Sensors
- Planet SuperDove: 3 m native resolution, near-daily global revisit (constellation of ~200 satellites), 8 spectral bands including red-edge at 705 nm. The daily cadence is the critical asset here: it can resolve a planting event to within 1–3 days in cloud-free conditions, which no freely available optical constellation matches.
- Sentinel-2 MSI (ESA): 10 m resolution in visible and near-infrared bands, 20 m in red-edge and shortwave infrared. Nominal 5-day revisit at the equator (two satellites combined). In practice, cloud cover often extends the effective gap to 10–20 days in humid tropics, which can blur planting-date estimates by a similar margin.
- Sentinel-1 SAR (ESA): C-band (5.405 GHz), 10 m resolution in Interferometric Wide swath mode, 6-day repeat per satellite (12-day per satellite alone, 6-day with both). Backscatter from emerging canopy rises measurably once leaf area index exceeds roughly 0.3–0.5, making this the primary tool when cloud contamination blocks optical sensors for weeks.
- Landsat 8/9 OLI (USGS/NASA): 30 m resolution, 8-day combined revisit (two satellites). Useful for historical archive work back to 1984 (Landsat 5 TM onwards) and for cross-calibrating Sentinel-2 time series. Too coarse for individual smallholder plots below roughly 0.5 ha, and the revisit is too slow to pin planting dates tightly in a single season.
What the signal actually looks like
When a field transitions from bare soil or crop residue to an established seedling canopy, near-infrared reflectance rises sharply while red reflectance falls. The normalised difference vegetation index (NDVI) captures this ratio, but the red-edge bands available on Sentinel-2 (at 705 nm and 740 nm) and Planet SuperDove are more sensitive at low canopy densities, where NDVI is still suppressed by soil background. This matters because the signal of interest is the first 7–14 days after emergence, not the peak of the season.
In SAR, the mechanism is different. Bare soil produces a relatively stable backscatter signal governed by surface roughness and moisture. As a seedling canopy develops, volume scattering from leaves and stems adds to that return, and the VV/VH cross-polarisation ratio shifts. The change is detectable but modest at C-band during early establishment, typically a 1–3 dB rise in VH backscatter. It becomes more reliable once the canopy reaches a leaf area index of around 0.5, which for dense-planted cereals may occur within 10–14 days of emergence.
Why temporal density is the binding constraint
A single cloud-free image tells you a crop exists. A dense time series tells you when it started. The difference matters enormously for area estimation, growing-degree-day modelling, and harvest-date forecasting downstream.
The minimum temporal density needed for a planting-date estimate accurate to within roughly one week is generally accepted in the literature as one clear observation per 5–8 days during the planting window. Sentinel-2 alone achieves this in semi-arid regions with low cloud frequency. In humid tropical environments, effective clear-sky observations can drop to one or two per month, which degrades planting-date precision to two to four weeks. Planet SuperDove's daily cadence substantially narrows this gap, but even Planet imagery is cloud-affected; the advantage is statistical. With daily attempts, the probability of capturing at least one clear observation within any five-day window is much higher than with a five-day sensor that may land entirely on a cloudy day.
The practical answer for high-value applications is sensor fusion: use Sentinel-1 SAR to maintain a continuous signal through cloud, anchor the phenological curve with clear Sentinel-2 or Planet observations on either side of the cloud gap, and fit a change-point model to the combined series. This is not a theoretical approach; it is operationally deployed in programmes such as the EU's LUCAS crop monitoring work and in FAO's WaPOR data service.
Transplanted crops do not look like direct-seeded ones
Direct-seeded crops (maize, wheat, sorghum, most cereals) produce a gradual emergence signal: bare soil gives way to sparse seedlings over 7–14 days, and the vegetation index rises smoothly. Transplanted crops (paddy rice in many Asian systems, tobacco, some vegetables) behave differently. The field is flooded or prepared, seedlings are grown in a nursery off-site, and then a large number of established plants are inserted into the field in a single day or two. The NDVI jump is therefore abrupt and large, sometimes mimicking a step function rather than a ramp.
This distinction matters for algorithm design. A change-point detector tuned for gradual emergence will misdate or miss transplanting events. Paddy rice is the most documented case: the flood-drain-transplant cycle produces a characteristic SAR backscatter signature (high backscatter during flooding, drop at transplanting as the water surface is partly obscured, then rise as canopy develops) that has been used to map rice planting dates across South and Southeast Asia at national scale, including in studies using Sentinel-1 published through ESA's research programme. Failing to separate these two establishment modes introduces systematic bias into any area or production estimate.
Honest limits of the method
Cloud contamination is the dominant source of error in optical-only approaches, and it is not uniformly distributed. West African and South Asian planting seasons frequently coincide with peak cloud cover, which is also when planting-date detection matters most. SAR fills the gap but introduces its own ambiguities: soil moisture changes, flooding, and crop residue all shift C-band backscatter in ways that can be confused with canopy emergence.
