National crop calendar validation against satellite phenology
Harmonic analysis of MODIS and Sentinel-2 NDVI time series extracts green-up, peak and senescence dates at administrative-unit level, then tests them against FAO GAEZ and national ministry calendars to surface reporting errors and climate-driven shifts.
Sensors
- MODIS MOD13Q1 (Terra/Aqua): 250 m NDVI and EVI composites at 16-day intervals, global coverage. The long archive from 2000 onward makes it the standard backbone for multi-year phenological trend analysis, though 250 m is too coarse to resolve smallholder fields below roughly 1–2 ha reliably.
- Sentinel-2 MSI (Sentinel-2A/2B): 10 m (visible/NIR) and 20 m (red-edge) multispectral imagery with a combined 5-day revisit at the equator. Red-edge bands (B5, B6, B7 at 705–783 nm) are particularly sensitive to chlorophyll content, improving green-up detection in sparse canopies. Cloud contamination can reduce effective revisit to 20–30 days in persistently overcast regions.
- Landsat 8/9 OLI: 30 m resolution, 16-day revisit per satellite (8-day combined). Consistent radiometric calibration and a continuous archive back to 1984 allow long-baseline phenological comparisons. Useful for validating whether calendar shifts are recent or decades-old. Temporal density is lower than Sentinel-2, making it less suited to capturing rapid green-up events.
- MODIS MOD09GA daily surface reflectance: 500 m daily observations used to fill gaps in the 16-day composite product and to detect abrupt within-season anomalies. Spatial resolution limits its use to large administrative units or regions with homogeneous crop systems.
What a phenological mismatch actually tells you
A national crop calendar is a policy document. It encodes when farmers are assumed to plant and harvest, and it underpins everything from import forecasts to food-aid pre-positioning. The problem is that many calendars were written once, validated loosely, and have not been updated to reflect two decades of shifting monsoon onset, rising temperatures and expanding irrigation. When satellite phenology disagrees with the official calendar by more than a week or two, something real is happening: either the calendar was always wrong, or conditions have changed.
The distinction matters operationally. A systematic bias, where the satellite consistently shows green-up 15 days earlier than the calendar across many years, suggests the original calendar was based on poor ground data. An accelerating trend, where the gap widens by roughly one to two days per year, points toward a climate-driven shift. A single-year anomaly points toward drought, flood or pest pressure. Each demands a different policy response, and satellite time series are the only source with the spatial coverage and temporal depth to separate them.
How harmonic analysis extracts phenological dates
The standard approach fits a sum of sinusoidal functions, typically two or three harmonics, to the NDVI or EVI time series for each pixel or administrative-unit mean. The fitted curve is smooth enough to suppress noise from cloud contamination and atmospheric scattering, yet retains the seasonal shape. From the fitted curve, four key dates are extracted: green-up onset (the point where NDVI crosses a threshold, often 20% of the seasonal amplitude above the base), peak greenness, senescence onset, and dormancy. These are compared directly to the planting and harvest windows in the FAO Global Agro-Ecological Zones database or the relevant national ministry publication.
MODIS MOD13Q1 is the workhorse for this at national scale. Its 16-year-plus archive and consistent 16-day compositing make it suitable for trend detection. Sentinel-2 adds spatial specificity: where MODIS flags a discrepancy at district level, Sentinel-2 at 10 m can show whether the anomaly is concentrated in irrigated perimeters, rainfed uplands or specific crop types. The two sensors are not interchangeable. MODIS gives you the trend; Sentinel-2 gives you the geography of the deviation.
One honest caveat: harmonic fitting assumes a roughly periodic signal. In regions with two cropping seasons of unequal intensity, or where intercropping blurs the NDVI profile, the fitted curve can misattribute senescence from one season as green-up from another. Quality-control steps, including land-cover masking and manual inspection of flagged administrative units, are not optional.
The FAO GAEZ and WFP CropMonitor as reference baselines
FAO's Global Agro-Ecological Zones database provides crop-specific growing period windows at roughly 5 arc-minute resolution, derived from climate normals and agronomic models. WFP's CropMonitor for Food Security publishes monthly assessments that already integrate satellite NDVI with reported planting progress for major food-insecure countries. These are the two most widely used reference baselines for calendar validation work.
Neither is infallible. GAEZ growing periods are modelled, not observed, and they carry the assumptions of the underlying climate data, which in data-sparse regions can be 30-year normals that no longer reflect current conditions. WFP CropMonitor assessments are expert-synthesised and inherently qualitative at the sub-national level. Satellite harmonic analysis offers a reproducible, quantitative complement: it does not replace expert judgement, but it gives that judgement a consistent empirical anchor.
Resolution floors, cloud and the limits of the method
At 250 m, MODIS cannot resolve fields smaller than roughly 6 hectares without significant mixed-pixel contamination. In fragmented smallholder landscapes, particularly across much of sub-Saharan Africa and South and Southeast Asia, the NDVI signal from a single MODIS pixel may blend crop, fallow, trees and bare soil. This tends to dampen the apparent seasonal amplitude and push detected green-up dates later than the true crop green-up. Sentinel-2 at 10 m largely resolves this, but its shorter archive (operational from 2017) limits trend analysis.
