Abstract:
Forests soils are exposed to changing environmental conditions on a global level and influenced by atmospheric element input and acidification processes. For forest management, the ongoing changes depict a major challenge in both the present and future. Thus, mitigation strategies are needed, which rely on up-to-date information about soil and humus conditions. Conventional mapping schemes are time-consuming and partly unreliable. Digital soil mapping approaches, in combination with proximal soil sensing procedures like visual and near-infrared spectroscopy (vis-NIR), have the potential to overcome the mentioned shortcomings. They could provide information at the forest stand scale, where currently no information about humus condition is available. These methods are already widely used for investigations on agricultural areas, but studies focusing on forests and organic humus layers are scarce. The objective of this thesis is therefore to investigate the potential and application of vis-NIR spectroscopy to predict and asses humus properties of forest soils, with respect to the utilisation of the results in silviculture. The included studies present the possibilities of using vis-NIR spectroscopy to predict various properties of the organic layer of forest soils in Saxony and assess different regression and machine learning methods, devices, and spatial scales. Practical application is shown by presenting an integrated framework that derives spatial information about humus conditions at forest stand scale, closing a known data gap. Our results show that vis-NIR spectroscopy is suitable to estimate various soil properties (C and N content, C/N ratio, pH value, cation exchange capacity, base saturation) for the so far not investigated humus layer of forest soils. The application offers a reduction of the required laboratory analysis. No clear preference could be determined regarding the different algorithms used for the humus property estimation. We found that micro-electro- mechanical-systems devices are suitable for humus property prediction. Their performance can compete with benchtop devices. The usage of the NIR region is crucial to achieve satisfying prediction accuracy. The presented framework is applicable and derives spatial predictions of humus properties at forest stand level. Vis-NIR spectroscopy can facilitate the data acquisition procedure. The improved data availability compensates the uncertainties of the predictions. Both vis-NIR spectroscopy and digital soil mapping approaches can be applied for humus property prediction of forest soils. The methods have the potential to play an important role for forest soil characterization in the future by enhancing data availability and support forest management regarding ongoing environmental changes and challenges.