Drone crop yield estimation uses drone imagery of a field or orchard to measure plant or tree counts, canopy size and crop vigour, then combines those measurements with sample harvest data from the ground to calculate the expected crop before harvest. Unlike an estimate by eye, it measures the whole area and states its uncertainty as a range.
Who needs a yield figure?
A yield estimate is not just a grower's curiosity. A food processor running contract farming wants to know when raw material will reach the plant and in what quantity. An exporter works out how much it can commit to a buyer before quoting. A cooperative sizes harvest crews, crates, cold storage and haulage around it. A large orchard business wants to walk into a sales negotiation with a figure it can stand behind.
Today that figure is often produced by a few experienced people walking the field and concluding "a bit less than last year". That works on a small plot. Across dozens of hectares, several varieties and trees of different ages, it relies on whatever can be seen from the track. When the estimate is wrong, either the factory sits idle or the crop waits in a queue for storage.
We explained how drones map crop health in our drone NDVI analysis guide. Yield estimation starts from the same imagery but goes one step further, from "is the crop healthy?" to "how many tonnes will we harvest?"
What a wrong estimate costs
In Türkiye, harvests can swing sharply from one year to the next, which makes a good estimate all the more valuable:
According to the Ministry of Agriculture and Forestry's Crop Production Statistics 2025 bulletin, total crop production in 2025 fell 11.6% on the previous year to 121 million tonnes, and output in the fruit, beverage and spice crops group fell 30.4%.
The same bulletin reports that apple production fell 48.3% and wheat production 13.7% in 2025. A buyer or grower who can see a drop of that size weeks before harvest can adjust contracts, storage capacity and price negotiations. One who finds out late either cannot deliver what it promised or pays for capacity it does not use.
Losses do not stop at the field gate. The FAO SDG Data Portal's Global Food Loss Indicator (12.3.1) estimates that 13.3% of food was lost globally in 2023 between harvest and retail, with the highest losses, 25.4%, in fruit and vegetables. Planning harvest, transport and storage capacity without knowing the real crop is among the factors that can make those losses worse.
How does drone crop yield estimation work?
There is no single "yield camera". An estimate combines different measurements depending on the crop:
- Counting: in orchards, tree counts; in vegetables and field crops, plant counts or gap rates, measured from high-resolution imagery with computer vision. Missing trees, dead saplings and empty rows also show up at this stage.
- Canopy and plant size: the area and height of each tree canopy are measured from a surface model generated by photogrammetry. Canopy volume often correlates with yield per tree. The same surface model is used on construction sites for drone stockpile volume measurement.
- Crop vigour: vegetation indices such as NDVI, calculated from a multispectral camera, show vigour and stress. A time series of flights across the season means far more than a single flight.
- Field sampling: fruit counts, weighing or trial harvests on selected sample trees or plots. The model is calibrated against this "ground truth".
- History: previous years' harvest records, plot-level yield history and weather data help the model separate the effect of the season.
The output is expected yield per plot and in total. In a properly run study it is reported as a range, not a single number.
Comparing the methods
| Method | Strength | Weakness | Best suited to |
|---|---|---|---|
| Field walk and estimate by eye | Quick, no cost | Poorly representative, depends on the person | Small, uniform plots |
| Sample harvest | Direct measurement | Rests on a small number of points | Calibration data |
| Satellite imagery | Wide coverage, regular revisits | Resolution often too coarse for single trees; cloud cover | Large arable fields |
| Drone imagery | Tree- and plant-level measurement; you choose when to fly | Flight and processing effort; limited area per day | Orchards, contract farming, high-value crops |
The most reliable result comes from a mixed method in which drone measurements are calibrated against sample harvests. A drone alone does not answer "how many tonnes?", and sampling alone does not represent the whole area. For season-long monitoring of large arable fields, satellite imagery for agriculture complements drone flights.
Drone crop yield estimation: a seven-step method
- Be clear about the purpose and the crop. Whose decision will the estimate feed: the harvest plan, a sales commitment or the factory intake plan? Which crop, which variety, how many plots? The target precision and delivery date follow from that.
