{"id":2281,"date":"2025-02-12T09:30:34","date_gmt":"2025-02-12T09:30:34","guid":{"rendered":"https:\/\/research.reading.ac.uk\/lemontree\/?p=2281"},"modified":"2025-02-28T10:05:12","modified_gmt":"2025-02-28T10:05:12","slug":"high-tech-imaging-offers-new-insights-into-uk-grasslands","status":"publish","type":"post","link":"https:\/\/research.reading.ac.uk\/lemontree\/high-tech-imaging-offers-new-insights-into-uk-grasslands\/","title":{"rendered":"High-tech imaging offers new insights into UK grasslands"},"content":{"rendered":"

A new study by researchers at Imperial College London, published in Ecological Informatics<\/em>, shows that hyperspectral sensing\u2014a technology capable of detecting subtle differences in vegetation\u2014can be employed to predict grassland biomass and biodiversity.<\/p>\n

The findings underline the benefits for more effective monitoring of ecosystems at a time when climate change and land-use pressures are reshaping landscapes across the world.<\/p>\n

What is hyperspectral sensing?<\/h2>\n

Grasslands play a crucial role in carbon storage, agriculture, and biodiversity, yet monitoring them at scale has long been a challenge.<\/p>\n

Traditional field surveys are time-consuming, while many space-borne sensors lack the fine spatial resolution needed to marry with the sampling unit (e.g. field-plot).<\/p>\n

Hyperspectral senors can capture reflected radiation across hundreds of narrow wavelengths, making it possible to detect signals pointing to variations in plant health, species diversity, and biomass.\u00a0 Leaves typically reflect weakly in the blue and red wavelengths because of absorption by photosynthetic pigments and strongly in the near-infrared wavelengths owing to refraction by intracellular structures.\u00a0 As a consequence, the spectral absorption response of vegetation in the red channel is strongly correlated with chlorophyll content whilst reflectance in the NIR is positively related to vegetation density.\u00a0 The contrast of these two domains has informed the development of vegetation indices (VIs) using reflectances corresponding to wavelengths with maximum and minimum sensitivity to properties such as variation in pigment concentration.<\/p>\n

Inside the experiment<\/h2>\n

The study took place at Nash\u2019s Field<\/a>, a 30-year grassland experiment at Silwood Park, Imperial College\u2019s research site near Ascot, UK.<\/p>\n

The experimental site manipulates multiple treatments in a split-plot design .\u00a0 The treatments include herbivory, grazing, soil pH, plant competition and soil fertility. \u00a0\u00a0The full design encompasses 1,152 plots of 2 x 2 m, but constraints meant that this study could only undertake to sample a subset.<\/p>\n

\"\"
Figure 1 Location and design of the Nash\u2019s Field site made up of 16 Blocks.<\/figcaption><\/figure>\n

The team collected hyperspectral reflectance data from the grassland canopy between 25 May and 6 June 2018, using a handheld spectroradiometer. This period was chosen to coincide with peak vegetation growth under predominantly clear skies.<\/p>\n

To complement the spectral analysis, aboveground biomass samples were collected from each plot during one week in August 2018. \u00a0Before harvesting, scientists recorded the number of plant species within each plot, identifying dominant taxa to assess biodiversity.\u00a0 This allowed the team to assess how different environmental factors influenced vegetation growth and diversity.<\/p>\n

<\/h2>\n

Key findings<\/h2>\n

For response variables of plant biomass and species diversity, the study investigated whether an hyperspectral Partial Least Squares Regression (PLSR) model offered better predictions than a suite of popular two-band VIs.\u00a0 The PLSR technique has been designed to analyse data with numerous, collinear input variables and uses all available spectral wavelengths simultaneously.\u00a0 By contrast, two-band indices utilise only a fraction of the reflectance information available in hyperspectral data.<\/p>\n

\"\"
Figure 2. Goodness of fit plots for the PLSR model predictions. Each point represents one sub-plot (n = 86). Model predictions are shown on the x-axes against (a) harvested biomass; (b) species diversity. The two experimental Blocks (K and R) are differentiated by colour.<\/figcaption><\/figure>\n

Among the results:<\/p>\n