Olive grove alternate-bearing cycle detection from multi-year NDVI series
Olive groves oscillate between heavy-fruit and light-fruit years, and that rhythm shows up in multi-year NDVI time series. Sentinel-2 at 10 m resolves individual grove blocks; the Landsat archive reaches back to 1984 for multi-cycle validation.
Sensors
- Sentinel-2 MSI: 10 m resolution in red and NIR bands (B04, B08) used for NDVI; 5-day revisit at mid-latitudes with two satellites. Dense enough time series to resolve olive phenological stages within a single season, and fine enough spatially to isolate parcels as small as 0.5 ha in fragmented Mediterranean landscapes.
- Landsat 4–9 archive: 30 m resolution, continuous record from 1984. Sufficient to capture multi-decadal NDVI oscillations across grove blocks larger than roughly 1–2 ha. The archive depth is the main asset here: it allows verification of three to four complete alternate-bearing cycles before Sentinel-2 existed.
- MODIS MOD13Q1: 250 m NDVI composite at 16-day intervals. Too coarse to resolve individual parcels in fragmented groves, but useful for regional-scale cycle synchrony analysis across entire olive-producing provinces, where mixed-pixel averaging still captures the landscape-level signal.
- Planet SuperDove: 3 m resolution, daily revisit where tasked. Resolves individual tree canopies rather than grove blocks, which allows canopy-level greenness tracking within a parcel. Useful for ground-truthing ambiguous Sentinel-2 pixels at parcel boundaries or in very small holdings, but commercial tasking adds cost and coverage is not global by default.
What the canopy is actually doing in an on year
Olive alternate bearing is a physiological cycle, not a management choice. A heavy fruit load in an on year draws carbohydrates away from vegetative growth. The tree produces fewer new shoots, and the canopy thins relative to an off year. Late-season NDVI, measured in August and September across the Mediterranean, is measurably lower in on years because the fruit competes with leaf maintenance and because post-harvest defoliation stress is more pronounced when the crop is large.
The signal is not dramatic. Typical on-versus-off NDVI differences at the grove level are in the range of 0.03 to 0.08 NDVI units in late summer, based on published phenological studies of Olea europaea in Spain, Greece and Tunisia. That is a modest contrast, which means sensor calibration consistency matters enormously. Sentinel-2 surface-reflectance products (Level-2A) processed through the Sen2Cor atmospheric correction pipeline are the standard starting point. Mixing Level-1C and Level-2A data, or blending uncorrected Landsat Collection 1 with Collection 2, introduces artefacts that can exceed the signal you are trying to detect.
Building a time series that spans multiple cycles
A single year of NDVI data cannot confirm alternate bearing. You need at least four to six years to observe two full on/off cycles and distinguish the biological rhythm from drought years, late frosts or management changes such as irrigation installation. The practical workflow stacks annual late-summer NDVI composites, typically the median of all cloud-free observations between 1 July and 30 September, for each parcel or grove block.
Sentinel-2 provides the spatial resolution to delineate individual parcels, but its record only begins in 2015 for Sentinel-2A and 2017 for Sentinel-2B. For multi-cycle validation, Landsat Collection 2 Surface Reflectance fills the gap back to 1984, though at 30 m the smallest parcels become unreliable. A common approach is to use Landsat for cycle-phase identification at landscape scale and Sentinel-2 for parcel-level confirmation of the current cycle position.
Cloud cover is the persistent constraint across the Mediterranean. Southern Spain and Tunisia have relatively low cloud frequency in summer, but Greece and parts of Italy can lose 20 to 40 percent of July-September observations in some years. A single cloudy summer can break the time series exactly when you need it. Median compositing over the full July-September window, rather than relying on single-date imagery, reduces but does not eliminate this risk.
Separating the cycle signal from drought and irrigation
The honest difficulty is that drought stress and alternate bearing produce similar late-season NDVI depressions. A severe drought year can mimic an on-year signal in an off-year grove. Conversely, irrigated groves show compressed NDVI variation across the cycle because supplemental water buffers the carbohydrate stress. This means the method works most cleanly in rainfed, traditionally managed groves and becomes less reliable as irrigation intensity increases.
One published approach to disambiguation uses the phase relationship between consecutive years. Drought effects tend to depress NDVI across all parcels in a region simultaneously, regardless of their cycle phase. Alternate bearing produces a spatial mosaic: adjacent parcels that are out of phase with each other show opposite NDVI patterns in the same year. Detecting that mosaic structure, rather than a region-wide depression, is stronger evidence of the biological cycle. MODIS regional composites are useful here precisely because they aggregate across many parcels and reveal whether a region-wide or parcel-specific pattern dominates.
What the output is actually good for
Knowing the current cycle phase of a grove, confirmed from a multi-year series, has direct value for yield forecasting. If a grove is confirmed to be entering an on year, the prior probability of a substantial crop is high before a single flower has opened. That is useful for traders, cooperatives and agricultural ministries building early-season supply estimates. It is also useful for agronomists advising on thinning interventions, which can moderate the cycle and improve oil quality in on years.