- Map plot boundaries and attributes. Each plot's boundary, variety, planting year and irrigation status should sit on one map. This is built on the same logic as any geographic information system that manages spatial data. Keeping one plot code across every record is a data governance decision.
- Schedule flights around crop stages. Counting needs a flight before or after leaf-out; tracking development needs regular flights through the season. In Türkiye, drone flights are regulated by the Directorate General of Civil Aviation (SHGM); depending on the aircraft class and flight area, registration, pilot competency and a flight permit may be required.
- Choose the right sensor. A high-resolution RGB camera is often enough for tree counts and canopy measurement; crop vigour and stress need a multispectral camera.
- Design the field sampling. Sample trees or plots are chosen from the map to represent good, average and weak zones. How sampling is done affects the reliability of the model more than the imaging technology. Training a counting model needs labelled images, which we cover in data labelling for AI.
- Build the model and report a range. Counts multiplied by estimated yield per tree, or the relationship between a vegetation index and yield, are calibrated with the sample data. Results are presented per plot with an uncertainty range.
- Compare after harvest. Actual harvest is compared with the estimate plot by plot, and the reasons for any gap are recorded. Each season's data strengthens next year's model.
Combining imagery with sensor data
A drone image is a snapshot, while soil moisture, temperature and frost events change constantly. Combined with data from in-field sensors, the model can account for stress that happens after the imagery was captured. You can read about setting up that sensor layer in our precision agriculture and soil moisture sensors guide. Once the yield figure feeds the processor's demand and capacity plan, it can also be used alongside AI demand forecasting.
From estimate to harvest and haulage plan
A yield estimate creates value when it becomes a decision table. With expected volume and ripening order known per plot, harvest crew numbers, crate needs, cold store capacity and trucks can be planned weeks ahead. Route optimisation works from the same data to sequence and route vehicles from field to factory or store.
Carrying the plot ID onto the crate and batch record of the harvested crop brings another benefit: the product becomes traceable to the plot and date it came from. For food companies this ties directly into food traceability and digital HACCP.
Yield estimates are far more accurate when plot, input and harvest records are already digital. We explain how to set up that record keeping in our farm digital transformation guide. When designing the screen that tracks estimated against actual harvest by plot, our guide to choosing KPIs for a BI dashboard is a useful starting point.
How we do this at Digital Bridge
We build yield estimation as a data process that runs all the way to the moment of decision, not as a single flight and a map:
- Discovery and requirements analysis. We agree with you which decision the estimate serves, by what date and to what precision, and review the crop, varieties, plot structure and any harvest records you hold. Our approach for farm businesses is summarised on our agriculture and livestock solutions page.
- Flights and image processing. We plan flights around the crop calendar; if your own operator or a contracted drone firm does the flying, we share the required image overlap, flight altitude and camera settings in advance. We produce orthophotos, surface models and vegetation index maps through our drone data processing service. For tree and plant counting we develop computer vision object-detection models tailored to your crop.
- Pilot and calibration. In the first season we start with a limited number of plots and calibrate the model against sample harvest data. After harvest we report the estimate and the actual side by side, and base the decision to expand on that comparison.
- One map, one dashboard. We bring plot boundaries and results together in a geographic information system (GIS) layer and in-field sensors on our precision agriculture IoT platform. For management, plot and total yield appear on a single dashboard.
Flight reports, orthophoto outputs and field sampling sheets pile up season after season. SmartFiles, part of our own Smart360 product family, keeps PDF, Excel and image files in project areas with version history; every file uploaded under the same name becomes a new version, so last season's report is never overwritten.
Your next step
Start by listing what you already have: plot boundaries, variety and planting year, and plot-level harvest records for the last three years. The tidier those records, the stronger the first season's model. Send the list through our contact page and we will propose, in writing, a pilot scope that fits your crop and calendar. More data and analytics guides are on our Data & Analytics hub.