Spatial resolution sets a hard floor on the minimum field size that can be reliably monitored. At 10 m (Sentinel-2), fields smaller than roughly 0.1 ha are dominated by edge mixing with adjacent land cover. At 30 m (Landsat), that floor rises to around 0.5–1 ha. Smallholder agriculture across sub-Saharan Africa and South Asia is heavily concentrated below these thresholds, which means that regional planting-date maps derived from freely available data carry real uncertainty at the farm level even when they are accurate at district or national scale.
Archive depth is an underappreciated constraint. Sentinel-2 data begins in 2015 (Sentinel-2A) and 2017 (Sentinel-2B). Sentinel-1 begins in 2014. For multi-year baseline work, Landsat's archive back to the early 1980s remains irreplaceable, even at coarser resolution.
From detection to a decision-ready product
A planting-date map is not itself a decision. It becomes useful when it feeds a growing-degree-day accumulation model (to estimate likely maturity date), a planted-area estimate (to anchor production forecasts), or an insurance trigger (to verify that planting occurred within a policy window). Each downstream use imposes different accuracy requirements on the planting-date estimate itself.
Satellize's crop-estimation work for the Kingdom of Tonga illustrates the chain: dense time series from open constellations, combined with local agronomic knowledge about planting calendars, feeds an area and condition estimate that would otherwise require ground surveys the country lacks the capacity to run. The analytic output is a dated planted-area layer, updated at each clear-sky pass, with a confidence interval that widens honestly when cloud contamination has been heavy. That confidence interval is part of the product, not a footnote.
Readers building a monitoring system should specify their required planting-date accuracy (in days), their minimum field size, and their cloud climatology before choosing a sensor stack. Those three parameters determine the answer almost completely.
Typical figures
| Spatial resolution (optical) | 3 m (Planet SuperDove), 10–20 m (Sentinel-2 MSI), 30 m (Landsat 8/9 OLI) |
| Spatial resolution (SAR) | 10 m (Sentinel-1 IW mode, ground range) |
| Revisit cadence | Near-daily (Planet); 5-day nominal, often longer effective (Sentinel-2); 6-day (Sentinel-1, two-satellite); 8-day (Landsat 8+9 combined) |
| Planting-date detection accuracy (typical) | ±3–7 days in low-cloud environments with daily optical; ±10–20 days in humid tropics with optical only; improves with SAR fusion |
| Minimum detectable field size | ~0.1 ha at 10 m resolution; ~0.5–1 ha at 30 m; sub-0.05 ha feasible with Planet 3 m |
| Key spectral bands | Red (665 nm), NIR (842 nm), red-edge (705, 740 nm) for optical; C-band VV/VH for SAR |
| Archive depth | Sentinel-2 from 2015; Sentinel-1 from 2014; Landsat from 1984 (TM); Planet from ~2016 (limited early archive) |
| Latency (open data) | Sentinel-1/2: typically 1–3 hours after acquisition via Copernicus Data Space. Landsat: same-day via EarthData. Planet: varies by licence. |
| Delivery formats | GeoTIFF (planting-date raster, days-of-year), GeoPackage or Shapefile (field polygons with date attributes), CSV (district-level summaries), JSON feed (alert triggers) |
Analytics Satellize can run
| Pixel-level planting-date map | Change-point detection on dense NDVI or red-edge VI time series (e.g. BFAST, EWMACD, or logistic-curve fitting to phenological inflection); cloud-gap filling via harmonic regression or SAR fusion | GeoTIFF raster, one value per pixel in day-of-year, with per-pixel confidence score; updated each clear-sky pass |
| Planted-area estimate by administrative unit | Thresholding of planting-date map against agronomic calendar windows; area summation within polygon boundaries | CSV or dashboard table showing hectares planted by district and crop calendar week, with uncertainty range |
| Transplant vs. direct-seed classification | SAR backscatter time series analysis for flood-transplant signature (paddy rice); step-function vs. ramp-function discrimination in optical VI series | Binary GIS layer (transplanted / direct-seeded) overlaid on planting-date map |
| Planting-date anomaly alert | Comparison of current-season planting-date raster against multi-year median derived from Sentinel-2 and Landsat archive; anomaly flagged where deviation exceeds one standard deviation | Automated alert (email or API push) with anomaly map showing early, on-time, or delayed establishment by zone |
| Projected maturity date layer | Growing-degree-day accumulation from planting-date estimate, driven by ERA5 or CHIRPS temperature data, parameterised with published crop-specific thermal thresholds | GeoTIFF of estimated maturity date (day-of-year) and associated harvest-window report |
| Insurance planting-verification report | Intersection of policy holder's declared field boundary with planting-date raster; verification that establishment occurred within policy window at stated location | Per-policy PDF or structured JSON record with satellite-derived planting date, confidence interval, and cloud-contamination flag |
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.