Background and Research Approach

My Master’s thesis focuses on the recovery of forest disturbance areas of previously spruce stands in low mountain ranges of Central Germany, specifically the Southern Harz Mountains, the Thuringian Forest, and the Thuringian Slate Mountains. Common disturbance products from remote sensing focus on either disturbed or undisturbed forest, while nuances in deadwood retention are usually not considered.

Based on remote sensing time series of biophysical variables and auxiliary data, I aim to develop a machine-learning (ML)-based model to classify and distinguish existing post-disturbance management variants. The analysis focus on Sentinel-derived LAI, topography, soil, and climate data.

To support this approach, detailed categorical training data was collected in the field, extending the spatial scope of the existing research project “ResEt-Fi” (BMFTR, REGULUS program) from individual plots to the regional scale. This fieldwork was supported by funding from the Competence Centre for Landscape Resilience and the Eva Mayr-Stihl Foundation, which enabled independent empirical data collection beyond the regular scope of the degree program. This comprehensive data collection enables robust model training and validation across all three study regions and management types.

By analyzing how LAI recovery curves and environmental factors differ between management strategies, this work will not only improve the understanding of early succession and support recovery monitoring, but also enable new insights into restoring ecosystem functioning in the context of assessing landscape resilience.

Methods

In preparation for the fieldwork, potential disturbed forest areas were identified using satellite imagery in the vicinity of existing “ResEt-Fi” plots within each of the nine study areas, comprising three areas per major region. Over five days of fieldwork, these areas, as well as additional sites identified on site, were classified into one of three post-disturbance management categories: clearing, high stumps, and standing deadwood.

Sites were digitized using two complementary approaches. Where feasible, site boundaries were mapped directly in the field by walking the perimeter of each area and recording its spatial extent using GPS data. In cases where direct field mapping was not possible or practical, site extents were delineated based on field-collected reference points and digitized using satellite imagery. This combined approach ensured a consistent and accurate spatial representation of site boundaries.

Achievements and Challenges

Overall, the fieldwork progressed well, and a large proportion of the predicted sites could be successfully included in the dataset. In the Thuringian Slate Mountains, data collection was facilitated by the high availability of large disturbed forest areas, in some cases extending over very large spatial scales. However, due to time constraints, not all of these areas could be fully digitized.

In contrast, in the Thuringian Forest, considerable effort was required to identify a comparatively small number of suitable disturbed forest areas. In retrospect, additional sampling time could have been allocated to the Thuringian Slate Mountains, where a substantially larger number of suitable sites was available; however, this would have conflicted with the aim of maintaining a balanced representation of sites across the study regions.

Minor logistical difficulties were encountered due to roads blocked by fallen trees or generally unsuitable road conditions for the vehicle used during fieldwork. Owing to the timing of the survey in late October, all sites were easily accessible on foot, although steep slopes under rainy conditions required increased caution and additional time to traverse.

Outputs and Outlook

The outcome of the fieldwork is a comprehensive dataset comprising approximately 500 spatially delineated sites representing different post-disturbance forest management practices. These sites are management-defined units and therefore vary considerably in size and shape. This heterogeneity reflects real-world management conditions and is explicitly accounted for in subsequent model development.

Depending on the spatial filtering criteria applied during preprocessing, approximately 15% of the sites are expected to be excluded from model training due to insufficient spatial extent. This limitation was anticipated during fieldwork; nevertheless, these sites were digitized to retain their potential value for alternative analyses and future applications where minimum area requirements may be less restrictive.

Overall, the resulting dataset provides the essential categorical training data required to develop and validate the planned machine-learning classification model and supports improved understanding of early forest succession processes and forest landscape resilience.

• Collaboration: Collaboration with local forest personnel proved highly valuable during fieldwork. In particular, support from a forest trainee in the Etterwinden study area facilitated efficient site identification and provided important contextual knowledge on local disturbance patterns and management practices, thereby improving the quality and relevance of the collected training data.
• Regional disturbance dynamics: A pronounced contrast in the availability of disturbed forest areas between the study regions was observed during fieldwork. While suitable sites were scarce and required extensive searching in the Thuringian Forest, the Thuringian Slate Mountains exhibited a high density of disturbed areas. This confirmed initial assumptions regarding regional disturbance dynamics and underscored the suitability of the selected study regions for comparative analysis.
• Fieldwork: Fieldwork provided valuable practical insights into logistical constraints associated with mapping disturbed forest areas in remote and heavily forested terrain. Limitations related to accessibility and transportation highlighted the importance of accounting for logistical factors in fieldwork planning and time allocation, particularly when conducting regionally extensive surveys.