For national agricultural statistics offices, a map of cycle phase across the olive-growing area provides a structural prior that improves in-season yield models. Rather than treating each year independently, the model can condition on the known phase. This is the logic behind Satellize's crop-estimation work, including the Tonga programme, where multi-year phenological context improves single-season estimates. The same principle applies to olive, though the Mediterranean setting brings its own fragmentation and cloud challenges.
Insurance and lending applications are also plausible. A lender financing an olive cooperative can assess whether the coming season is structurally an on or off year before committing credit. The satellite record does not replace agronomic field assessment, but it provides an independent, spatially continuous check that is not subject to the same reporting incentives as farmer declarations.
Resolution floors and what the method cannot do
At 10 m, Sentinel-2 resolves grove blocks reliably down to about 0.3 to 0.5 ha in practice, accounting for mixed pixels at parcel edges. Smaller holdings, common in parts of Greece and the Levant, are problematic. The NDVI of a 0.1 ha parcel measured at 10 m resolution is substantially contaminated by surrounding soil or other crops, and the alternate-bearing signal may be undetectable.
The method also cannot determine variety. Different olive cultivars have different alternate-bearing intensities. Arbequina, widely planted in super-high-density Spanish orchards, shows a weaker cycle than Koroneiki or Chemlali. A grove of low-bearing-intensity cultivars may show no detectable NDVI oscillation even when the biological cycle is present. Without ground-truth data on cultivar distribution, the satellite analysis will undercount cycle intensity in some regions.
Finally, this approach detects the cycle; it does not measure oil content, fruit size or harvest timing with precision. Those require additional inputs, whether in-season SAR backscatter, thermal data or field sampling. The NDVI time series is a structural diagnostic, not a complete yield model.
Typical figures
| Primary spatial resolution | 10 m (Sentinel-2 MSI, red/NIR bands) |
| Secondary spatial resolution | 30 m (Landsat 4–9); 250 m (MODIS MOD13Q1 for regional context) |
| Revisit frequency | 5 days at mid-latitudes (Sentinel-2A+B combined); 16 days (Landsat 8+9 combined); 16-day composites (MODIS MOD13Q1) |
| Key spectral bands | Red (B04, ~665 nm) and NIR (B08, ~842 nm) for NDVI on Sentinel-2; equivalent bands on Landsat OLI/TM/ETM+ |
| Archive depth | Sentinel-2: 2015–present; Landsat: 1984–present (Collection 2 Surface Reflectance) |
| Minimum detectable parcel size (reliable) | Approximately 0.3–0.5 ha at 10 m resolution; smaller parcels subject to mixed-pixel contamination |
| Typical on/off NDVI contrast | 0.03–0.08 NDVI units in late-summer composites (July–September); varies by cultivar and irrigation regime |
| Minimum cycle length for detection | 4–6 years of annual composites (two full on/off cycles) |
| Cloud-cover constraint | 20–40% of summer observations lost in cloudier Mediterranean sub-regions; median compositing over July–September mitigates but does not eliminate gaps |
| Delivery formats | GeoTIFF raster (annual NDVI composites and cycle-phase maps), vector parcel-level attribute tables (GeoPackage or Shapefile), CSV time series per parcel |
Analytics Satellize can run
| Annual late-summer NDVI composite per parcel | Median compositing of all cloud-free Sentinel-2 Level-2A observations in July–September window; NDVI = (NIR − Red) / (NIR + Red) | Annual GeoTIFF stack with per-parcel zonal statistics exported as attribute table |
| Multi-year NDVI time series and cycle-phase classification | Harmonic or anomaly-based time-series decomposition applied to annual composites; parcels classified as on-year, off-year or indeterminate based on oscillation phase and amplitude | Parcel-level GIS layer with cycle-phase attribute and confidence score; updated annually after each summer compositing window |
| Landsat-extended historical cycle record | Cross-calibrated NDVI time series merging Landsat Collection 2 (1984–present) with Sentinel-2; used to validate current phase against prior cycles | Per-parcel CSV time series with flagged on/off years and data-quality indicators (cloud fraction, observation count per window) |
| Regional cycle-synchrony map | MODIS MOD13Q1 250 m NDVI composites aggregated to administrative or watershed units; spatial autocorrelation analysis to distinguish region-wide drought signal from parcel-mosaic alternate-bearing pattern | Province-level summary report with synchrony index and drought-versus-cycle disambiguation flag |
| Early-season on/off year prior for yield model | Cycle-phase classification from prior years used as Bayesian prior in in-season yield estimation; combined with current-season NDVI trajectory as season progresses | Structured data feed (JSON or CSV) ingested by client yield-forecasting system; updated at each new Sentinel-2 acquisition |
| Irrigated versus rainfed grove segmentation for method applicability flagging | Summer NDVI variance across years used as proxy for irrigation buffering; high inter-annual stability flagged as likely irrigated and excluded from cycle-phase confidence tier | Binary irrigation-likelihood attribute appended to parcel layer, with recommendation on reliability of cycle-phase output per parcel |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.