Sensing Biodiversity Across Scales – Biodiversity Exploratories

Picture: The photo shows an unmown meadow with an area marked with white tape. In the foreground are project leader Professor Anja Linstädter and PhD student Florian Männer, both wearing hats to protect them from the sun. Ms. Linstädter is holding a black flat box in her hands at waist level. Inside the box is a white reference plate over which Mr. Männer is holding a field spectrometer to calibrate it. In the background of the image is a fenced climate measurement station and a row of bushes and trees. Meadows and forests can be seen on the horizon

Background

It is generally established that through the intensification of fertilisation and mowing practices in grasslands, the provision of ‘feed production’ as ecosystem service increases, but at the cost of biodiversity and other ecosystem functions. Remote sensing is useful to map certain plant traits and classify the vegetation into functional groups at coarse scales. However, due to the spatial discrepancies between ecological processes and management units in coupled social-ecological systems, we still have a limited understanding of the effects of land use on the relationship between biodiversity – ecosystem functions – and ecosystem services. For instance, while field ecologists measure traits at the species scale, satellites measure traits at pixel scale. Thus, pixels represent inter-pixel variance instead of inter-specific variance. Also, biodiversity and ecosystem functions can vary in space and time with a large number of drivers (e.g., succession stages, climatic gradients and variability, soil properties, land use, etc.) making extrapolations challenging.

Goals

We aim to improve the mechanistic understanding of the effects of land use on the interplay between biodiversity – ecosystem functions – and ecosystem services. For that, we are analysing the relationships between functional and structural diversity and the ecosystem service of forage production, and their temporal variation for three spatial scales (plot, farm, and landscape). We will achieve this by combining plot-based ecological and remote sensing research on land-use intensity and 5 Essential Biodiversity Variables (EBVs): Aboveground Biomass (AGB), Above Net Primary Productivity (ANPP), Leaf Area Index (LAI), plant phenology, and functional diversity.

Methods

The data collected from the 5 EBVs during a full annual cycle is allowing us to quantify forage production along with functional composition and diversity across the season at sub-plot level. At the same time and in collaboration with Core-3, we are retrieving multispectral and microwave data from UAVs and multiple satellites (Sentinel-1 & 2, Landsat-8, MODIS and PlanetScope). Following a multiscale sampling and upscaling strategy we are using machine learning and artificial intelligence algorithms to scale up the field information to the satellite pixels, as well as to a landscape scale.

Picture: The collage contains four photos of the research group's field work. Photo 1 shows three young scientists on an unmown meadow in the sunshine. On the left of the picture, two of the men are standing at a folding table on which a laptop is placed. To the right of the picture, the third man is walking, carrying a fixed-wing drone. A row of bushes and trees stretches along the left edge of the meadow. A few clouds can be seen in the blue sky. Photo 2 shows a young female scientist smiling into the camera on a mown meadow. In front of the woman is a quadrocopter on the ground. Tall deciduous trees can be seen in the background. Photo 3 shows an unmown meadow with an area marked with white tape. In the foreground are the project leader Professor Anja Linstädter and the PhD student Florian Männer, both wearing hats to protect them from the sun. Ms Linstädter is holding a black flat box in her hands at waist level. In the box is a white reference plate over which Mr Männer is holding a field spectrometer to calibrate it. In the background of the picture is a fenced climate measurement station and a row of bushes and trees. Meadows and forests can be seen on the horizon. Photo 4 shows a young female scientist and two young male scientists on an unmown meadow, all smiling at the camera. In front of the standing woman, one of the men is squatting and bagging something. To his left are several backpacks, a carrier bag, a box and a plastic bag. The second man kneels to the right of the picture, bent over with his hands in the grass. A white marking tape runs through the meadow. At the far end of the meadow, a row of trees and bushes can be seen. — Fig. 1. Different steps in our fieldwork. In cooperation with Core-3, multispectral data is collected with cameras on board of fixed-wing (a) and quadcopter (b) UAVs. A field spectrometer is used to collect spectral information in a wider range (c). Afterward, plant trait and biomass data are collected to be analysed later in the laboratory (d).

Hypotheses

We hypothesise that (i) the five EBVs can be derived on multiple spatial scales using multimodal satellite image time series data; and that (ii) the effects of land use on the relationship of biodiversity to ecosystem functions and services vary across spatial scales. Here, the functional and structural diversity is likely to play a key role in the level and temporal stability of feed production.

