Duckweed’s spectral signal can reveal pigments—but not reliably its biomass

Duckweed—tiny floating plants often seen as a green film on water—is being investigated as a potential food crop. To choose promising varieties, growers would like to measure traits such as pigment content, photosynthetic performance and biomass quickly, without repeatedly harvesting plants for laboratory tests. One possibility is to measure the light plants reflect: different wavelengths can carry clues about what is happening in plant tissue.

A study posted as a bioRxiv preprint tested whether that approach works for duckweed. The researchers examined 133 UK accessions representing six species, grown under low growth-light conditions. They measured growth, biomass, pigments and photosynthetic traits alongside reflectance across many wavelengths. They also tested a smaller panel of 24 accessions under higher light, where differences in light tolerance could be examined.

The team compared two ways of using the reflectance data. The first used vegetation indices: formulas combining a few selected wavelength bands, commonly used to estimate plant traits. The second used partial least squares regression, or PLSR, a statistical method that searches across a wider stretch of the spectrum for patterns associated with a measured trait. In plain terms, the researchers asked whether a few carefully chosen colors—or a richer spectral fingerprint—could act as a useful proxy for measurements made directly on the plants.

The duckweed varied substantially in its traits, but familiar vegetation indices were not reliable all-purpose shortcuts. They showed limited predictive accuracy for total chlorophyll and maximum efficiency of photosystem II, a part of the photosynthetic machinery. Their performance was weak for other traits, including biomass. In the high-light group, some indices helped distinguish accessions with differences in photosynthetic efficiency and pigment balance, but the associations were moderate rather than strong.

Using the broader spectral information improved estimates for some traits. In validation tests, PLSR models explained about half the variation in total chlorophyll, maximum non-photochemical quenching (NPQ), and the efficiency of that photoprotective process. NPQ is one way plants dissipate excess absorbed light energy as heat. The models did not perform equally well for every trait: predictions of maximum photosystem II efficiency were weaker in validation, and estimates of carotenoid content and pigment ratios remained poor.

The central limitation was growth. Neither the selected indices nor the full-spectrum models predicted biomass and growth rate reliably. For example, the PLSR validation model for fresh biomass had a negative R², meaning it performed worse than simply using the validation-set average as a predictor. The researchers found that green and red-edge spectral regions—the transition between red and near-infrared light—were associated with pigment and photosynthetic traits and indirectly linked to biomass. But association with a trait is not the same as accurately estimating it.

Why might biomass be hard to infer from reflected light? The authors suggest that dense duckweed mats may create signal saturation: once the surface is mostly covered, more biomass may not produce a proportionate change in the measured spectrum. Duckweed’s floating structure and uniform, flat canopy may also affect reflectance. These are interpretations offered by the study, not proof that any one mechanism sets a fixed limit. The results apply to the experimental conditions tested, including laboratory measurements of dense populations; they do not establish how well the approach would work in every pond or commercial growing system.

The practical message is measured optimism. Hyperspectral sensing could help monitor chlorophyll and some aspects of photosynthesis without destructive sampling, and full-spectrum modelling performed better than simple indices for several traits. But it should not be treated as a stand-alone biomass scale. The authors propose combining spectral data with other rapid measurements, such as imaging or fluorescence, and testing additional modelling approaches.

This work is a bioRxiv preprint posted November 11, 2026, and was not certified by peer review at the time of posting. Its findings should therefore be considered preliminary. The paper says the data are available from the corresponding authors upon request, rather than as a publicly downloadable dataset. Further testing—especially beyond the controlled conditions and accessions examined—will be needed before judging how useful these methods are for crop selection or commercial monitoring.