Soybean leaf chlorophyll and nitrogen status estimation from multispectral indices
Red-edge reflectance from Sentinel-2 and Planet SuperDove lets agronomists estimate soybean leaf chlorophyll concentration as a proxy for canopy nitrogen status, but saturation at high LAI and atmospheric correction quality set real limits on what satellite data can reliably retrieve.
Sensors
- Sentinel-2 MSI: 10 m visible bands, 20 m red-edge bands (B5 at 705 nm, B6 at 740 nm, B7 at 783 nm) and NIR; 5-day revisit at the equator with both satellites. The red-edge triplet is the primary input for CIre and MTCI indices. Free and open.
- Planet SuperDove: 3 m ground sampling distance, 8 bands including a red-edge band centred near 705 nm. Daily revisit over most agricultural regions. Narrower red-edge coverage than Sentinel-2's triplet limits multi-index approaches, but spatial detail resolves within-field gradients invisible to Sentinel-2.
- RapidEye (archive): 5 m, 5-band sensor with a red-edge channel at 690–730 nm. No longer operational but archive extends to 2009, useful for multi-season trend analysis or calibration dataset construction.
- Landsat 8/9 OLI: 30 m, 8-day revisit (16-day per satellite). No dedicated red-edge band; the transition from Band 4 (red, 630–680 nm) to Band 5 (NIR, 845–885 nm) straddles but does not sample the red-edge slope. Useful for NDVI-based nitrogen proxies in low-biomass conditions only. Free and open.
Why the red edge is where the physics lives
Chlorophyll a and b absorb strongly at around 670 nm. Move to 700–750 nm and reflectance rises steeply because chlorophyll absorption drops away while leaf internal scattering takes over. This slope, the red-edge inflection, shifts position and steepness with chlorophyll concentration. At low chlorophyll content the inflection sits closer to 700 nm; at high content it moves toward 740 nm. Satellite sensors that sample within this window can therefore infer chlorophyll without touching the leaf.
The connection to nitrogen matters because roughly half the nitrogen in a soybean leaf is bound up in photosynthetic enzymes, primarily RuBisCO, which contains chlorophyll. Leaf chlorophyll concentration measured with a SPAD-502 meter correlates with total leaf nitrogen across a wide range of growth stages, though the relationship is not linear and varies with genotype and sulphur supply. Satellite-derived chlorophyll indices are therefore nitrogen proxies, not nitrogen measurements. That distinction matters when interpreting results.
Indices that do the work, and what each one assumes
The Chlorophyll Index red-edge (CIre) is defined as (R783 / R705) minus 1, using Sentinel-2 bands B7 and B5. It is empirically and theoretically linked to canopy chlorophyll content per unit ground area, not per unit leaf area, which is an important distinction. The MERIS Terrestrial Chlorophyll Index (MTCI) uses (R754 minus R709) divided by (R709 minus R681), approximated with Sentinel-2 B6, B5 and B4. Both indices were shown in published literature to saturate at canopy chlorophyll contents above roughly 50–70 µg cm⁻², a threshold soybean crops can exceed at peak vegetative growth.
Saturation is the most consequential limit of these methods. Once the canopy closes and LAI exceeds approximately 3–4 m² m⁻², additional chlorophyll in lower canopy layers contributes little to top-of-canopy reflectance. The satellite effectively sees only the top fraction of the canopy. This means index values plateau even as leaf nitrogen continues to vary, and the method loses sensitivity precisely when the crop is most productive. Radiative-transfer modelling with PROSAIL or similar can partially correct for LAI effects if concurrent LAI estimates are available, but that adds a calibration step and its own uncertainty.
From top-of-canopy reflectance to a SPAD-equivalent number
Sentinel-2 Level-2A products deliver surface reflectance after atmospheric correction with the Sen2Cor processor. Quality is generally adequate for red-edge indices, though adjacency effects near field boundaries and residual aerosol errors can introduce noise of a few percent in reflectance, which translates to meaningful index uncertainty. Planet SuperDove surface reflectance products use their own atmospheric correction pipeline; inter-scene consistency is generally good but cross-calibration against Sentinel-2 is advisable before mixing data sources.
Calibration against ground SPAD readings requires sampling across the expected range of nitrogen status, ideally spanning at least three growth stages and multiple nitrogen treatments. A regression of index against SPAD typically achieves R² values of 0.70–0.85 in published soybean studies, with residual scatter driven by LAI variation, leaf angle distribution and measurement timing relative to overpass. Applying a calibration built in one season to another without revalidation is a common source of error. The calibration is site- and season-specific to a degree that users often underestimate.
Cloud cover, revisit, and the agronomic window
Nitrogen management decisions in soybean typically occur between growth stages V3 and R3, a window of four to eight weeks depending on variety and latitude. In humid tropical and subtropical growing regions, cloud cover during this period can block usable acquisitions for weeks at a time. Sentinel-2's five-day revisit does not guarantee a cloud-free image within an agronomic decision window. Planet SuperDove's daily revisit improves the odds substantially, but cloud probability is the same sky regardless of the sensor.
