Day 4

Detailed paper information

Back to list

Paper title Exploring capabilities of modelling forage provision using field spectroscopy and UAV imagery in a semi-arid rangeland
  1. Florian Männer University of Bonn Speaker
  2. Vistorina Amputu University of Tübingen
  3. Nichola Knox Namibia University of Science and Technology
  4. Katja Tielbörger Institute of Evolution and Ecology, Eberhard Karls University Tuebingen, Germany
  5. Anja Linstädter Institute for Biology and Biochemistry, University of Potsdam, Germany
Form of presentation Poster
  • A3. Biosphere
    • A3.06 Biomass monitoring
Abstract text Forage provision is an important indicator of rangeland health and is reliable for evaluating land degradation. In dry rangelands it is largely limited by moisture availability compounded with grazing pressure as it sustains a significant proportion of livestock-based systems. For sustainable and adaptive management, parameters such as biomass production and forage quality are of key interest. Yet, their quantification and monitoring still remains laborious and costly. Advancing remote sensing technologies such as hyperspectral readings and drone imaging enable rapid, repeatable and non-destructive estimations of these parameters that can be applied over large spatial scales. While these are increasingly being integrated for ecological research, robust prediction models supported by field data is still lacking, especially in highly dynamic systems like semi-arid savannahs. In our study we aim to answer the following research questions: (1) to what extend can we model forage provision (quality and quantity) from resampled hyperspectral data? (2) Can we model forage provision from UAV-based multispectral imagery calibrated with field spectrometer prediction models? (3) How do artificial hyperspectral data, interpolated from multispectral data enhance the prediction quality? (4) How does forage provision vary between two differently managed rangelands? To address these questions, we took hyperspectral readings with a field spectrometer from herbaceous canopies along transects in two management types in a Namibian semi-arid savannah. Plant biomass samples were collected at the reading areas to measure forage quantity and forage quality. Machine learning and deep learning methods were used to establish hyperspectral prediction models for both forage quality and quantity. We applied these models to hyperspectral readings from a broader area. For upscaling the hyperspectral models, we acquired drone multispectral imagery along the same transects. Multispectral prediction models were set up using the predicted values from the hyperspectral prediction model. As predictors for the model we used the pure spectra, derived vegetation indices and artificial hyperspectral data from interpolating the multispectral bands. We then created forage quantity and quality maps to visualize and compare forage provision dynamics in the two management systems. While field-based hyperspectral models offer greater spectral resolution for assessing complex forage quality parameters, and drone imagery offer unprecedented spatial and temporal data products for mapping forage parameters at a landscape level, independently they are limited. Thus, emerging UAV-based hyperspectral imagery minimizes these discrepancies, a technology that will catapult remote sensing to map even more complex variables and resolve ecological questions.