Näthe, Paul: Retrieval of information and data products for calibration, validation and spatial-temporal analyses from automated field spectrometers. - Bonn, 2024. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5-79850
@phdthesis{handle:20.500.11811/12564,
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5-79850,
author = {{Paul Näthe}},
title = {Retrieval of information and data products for calibration, validation and spatial-temporal analyses from automated field spectrometers},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2024,
month = nov,

note = {Intact ecosystems are the basis of human food security, air quality and provide naturally re-growing resources. Adaptive ecosystem management in the face of climate change requires detailed and continuous data about the ecosystems’ status. Optical remote sensing allows the non-invasive acquisition of information from terrestrial ecosystems and produces very large amounts of multi-dimensional data. However, it poses the following challenges at the same time: (1) Many different optical sensors use different measurement protocols, yet a joint exploitation of their data would improve temporal, spatial and spectral details available for further analyses. (2) Standardizing data products eases comparing and interpreting optical, remote-sensed data, but requires a specific processing chain, which also includes the propagation of uncertainties. (3) The retrieval of data-products is affected by systematic influences, e.g. from the atmosphere or the surface-properties, which bias the retrieved signal and require a correction approach.
The use of standardized, automated field spectrometers enables the continuous, unattended acquisition of hyperspectral data at very high temporal resolution in proximity sensing on the ground. The high dimensionality of hyperspectral down-welling and up-welling radiance recorded in the visible-near infrared (VIS-NIR) spectral range enables the retrieval of detailed atmosphere and vegetation properties. The application of Machine Learning (ML) algorithms is promising to disentangle multiple-redundant spectral information, isolate irrelevant or disturbing spectral information and find relevant, correlating spectral information, while offering detailed investigation of uncertainties and levels of confidence around the data products. Thus in the first study, Solar Induced chlorophyll Fluorescence (SIF), a proxy for photosynthesis in vegetation, is retrieved from hyperspectral field-measurements using a novel, ML-driven approach and avoiding atmospheric reabsorption. The second study in this thesis demonstrates farther the potential of exploiting continuous, hyperspectral VIS-NIR measurements using ML for the investigation of NOx concentration in the atmosphere. Furthermore, high-resolution field spectrometer measurements allow the convolution of multispectral sensor characteristic at overlapping spectral ranges. In consequence, the third study of this thesis harmonizes a network of standardized, automated field spectrometers in ten different locations around the world in comparison to Sentinel-2 bottom of atmosphere reflectance, and investigates effects of variable temporal-spatial heterogeneity. In the final study, automated field spectrometers were used as central transfer instruments inter-calibrating satellite and two airborne multispectral sensors, while correcting for continuous changes of down-welling radiance over time. Addressing the above stated challenges facilitated recommendations for the standardization of optical proximity sensing data and for using automated field spectrometers as a centerpiece of data fusion enabling a more holistic and more detailed ecosystem monitoring.},

url = {https://hdl.handle.net/20.500.11811/12564}
}

The following license files are associated with this item:

http://creativecommons.org/licenses/by-nc-nd/4.0/