Invasive aquatic vegetation mapping: water hyacinth and floating macrophytes
Water hyacinth and floating macrophytes spread fast enough to close navigation channels and crash fisheries within weeks. Sentinel-2 red-edge bands, Planet SuperDove daily revisit and Sentinel-1 SAR together provide the spectral discrimination and cloud-penetrating coverage the problem demands.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands; 20 m in red-edge bands (B5, B6, B7) and shortwave infrared. Five-day revisit at the equator with both Sentinel-2A and 2B operating. The red-edge and NIR combination is the primary spectral discriminator between floating macrophytes and open water or submerged vegetation. Cloud cover is the binding constraint in humid tropical zones.
- Planet SuperDove: 3 m resolution, daily revisit globally, eight spectral bands including two red-edge channels. The combination of sub-daily revisit and red-edge sensitivity allows detection of mat expansion events that would fall between Sentinel-2 acquisitions. Commercial licence required; latency can be under 24 hours with priority tasking.
- Sentinel-1 SAR (C-band): IW mode provides 10 m resolution with 6-day revisit and is unaffected by cloud or rain. Dense water hyacinth mats produce a measurable increase in C-band backscatter relative to open water, enabling detection when optical data is unavailable. SAR cannot reliably distinguish hyacinth from other dense floating macrophytes on backscatter alone; it works best as a cloud-gap filler alongside optical classification.
- DESIS (ISS hyperspectral): The DLR Earth Sensing Imaging Spectrometer on the ISS delivers 235 spectral bands at approximately 2.55 nm sampling, 30 m spatial resolution, covering 400–1000 nm. Hyperspectral data allows species-level discrimination between water hyacinth, water lettuce and water fern where multispectral data cannot separate them. ISS orbital inclination limits coverage to roughly 51.6° latitude, and revisit is irregular and non-schedulable by users.
Why a fortnight changes everything
Water hyacinth (Eichhornia crassipes) holds a documented doubling time of 11 to 18 days under warm, nutrient-rich conditions. That figure is not a worst case; it is the typical range reported across African and Asian reservoir studies. A mat that covers 5 % of a reservoir surface in early January can cover 40 % by mid-February. Fisheries collapse, navigation closes, and hydropower intakes block on timescales that monthly satellite composites simply cannot track.
The operational implication is blunt: a monitoring programme with a revisit longer than five to seven days will routinely miss the inflection point between nuisance and crisis. This is the primary reason Planet SuperDove's daily cadence matters for this application, even though Sentinel-2's spectral depth is superior for classification accuracy.
What the red-edge actually reveals
Floating macrophytes produce a distinctive spectral profile. Chlorophyll absorption creates a strong reflectance minimum around 670 nm and a steep rise through the red-edge (700–740 nm) into a high NIR plateau. Open water absorbs NIR almost completely, giving near-zero reflectance above 750 nm. Submerged vegetation is attenuated by the water column and lacks the sharp red-edge inflection. Emergent vegetation such as papyrus or reed beds shares some of these features but tends to have higher SWIR reflectance due to structural differences in the canopy.
Sentinel-2 bands B5 (705 nm), B6 (740 nm) and B7 (783 nm) sit precisely across this inflection. Indices derived from these bands, particularly the Red-Edge Chlorophyll Index and the Floating Algae Index, achieve published classification accuracies above 90 % for dense hyacinth mats in clear-sky conditions. Accuracy drops for thin or fragmented mats, for mixed pixels at the mat edge, and when atmospheric correction over water is imperfect. The 20 m resolution of the red-edge bands means mats smaller than roughly 0.04 hectares fall below reliable detection.
SAR as the cloud-gap filler, not the primary classifier
Tropical reservoirs spend months under persistent cloud cover. In the Congo Basin, Lake Victoria catchment and across South and Southeast Asia, optical revisit can effectively collapse to one usable scene per month during the wet season. That is precisely when hyacinth growth is fastest, because nutrient runoff peaks with rainfall.
Sentinel-1 C-band SAR fills the gap partially. Dense floating vegetation increases volume scattering and double-bounce returns relative to specular open water, which appears very dark in SAR imagery. Published studies on Lake Victoria and Tana Lake in Ethiopia have demonstrated that Sentinel-1 VV and VH polarisation backscatter can separate dense hyacinth mats from open water with overall accuracies in the range of 80 to 88 %. The limitation is real: SAR cannot distinguish water hyacinth from water lettuce or dense algal scum on backscatter alone. It tells you something is there; optical data tells you what it is. A combined optical-SAR workflow, using SAR detections to flag areas for optical follow-up and to interpolate between clear-sky optical scenes, is more informative than either sensor alone.
Hyperspectral data and the species question
Most management interventions are species-specific. Herbicide selection, biological control agents and mechanical harvesting strategies differ between water hyacinth, water lettuce (Pistia stratiotes) and giant salvinia (Salvinia molesta). Multispectral sensors struggle to separate these species when they co-occur, because their red-edge and NIR profiles overlap substantially.
