counting. The mobile app then guides the user to the exact sample tree in the orchard. Depending on the size of the orchard, there may be several trees to visit. This number can also be customised.
My favoured suggestion is at least one tree per two hectares - one tree per hectare is even better - but you will need to balance hectares covered against capacity. A careful mix of enough data to back the integrity of the results, and available resources for the data collection itself.
Once at the sample point, and depending on the size of the tree and the number of nuts on that tree, stripping the tree may be the best option.
This is a personal preference and some will rather favour speed over accuracy, and hurriedly count nuts by eye on branches, or guesstimate the totals. Of course, the more accurate you can be, the better. I suggest stripping the tree, placing nuts into crates, and counting out the nuts as you remove them. A clicker counter can help too. Anything to reduce any human error at this stage is great.
If you feel disheartened about taking a crop off the tree for this exercise, think to yourself, “Will an accurate forecast of nut counts across my operations be worth more than the loss of the crop on a few trees per orchard?” if the answer is, “Absolutely!” then strip the trees, gather the results, and let the second step in the process run its course. This step is of course the weighted extrapolations.
The tree-specific drone data like canopy size, stress, transpiration, etc are used together with an accurate tree count to provide an extrapolated figures of:
Total nuts per orchard, and
Average nuts per tree.
With a general idea (or trusty formula) that indicated what weight a certain number of nuts should be (for example 100 fresh nuts = X-kg’s) one can get a very good idea of tonnage per orchard or per hectare using the weighted extrapolation figures.
Additionally, and one of the most underrated outcomes of this exercise, is the opportunity to produces spacial information within the orchards. Where are trees bearing less, or more? Why? And then to compare nut count results against the drone data outputs for further analysis. In effect, taking action to address limiting factors and work on improving future yields and lowering orchard variance.
“Yield estimates remain a constant challenge. Green & Gold Macadamias that does our sales and marketing, contracts our crop based on estimates and we are all dependent on getting this as accurate as possible early in the season. There are a few growers that get their estimates accurate, but even they get it wrong sometimes. A reliable method based on science would be a game-changer in the macadamia industry.” - Barry Christie, Group Agricultural Technical Manager (Green Farms Nut Co.)
Conclusions:
Accurate yield forecasting in the nut industry has long been a challenge, with historical miscalculations leading to financial losses, supply chain disruptions, and misplaced market confidence.
Traditional models that rely heavily on historical data and early-season indicators often fail to account for unpredictable factors such as extreme weather, pest outbreaks, and shifting environmental conditions.
However, the integration of AI, remote sensing, and real-time data collection is reshaping how nut farmers approach forecasting.
In making use of AI-driven sampling, drone-based analysis, and real-time orchard monitoring, farmers can move beyond static predictions and adopt a dynamic, data-driven approach that continuously adapts to changing conditions.
Tools like Aerobotics’ TrueFruit app demonstrate the power of AI-assisted sampling and weighted extrapolations, ensuring that yield estimates remain as accurate as possible, even in the face of unforeseen challenges.
By embracing these technologies, the industry can reduce forecasting errors, improve market stability, and optimise resource management.
In an era where precision agriculture is becoming the standard, those who integrate AI and remote sensing into their forecasting strategies will gain a critical edge. Not just in yield prediction, but in overall farm profitability and sustainability.
The future of nut production is clear: technology and data-driven insights are no longer optional but essential for success.
And the tools are here to be used.
Contact me to find out how.
» Join me at the Amber Macs Expo 2025, where I will be presenting on AI and drone-backed yield forecasting techniques and strategies. Click the link above to see the program and where you can grab your tickets.
Read more:
