# Predicting species diversity and community traits from remote sensing in species-rich grasslands

## TL;DR

The study explores how remote sensing can help monitor species-rich grasslands in Scotland, finding limited success in predicting species diversity and community traits.

## Contribution

The study applies remote sensing techniques at a national scale to species-rich grasslands, revealing limitations in predicting diversity and traits.

## Key findings

- No significant relationship was found between species diversity and spectral diversity metrics across seven study sites.
- UAV-mounted multispectral sensors predicted grassland traits more accurately than satellite data.
- Remote sensing monitoring of species-rich grasslands is hindered by increased variation and confounding factors.

## Abstract

Species-rich grasslands (SRGs) provide unique ecosystem services, yet they are some of the most understudied environments in Scotland. This is because their true extent is unknown, and their locations are often in unreachable areas. Remote sensing may ameliorate this by helping to distinguish between SRG classes and increase monitoring efforts. However, these applications have not been conducted across highly diverse grasslands at national scales. We aimed to address this by applying remote sensing techniques to multiple classes of species-rich grasslands, nationally. Using an Unmanned Aerial Vehicle (UAV), we investigated the Spectral Variation Hypothesis, testing the relationship between species diversity (species richness) and spectral diversity (variation in surface reflectance using the coefficient of variation and standard deviation). Additionally, data was acquired from Sentinel-2 and Planetscope satellites, as well as the UAV data, assessing the prediction of grassland community traits (above ground biomass, sward height, and SPAD-measured chlorophyll) across multiple spatial (8 cm − 10 m) and spectral (8–13 bands) resolutions. The results contrast with prior studies of single sites and less diverse grasslands, indicating that there was no significant relationship between species diversity and the spectral diversity metrics, standard deviation (p = 0.120) and the coefficient of variation (p = 0.141), across seven of our study sites. The accuracy of grassland trait prediction varied largely with spatial and spectral resolution and the combination of predictor variables used. The UAV mounted multispectral Micasense predictor variables were most successful in predicting all traits: sward height, above ground biomass, and SPAD-measured chlorophyll (R2 = 0.545, R2 = 0.221, R2 = 0.167 respectively). The results suggest monitoring across species-rich grassland classes using remote sensing may be hampered by increased variation and confounding factors in these highly diverse habitats. Further methodological advancements are needed for wide scale cross-grassland habitat differentiation and monitoring, and field guidelines for remote sensing species-rich grasslands must be elaborated.

The online version contains supplementary material available at 10.1186/s12862-026-02500-4.