DESIS hyperspectral data, with its 235 narrow bands, allows derivative spectroscopy and spectral unmixing approaches that can separate co-occurring species at the sub-pixel level, at least for dense monoculture patches. The honest caveat is coverage: DESIS revisit over any given tropical lake is irregular and may be weeks apart. It is best used for periodic ground-truth and species-composition surveys rather than operational weekly monitoring. Airborne hyperspectral campaigns offer finer spatial resolution but at a cost and logistics burden that rules them out for most government programmes.
Building an operational monitoring workflow
An effective programme combines three layers. First, a Sentinel-2 baseline classification at five-day revisit, using red-edge indices and a supervised classifier trained on field-verified samples, produces a time series of mat extent polygons. Second, Sentinel-1 SAR scenes fill cloud-affected epochs with lower-confidence extent estimates flagged as SAR-only. Third, Planet SuperDove daily imagery, triggered when the Sentinel-2 time series shows rapid expansion, provides the high-cadence confirmation needed to dispatch physical response teams.
Change detection between successive classifications is more operationally useful than raw extent maps. A lake authority needs to know that the northern inlet gained 120 hectares in six days, not simply that 340 hectares are covered today. Alert thresholds tied to expansion rate rather than absolute area are more likely to prompt timely intervention.
Satellize has applied comparable multi-sensor classification workflows in its analytics work, including the Tonga crop-estimation programme, and can configure the same architecture for reservoir and river monitoring. Delivery formats range from GIS polygon layers to automated alert feeds, depending on what the client's operations centre can ingest.
Limits worth stating plainly
No satellite system currently resolves individual hyacinth rosettes, which are 20 to 50 cm across. Mapping is of mat-level aggregations. Thin mats of one to two plant layers are frequently misclassified as turbid water or sparse algae by multispectral classifiers. Atmospheric correction over inland water remains a source of systematic error, particularly in the blue and green bands used to distinguish water from vegetation at low densities.
SAR backscatter from floating vegetation is sensitive to wind roughening of the water surface and to the moisture state of the mat itself. A recently harvested or senescent mat may have backscatter values close to open water. Validation with concurrent field surveys or drone imagery is not optional for a programme that will be used to make enforcement or resource-allocation decisions.
Typical figures
| Optical spatial resolution (classification) | 10 m (Sentinel-2 RGB/NIR); 20 m (Sentinel-2 red-edge); 3 m (Planet SuperDove) |
| SAR spatial resolution | 10 m (Sentinel-1 IW mode, ground range) |
| Optical revisit (equatorial) | 5 days (Sentinel-2A+B combined); daily (Planet SuperDove) |
| SAR revisit | 6 days (Sentinel-1, single satellite) |
| Key spectral bands for classification | Red-edge: 705, 740, 783 nm (Sentinel-2 B5/B6/B7); NIR: 842 nm; SAR: C-band 5.405 GHz, VV and VH polarisation |
| Minimum detectable mat area (optical) | Approximately 0.04 ha at 20 m red-edge resolution; approximately 0.001 ha at 3 m (Planet) |
| Cloud penetration | Full (SAR only); optical sensors blocked by cloud |
| Hyperspectral option | DESIS: 235 bands, 400–1000 nm, ~30 m resolution, irregular ISS revisit |
| Archive depth | Sentinel-2: from 2015; Sentinel-1: from 2014; Planet SuperDove: from 2021 (8-band) |
| Typical classification latency | 12–48 hours post-acquisition for automated workflows on open data; under 24 hours with priority commercial tasking |
Analytics Satellize can run
| Mat extent map (per acquisition) | Supervised random forest or support vector machine classifier trained on red-edge and NIR indices (Red-Edge Chlorophyll Index, Floating Algae Index) applied to Sentinel-2 or Planet imagery | GeoTIFF and vector polygon layer per scene, with confidence score per polygon |
| Expansion rate alert | Change detection between successive classified scenes; expansion rate computed in hectares per day; alert triggered when rate exceeds user-defined threshold | Automated alert feed (email or API webhook) with affected grid cells and expansion vector |
| Cloud-gap SAR extent estimate | Sentinel-1 VV/VH backscatter thresholding and texture analysis to separate dense floating vegetation from open water during cloud-affected periods | SAR-derived extent polygon flagged as lower-confidence, delivered into the same time series as optical classifications |
| Monthly time-series report | Aggregation of per-scene classifications into monthly area statistics with cloud-cover fraction and SAR gap-fill fraction reported per epoch | PDF and spreadsheet report with area-over-time chart, suitable for lake authority or ministry submission |
| Species composition assessment (periodic) | Spectral unmixing applied to DESIS hyperspectral data to estimate fractional cover of water hyacinth, water lettuce and other co-occurring macrophytes | Species-fraction raster and summary table, recommended quarterly or following major infestation events |
| Biomass proxy index | Regression of canopy chlorophyll content against red-edge band ratios, following published relationships between NDRE and fresh biomass density for floating macrophytes | Relative biomass density raster (low/medium/high classes) to prioritise mechanical harvesting zones |
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.