Cloud cover is the other persistent constraint. In humid tropical regions with long wet seasons, effective Sentinel-2 revisit can drop to one clear observation per month or fewer during the growing season, which is precisely when phenological timing matters most. Synthetic aperture radar backscatter can partially substitute for optical NDVI in cloud-affected periods, but SAR-derived phenological metrics are noisier and less directly comparable to the NDVI-based calendar definitions used by FAO and WFP. This page does not cover SAR phenology in detail; that ground belongs to the crop yield and irrigated-area pages in this library.
Finally, NDVI is a proxy. It measures canopy greenness, not crop identity. A green-up signal in a district that the calendar assigns to maize might actually be from a weed flush after early rains, a volunteer crop or a different species entirely. Validation against crop-type maps is essential before drawing conclusions about a specific commodity calendar.
Turning a discrepancy into a recommendation
The output of a calendar validation exercise is not a list of pixels. It is a structured comparison table: for each administrative unit and each crop season, the satellite-derived date, the official calendar date, the difference in days, and a classification of the discrepancy as systematic bias, trend, or single-year anomaly. That table is what a ministry of agriculture or a WFP country office can act on.
Systematic biases of more than two weeks in a high-production zone warrant a formal calendar revision. Trends of one to two days per year of earlier green-up, consistent with documented warming in the published literature on agricultural phenology, warrant a monitoring protocol that revisits the calendar every three to five years. Single-year anomalies feed directly into early-warning systems: a district where harvest is running three weeks late relative to both the official calendar and the multi-year satellite baseline is a district that may need food-security attention before the next assessment cycle.
Satellize ran a related exercise for the Kingdom of Tonga crop-estimation programme, where sparse ground data made satellite phenology the primary source of seasonal timing information. The methodological lessons from that engagement, particularly around harmonic fitting in small-island, multi-crop environments, inform how we approach calendar validation in similarly data-sparse settings. Analysts wanting to explore a specific country calendar can request a scoping call with our agriculture analytics team.
Typical figures
| Spatial resolution (phenological mapping) | 250 m (MODIS MOD13Q1); 10–20 m (Sentinel-2 MSI); 30 m (Landsat 8/9) |
| Temporal compositing interval | 16 days (MODIS MOD13Q1 standard product); 5-day effective revisit (Sentinel-2A+2B combined, cloud-free conditions) |
| Archive depth | MODIS: 2000 to present; Landsat: 1984 to present; Sentinel-2: 2017 to present |
| Key spectral bands for NDVI | Red (620–670 nm) and NIR (841–876 nm) for MODIS; B4/B8 for Sentinel-2; Band 4/5 for Landsat OLI |
| Minimum field size for reliable NDVI extraction | Approximately 6 ha at MODIS 250 m; approximately 0.1 ha at Sentinel-2 10 m (mixed-pixel effects increase below these thresholds) |
| Phenological date precision (published studies) | Typically ±5–10 days for green-up onset under good data conditions; degrades to ±15–20 days in cloud-affected or fragmented landscapes |
| Reference calendar sources | FAO GAEZ v4 growing period database; WFP CropMonitor for Food Security; national ministry of agriculture publications |
| Delivery formats | GeoTIFF phenological date layers; CSV/Excel comparison tables by administrative unit; PDF summary report; GIS-ready shapefiles |
Analytics Satellize can run
| Administrative-unit phenological date extraction | Harmonic regression (double-logistic or Fourier series) on MODIS MOD13Q1 16-day NDVI composites, producing green-up onset, peak, senescence and dormancy dates per unit | GeoTIFF and CSV table of four phenological dates per administrative unit per crop season |
| Calendar discrepancy classification | Difference calculation between satellite-derived dates and FAO GAEZ or national ministry calendar windows; classified as systematic bias, monotonic trend or single-year anomaly using multi-year time series | Structured comparison table (Excel/CSV) with discrepancy magnitude, direction and classification per unit |
| High-resolution phenological verification | Sentinel-2 MSI NDVI and red-edge index time series (10–20 m) applied to flagged administrative units to localise discrepancies within sub-district crop zones | GeoTIFF phenological maps at 10 m for flagged districts; annotated PDF report |
| Multi-year trend analysis | Mann-Kendall trend test on annual green-up and harvest dates derived from MODIS and Landsat archives (2000 to present), identifying statistically significant calendar shifts | Trend magnitude (days per decade) and significance level per administrative unit; GIS layer and summary chart |
| Cloud-gap-filled NDVI time series | Temporal interpolation and Savitzky-Golay smoothing applied to Sentinel-2 L2A surface reflectance to reconstruct a dense, gap-filled NDVI stack for cloud-affected regions | Gap-filled NDVI GeoTIFF stack at 10 m, ready for downstream harmonic fitting |
| Early-warning flag for late or failed seasons | Real-time comparison of current-season NDVI trajectory against the multi-year satellite baseline; alert triggered when green-up or peak date deviates beyond a configurable threshold | Automated alert report (PDF or API feed) listing flagged administrative units with deviation magnitude and historical context |
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.