Practical programmes therefore build compositing strategies: selecting the least-cloudy pixel from a rolling window of acquisitions rather than relying on a single date. This introduces phenological mixing if the window is too wide. A ten-day composite may blend V4 and V6 reflectance from different parts of a field, which is acceptable for slow-changing chlorophyll status but introduces error if the field is spatially variable. There is no clean answer here; the right window width depends on local cloud climatology and how quickly the crop is developing.
What the method can and cannot tell a grower
A well-calibrated CIre or MTCI layer at 10–20 m resolution can identify zones within a field where chlorophyll, and by inference nitrogen, is substantially below the field mean. Differences of around 5 SPAD units or more are generally detectable above the noise floor of the satellite-plus-calibration pipeline. Smaller differences are not reliably distinguishable from atmospheric and calibration artefacts.
The method cannot diagnose the cause of low chlorophyll. Iron deficiency chlorosis, sulphur deficiency, waterlogging stress and early soybean cyst nematode infestation all reduce chlorophyll and produce similar index signatures. Ground verification remains necessary before any variable-rate nitrogen application decision. Satellize runs CIre and MTCI analytics on Sentinel-2 and Planet data for crop monitoring programmes, including its crop-estimation work in Tonga, and the same pipeline applies to soybean nitrogen monitoring with the calibration steps described above. The satellite layer narrows the area requiring ground inspection; it does not replace it.
Building a programme that produces actionable numbers
A credible soybean chlorophyll monitoring programme needs four things: consistent atmospheric correction, a local calibration dataset spanning the expected SPAD range, a cloud-management strategy matched to local climatology, and an honest uncertainty estimate attached to every output layer. Skipping any of these steps produces numbers that look precise but mislead.
Sentinel-2 Level-2A is the practical starting point for most programmes because it is free, the red-edge triplet is well-suited to CIre and MTCI, and the archive extends to 2017, allowing multi-season analysis. Planet SuperDove adds spatial detail and revisit frequency at cost. RapidEye archive data can extend the historical record for fields where long-term nitrogen management history matters. Landsat is best reserved for coarser regional assessments where its lack of a red-edge band is less limiting, such as tracking inter-field variability across a district rather than within-field prescription mapping.
Typical figures
| Spatial resolution (Sentinel-2 red-edge) | 20 m (B5, B6, B7); resampled to 10 m in some workflows |
| Spatial resolution (Planet SuperDove) | 3 m ground sampling distance |
| Revisit frequency | 5 days (Sentinel-2 constellation); daily (Planet SuperDove) |
| Key spectral bands | Red ~670 nm, red-edge 705–783 nm, NIR ~842 nm (Sentinel-2); red-edge ~705 nm (SuperDove) |
| Minimum detectable SPAD difference | Approximately 5 SPAD units above noise floor; smaller differences unreliable |
| Index saturation threshold | Canopy chlorophyll content roughly 50–70 µg cm⁻²; LAI above ~3–4 m² m⁻² |
| Sentinel-2 archive depth | From 2015 (Sentinel-2A launch); consistent Level-2A products from 2017 |
| Atmospheric correction | Sen2Cor (Sentinel-2 Level-2A); Planet surface reflectance pipeline (SuperDove) |
| Typical calibration accuracy | R² 0.70–0.85 against ground SPAD in published soybean studies; site- and season-specific |
| Delivery formats | GeoTIFF index layers, field-mean time series (CSV), zonal statistics by management zone |
Analytics Satellize can run
| CIre canopy chlorophyll map | Sentinel-2 B7/B5 ratio minus 1; surface reflectance from Level-2A; cloud masking with Scene Classification Layer | GeoTIFF layer per acquisition date, field-boundary zonal statistics table |
| MTCI time series for growth-stage tracking | MERIS Terrestrial Chlorophyll Index approximated from Sentinel-2 B4, B5, B6; composited over rolling 10-day window | Per-field CSV time series with confidence flags for cloud fraction and LAI saturation risk |
| Within-field nitrogen-status zone map | K-means or threshold classification of CIre at 20 m; zones labelled low/medium/high relative to field mean | Shapefile or GeoPackage of management zones, suitable for variable-rate equipment import |
| SPAD-equivalent estimation layer | Linear or polynomial regression of CIre/MTCI against client-supplied ground SPAD dataset; uncertainty bounds propagated from calibration residuals | Raster layer with per-pixel SPAD estimate and ±1 standard error band; calibration report |
| Cloud-gap analysis and acquisition planning report | Historical cloud-cover frequency from Sentinel-2 Scene Classification Layer archive over client field locations | PDF report showing probability of cloud-free acquisition by week across the agronomic season |
| Multi-season chlorophyll trend analysis | Annual peak-season CIre from Sentinel-2 archive (2017 to present); anomaly detection relative to field-level baseline | Multi-year GeoTIFF stack and trend summary report identifying fields with declining or improving nitrogen status |
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.