Crop Monitoring With Drones and Satellite Imagery (NDVI)
Advanced4 minInnovialabs editorial team
In short
NDVI compares how plants reflect red and near-infrared light to show where crops are weaker; satellites give regular coverage, drones give detail, and the result becomes prescription maps for machinery.
What NDVI is
NDVI (Normalized Difference Vegetation Index) shows how active vegetation is. Healthy leaves strongly absorb red light for photosynthesis and strongly reflect near-infrared (NIR). A stressed plant reflects less NIR.
The formula is simple:
ndvi = (nir - red) / (nir + red)
Values range from -1 to 1. Bare soil and water give low values, dense healthy crops give high ones. The absolute number alone says little: what matters is the variation within a field and the trend over time.
Other indices
NDVI is neither the only index nor always the best:
| Index | Uses | When it helps |
|---|---|---|
| NDVI | Red and NIR | General overview, early and mid stages |
| NDRE | Red edge and NIR | Dense canopies where NDVI saturates, nitrogen status |
| SAVI / EVI | Soil and atmosphere corrections | Sparse crops, lots of exposed soil |
| NDWI / NDMI | NIR and shortwave IR | Plant water status |
At later stages, when leaves fully cover the soil, NDVI barely rises any more. Differences then show up better in NDRE.
Satellite or drone
| Parameter | Satellite | Drone |
|---|---|---|
| Resolution | Meters per pixel | Centimeters per pixel |
| Coverage | All fields at once | Limited by battery and flight time |
| Frequency | Every few days, if there are no clouds | On demand, weather permitting |
| Clouds | The main problem | Flies below the clouds |
| Costs | Free open data exists | Aircraft, multispectral camera, pilot, processing |
| Best for | Routine monitoring, finding anomalies | Detailed checks, plant counts, spot problems |
In practice they are combined: the satellite shows every few days where something is wrong, and the drone flies to that specific area for detail. Open satellite data, such as the Sentinel-2 mission of the Copernicus programme, is enough to start without paying for imagery.
Detecting stress and disease
An index shows a deviation, not its cause. A low-NDVI patch can mean:
- water shortage or waterlogging;
- nitrogen or other nutrient deficiency;
- disease and pests;
- seeding gaps, soil compaction, hail damage.
A working process:
- Compare the area with the rest of the field and with earlier images — the trend matters.
- Discard images with clouds and shadows: a cloud mask is a must.
- Overlay other layers: soil map, topography, operations history, weather station data.
- Send the agronomist to scout the anomaly using its coordinates.
- Record the cause in your system so rules become more accurate over time.
Machine learning models can help classify problems, but they need labeled examples confirmed in the field.
From image to prescription map
A prescription map turns an index into application rates for machinery:
- Choose an image at a suitable crop growth stage.
- Split the field into zones by index values, usually a few classes from weak to strong areas.
- Assign a rate to each zone. The logic depends on the goal: nitrogen is sometimes added to weak areas and sometimes to high-potential ones. This is an agronomic decision, not an automatic one.
- Smooth out small patches: machinery cannot change the rate every meter.
- Export the map in a format the machine's controller reads (often shapefile or ISOXML).
Feeding results into farm software
Monitoring delivers the most when it is connected to your records:
- images and indices are tied to field boundaries from the farm management software;
- a detected anomaly creates a scouting task for the agronomist;
- the treatment that follows is logged as an operation with costs;
- at season end, indices are compared with the yield map.
Automation usually relies on satellite platform APIs: the system fetches new images on a schedule, calculates indices per field and sends an alert when a value drops sharply.
Common mistakes
- Comparing absolute NDVI values across different crops and dates.
- Ignoring clouds, haze and sun angle.
- Building a prescription map from a single image without field checks.
- Storing imagery separately from the operations history.