Picture: The collage contains four photos taken from above, representing proportions. Photo 1 shows a narrow feathered leaf about 5 centimeters long. Next to the leaf is a red square of paper, the edge length of which is indicated as 2 centimeters. Photo 2 shows a folded out folding ruler on a low overgrown meadow, forming a square with an edge length of one meter. Photo 3 shows in green and magenta hues the multispectral image of a drone from an open field. Drawn into the photo is a yellow square, the edge length of which is indicated as fifty meters. Photo 4 shows in brown and blue-gray hues the multispectral image of a satellite from a landscape with open fields. Drawn into the photo are three widely spaced yellow squares, the edge length of which is fifty meters. — Fig. 2. upscaling plant functional features (a) to cover estimates (b), UAVs with multispectral sensor (c) and multispectral satellite images from Planet Cubesat (d)

Results

In our project SEBAS, we have so far addressed the first hypothesis and successfully upscaled some of the five EBVs with satellite image time series. For this purpose, several deep learning models and random forest algorithms were trained and validated with vegetation data collected by the SEBAS and the Botany team. Good predictions were obtained for aboveground plant biomass and plant species richness in grasslands across the three exploratory regions (Muro et al. 2022). Predictions based on time series also allow us to study management effects over the growing season.

Furthermore, we produced good spatial models on differences in plant species composition (beta diversity). In contrast, the upscaling of community-weighted plant functional traits using Sentinel-2 satellite imagery was not successful, e.g. regarding specific leaf area, leaf dry matter content and chlorophyll content, as heterogeneity within pixels may not be captured.

Field spectroscopy as another remote sensing technique was used to predict forage quality at the plot scale and is expected to replace destructive and costly measurements in the future.

Please find further information about our results in the references under ’Publications’.

Publications

Predicting Species and Structural Diversity of Temperate Forests with Satellite Remote Sensing and Deep Learning

Hoffmann J., Muro J., Dubovyk O. (2022): Predicting Species and Structural Diversity of Temperate Forests with Satellite Remote Sensing and Deep Learning. Remote Sensing 14 (7), 1631. doi: 10.3390/rs14071631

Predicting plant biomass and species richness in temperate grasslands across regions, time, and land management with remote sensing and deep learning

Muro J., Linstädter A., Magdon P., Wöllauer S., Männer F., Schwarz L.-M., Ghazaryan G., Schultz J., Malenovsky Z., Dubovyk O. (2022): Predicting plant biomass and species richness in temperate grasslands across regions, time, and land management with remote sensing and deep learning. Remote Sensing of Environment 282, 113262. doi: 10.1016/j.rse.2022.113262

Predicting tree species and structural diversity across temperate forest types with Satellite Remote Sensing and deep learning

Hoffmann J. (2022): Predicting tree species and structural diversity across temperate forest types with Satellite Remote Sensing and deep learning. Bachelor thesis, University Bonn

Untersuchung des Zusammenhangs zwischen den Fernerkundungsdaten von Sentinel-1 sowie Sentinel-2 und der Landnutzung bei gemäßigten Grünlandflächen in Deutschland

Rubach M. (2021): Untersuchung des Zusammenhangs zwischen den Fernerkundungsdaten von Sentinel-1 sowie Sentinel-2 und der Landnutzung bei gemäßigten Grünlandflächen in Deutschland. Bachelorarbeit, University Bonn

Assessing the possibilities of Sentinel-1 time-series and Omnibus algorithm to detect grassland management practices: A case study in the Swabian Jura

Bott J. (2021): Assessing the possibilities of Sentinel-1 time-series and Omnibus algorithm to detect grassland management practices: A case study in the Swabian Jura. Bachelor thesis, University Bonn

Grassland farmers’ perception of biodiversity patterns and trends: The role of land use practices and the ecological context

Klimke M. (2021): Grassland farmers’ perception of biodiversity patterns and trends: The role of land use practices and the ecological context. Master thesis, University of Cologne

The potential of Sentinel-1 radar images for inferring vegetation structure in grasslands – A case study of biodiversity exploratories in Germany

Blum D. (2021): The potential of Sentinel-1 radar images for inferring vegetation structure in grasslands - A case study of biodiversity exploratories in Germany. Master thesis, University Bonn

Deriving Essential Biodiversity Variable on ecosystem fragmentation with Remote Sensing in German grasslands

Ableitung einer essentiellen Biodiversitätsvariable für Ökosystemfragmentierung mittels Fernerkundung in deutschen Grünlandflächen

Roth M. (2020): Deriving Essential Biodiversity Variable on ecosystem fragmentation with Remote Sensing in German grasslands. Bachelor thesis, University of Bonn